[House Hearing, 118 Congress]
[From the U.S. Government Publishing Office]
AI INNOVATION EXPLORED:
INSIGHTS INTO AI APPLICATIONS
IN FINANCIAL SERVICES AND HOUSING
=======================================================================
HEARING
before the
COMMITTEE ON FINANCIAL SERVICES
U.S. HOUSE OF REPRESENTATIVES
ONE HUNDRED EIGHTEENTH CONGRESS
SECOND SESSION
__________
JULY 23, 2024
__________
Serial No. 118-104
Printed for the use of the Committee on Financial Services
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
www.govinfo.gov
_______
U.S. GOVERNMENT PUBLISHING OFFICE
56-909 PDF WASHINGTON : 2026
HOUSE COMMITTEE ON FINANCIAL SERVICES
PATRICK McHENRY, North Carolina, Chairman
FRENCH HILL, Arkansas, Vice MAXINE WATERS, California, Ranking
Chairman Member
FRANK D. LUCAS, Oklahoma SYLVIA R. GARCIA, Texas, Vice
PETE SESSIONS, Texas Ranking Member
BILL POSEY, Florida NYDIA M. VELAZQUEZ, New York
BLAINE LUETKEMEYER, Missouri BRAD SHERMAN, California
BILL HUIZENGA, Michigan GREGORY W. MEEKS, New York
ANN WAGNER, Missouri DAVID SCOTT, Georgia
ANDY BARR, Kentucky STEPHEN F. LYNCH, Massachusetts
ROGER WILLIAMS, Texas AL GREEN, Texas
TOM EMMER, Minnesota EMANUEL CLEAVER, Missouri
BARRY LOUDERMILK, Georgia JAMES A. HIMES, Connecticut
ALEXANDER X. MOONEY, West Virginia BILL FOSTER, Illinois
WARREN DAVIDSON, Ohio JOYCE BEATTY, Ohio
JOHN W. ROSE, Tennessee JUAN VARGAS, California
BRYAN STEIL, Wisconsin JOSH GOTTHEIMER, New Jersey
WILLIAM R. TIMMONS, IV, South VICENTE GONZALEZ, Texas
Carolina SEAN CASTEN, Illinois
RALPH NORMAN, South Carolina AYANNA PRESSLEY, Massachusetts
DANIEL MEUSER, Pennsylvania STEVEN HORSFORD, Nevada
SCOTT FITZGERALD, Wisconsin RASHIDA TLAIB, Michigan
ANDREW R. GARBARINO, New York RITCHIE TORRES, New York
YOUNG KIM, California NIKEMA WILLIAMS, Georgia
BYRON DONALDS, Florida WILEY NICKEL, North Carolina
MIKE FLOOD, Nebraska BRITTANY PETTERSEN, Colorado
MICHAEL LAWLER, New York
ZACHARY NUNN, Iowa
MONICA DE LA CRUZ, Texas
ERIN HOUCHIN, Indiana
ANDREW OGLES, Tennessee
Matthew Hoffmann, Staff Director
C O N T E N T S
----------
Tuesday, July 23, 2024
Page
OPENING STATEMENTS
Hon. Patrick T. McHenry, Chairman of the Committee on Financial
Services, a U.S. Representative from North Carolina............ 1
Hon. Maxine Waters, Ranking Member of the Committee on Financial
Services, a U.S. Representative from California................ 2
STATEMENTS
Hon. French Hill, Chairman of the Subcommittee on Digital Assets,
Financial Technology, and Inclusion, a U.S. Representative from
Arkansas....................................................... 4
Hon. Stephen F. Lynch, Ranking Member of the Subcommittee on
Digital Assets, Financial Technology, and Inclusion, a U.S.
Representative from Massachusetts.............................. 4
WITNESSES
Mr. Vijay Karunamurthy, Chief Technology Officer, Scale AI....... 5
Prepared Statement........................................... 8
Mr. Ondrej Linda, Senior Director, Personalization AI, Zillow.... 16
Prepared Statement........................................... 18
Ms. Elizabeth Osborne, Chief Operations Officer, Great Lakes
Credit Union................................................... 22
Prepared Statement........................................... 24
Mr. Frederick Reynolds, Deputy General Counsel for Regulatory
Legal and Chief Compliance Officer, FIS Global................. 36
Prepared Statement........................................... 38
Ms. Lisa Rice, President and CEO, National Fair Housing Alliance
(NFHA)......................................................... 43
Prepared Statement........................................... 45
Mr. John Zecca, Executive Vice President, Global Chief Legal,
Risk and Regulatory Officer, NASDAQ............................ 78
Prepared Statement........................................... 80
APPENDIX
ADDITIONAL MATERIAL SUBMITTED FOR THE RECORD
Hon. Maxine Waters:
American Bankers Association................................. 138
National Association of Residential Property Owners (NARPM).. 148
U.S. Chambers of Congress.................................... 150
RESPONSES TO QUESTIONS FOR THE RECORD
Written responses to questions for the record from Mr. Vijay
Karunamurthy
Representative Maxine Waters................................. 154
Written responses to questions for the record from Mr. Ondrej
Linda
Representative Maxine Waters................................. 157
Written responses to questions for the record from Ms. Elizabeth
Osborne
Representative Maxine Waters................................. 158
Written responses to questions for the record from Mr. Frederick
Reynolds
Representative Maxine Waters................................. 160
LEGISLATION
H.Res.------, Expressing the sense of the House of
Representatives with respect to the use of artificial
intelligence in the financial services and housing industries.. 161
H.R. 5808, the "Preventing Deep Fake Scams Act".................. 165
H.R. 7781, the "Artificial Intelligence Practices, Logistics,
Actions, and Necessities Act".................................. 170
H.R. ------, a bill To require the Federal financial agencies to
carry out a study and report on standardized descriptions for
vendor-provided artificial intelligence systems, and for other
purposes....................................................... 174
AI INNOVATION EXPLORED:
INSIGHTS INTO AI APPLICATIONS
IN FINANCIAL SERVICES AND HOUSING
----------
Tuesday, July 23, 2024
U.S. House of Representatives,
Committee on Financial Services,
Washington, DC.
The committee met, pursuant to notice, at 10:04 a.m., in
room 2128, Rayburn House Office Building, Hon. Patrick McHenry
[chairman of the committee] presiding.
Present: Representatives McHenry, Lucas, Sessions, Posey,
Luetkemeyer, Huizenga, Wagner, Barr, Williams of Texas, Hill,
Emmer, Loudermilk, Davidson, Rose, Steil, Timmons, Norman,
Meuser, Fitzgerald, Kim, Donalds, Flood, Lawler, Nunn, De La
Cruz, Houchin, Ogles, Waters, Velazquez, Sherman, Lynch, Green,
Himes, Foster, Beatty, Vargas, Gottheimer, Gonzalez, Casten,
Pressley, Horsford, Tlaib, Garcia, Williams of Georgia, and
Nickel.
Chairman McHenry. The committee will come to order.
Without objection, the chair is authorized to declare
recess of the committee at any time.
The hearing is titled, ``AI Innovation Explored: Insights
into AI Applications in Financial Services and Housing.''
Without objection, all members will have 5 legislative days
within which to submit extraneous materials to the chair for
inclusion in the record
I will now recognize myself for 4 minutes for an opening
statement.
OPENING STATEMENT OF HON. PATRICK T. McHENRY, CHAIRMAN OF THE
COMMITTEE ON FINANCIAL SERVICES, A U.S. REPRESENTATIVE FROM
NORTH CAROLINA
I want to thank our witnesses for being here today. This
hearing is an historic milestone for the committee. Today we
will examine use cases of artificial intelligence in financial
services and housing. At its best, artificial intelligence
holds the promise to enhance, not replace human progress.
Unfortunately, its novelty and perceived risks have delayed
adoption in many sectors of our economy.
In January, Ranking Member Waters and I formed a bipartisan
Working Group to explore the use cases of AI within our
committee's jurisdiction. I appreciate the ranking member and
her staff for their participation efforts in this endeavor. I
also want to thank Working Group co-leads, Congressman Hill,
the vice chair of the full committee, and Congressman Lynch for
their leadership. This Working Group builds on the success of
the Task Force on Artificial Intelligence in previous
Congresses. The bipartisan Working Group embarked on a month's
long fact-finding mission, including six roundtables and a trip
to Massachusetts Institute of Technology (MIT), which I was
pleased to be a part of, and countless other meetings. Members
engaged with regulators, technologists, market participants,
and consumer advocates to gain a deeper understanding of how AI
is currently impacting our financial system, as well as how it
may do so in the future. This was a comprehensive process,
which is outlined in the bipartisan report released last week.
This collaborative approach will benefit Congress as we work to
keep pace with this evolving technology.
To be sure, AI is not new in the financial services
industry. Financial firms have used algorithms for decades,
whether it is for analyzing large datasets or for trading.
However, new developments in generative AI, or GenAI, have
excited the imaginations of many. The financial services
industry, one of the most highly regulated in America, is a
clear entry point as policymakers attempt to tackle the thorny
questions AI presents. We should be leery of rushing
legislation. It is far better we get this right rather than to
be first. In other words, policymakers should not measure twice
and cut once. We want to get this right. American financial
regulations are technology neutral and should remain that way.
To be clear, the use of GenAI does not absolve a firm from
complying with existing consumer protection laws. At the same
time, our regulators must ensure that they are equipped to take
on this new technological frontier. This committee should
examine whether current regulation needs to be clarified and
carefully consider if targeted legislation to close regulatory
caps may be needed. Today we will discuss the importance of
quality data, credit access model creation, model risk
management, cybersecurity, human touch points, and a whole lot
more. This hearing will advance the conversation on the use of
AI in financial services and cement this committee's leadership
on artificial intelligence policy.
I will close with this. GenAI is here. We cannot put it
back in the box. It will become more widely adopted to the
point that it is embedded in our everyday lives. We cannot
allow the fear of the unknown to thwart the United States' role
as a hub for technological innovation. We should embrace it, we
should be at the forefront of it, and we should ensure that all
technologists want to be a part of the American ecosystem, our
rule of law, and our regulatory structures. Far greater than
the risks associated from AI itself are the risks of allowing
foreign competitors and adversaries to lead the development,
adoption, and terms of use. The United States should be at the
forefront and remain at the forefront as we are right now. I
yield back.
I will now recognize the ranking member of the full
committee, the gentlewoman from California, Ms. Waters, for 4
minutes.
OPENING STATMENT OF HON. MAXINE WATERS, RANKING MEMBER OF THE
COMMITTEE ON FINANCIAL SERVICES, A U.S. REPRESENTATIVE FROM
CALIFORNIA
Ms. Waters. Thank you, Mr. Chairman. Good morning. I would
like to start by applauding and thanking Chairman McHenry for
working with me to establish the committee's bipartisan Working
Group on Artificial Intelligence. After months of roundtables
moderated by Representatives Hill and Lynch with regulators and
experts, we released a report detailing the implications of AI,
including newer generative AI models on financial services and
housing. I am pleased that this report lays out several
recommendations for the committee to better protect consumers
and our markets as AI is increasingly adopted. For example, the
report makes clear that this committee must lead the House in
overseeing AI, and that we must ensure our regulators enforce
existing laws, including antidiscrimination laws and policies,
so that any benefits of AI are broadly shared.
Today's bipartisan hearing, the Working Group, and the
report successfully build off the work this committee started
under my leadership as chair when I stood up in Congress with
the first ever Task Force on AI in 2019, chaired by Mr. Foster.
This task force held more than a dozen hearings, exploring the
potential risks and benefits of AI. This task force is part of
a long list of efforts by Democrats to oversee new technologies
and to ensure they are developed in the best interests of
consumers.
As companies forge ahead with AI, it is more important than
ever that this committee and Congress not only continue this
kind of oversight, but prioritize its review of AI and
diversity, equity, and inclusion. As we know, AI is built by
humans and relies on data that may reflect bias and systemic
inequities, or perpetuate discrimination. For instance, people
of color looking to secure mortgage loans have been denied or
are overcharged significantly as a result of discriminatory
algorithms. However, responsibly crafted, AI may help expand
financial access for consumers, aspiring homeowners, renters,
and businesses, while removing barriers to economic mobility.
For this reason, I am pleased that this hearing is also
considering my draft legislation, which would better inform
consumers when products and services incorporate AI. The data
is used to train AI-based decisions and provide regulators with
the origin of the data used by AI. The Treasury issued this
recommendation in its report entitled, ``Managing AI-Specific
Cybersecurity Risk in the Financial Sector'' would benefit from
a standardized description, similar to the food nutrition
label, for vendor-provided AI systems and data providers.'' I
am also pleased that we are considering Representative
Pettersen's legislation, Preventing Deep Fake Scams Act, in
this hearing, which sets up a task force to examine how banks
and credit unions can protect themselves and their customers
and members from fraud associated with AI.
Through efforts like these, we can build more transparent
and equitable systems, as well as trust and safety in an
increasing AI-driven world, so I look forward to today's
discussion, and I yield back.
Chairman McHenry. The gentlelady yields back. The chair
recognizes the chairman of the Subcommittee on Digital Assets,
Financial Technology, and Inclusion, the gentleman from
Arkansas, Mr. Hill, who is co-lead of the Task Force on AI as
well. He will be recognized for 1 minute.
STATEMENT OF HON. FRENCH HILL, CHAIRMAN OF THE SUBCOMMITTEE ON
DIGITAL ASSETS, FINANCIAL TECHNOLOGY, AND INCLUSION, A U.S.
REPRESENTATIVE FROM ARKANSAS
Mr. Hill. Thanks, Mr. Chairman. I am proud to be the co-
chair with my good friend Mr. Lynch of Massachusetts of this
Working Group on Artificial Intelligence. I see Mr. Lynch got
the wardrobe memo today. Good. Dressing very sharp today.
Tackling this complex issue in a bipartisan manner is not
easy, but it is critically important. Leading up to this
hearing, as the chairman said, we had six roundtable
discussions and an outstanding deliberative visit to the
academic experts at MIT. We also have a Working Group that Dr.
Foster and I serve on behalf of Speaker Johnson and Minority
Leader Jeffries at the House-wide level, demonstrating the
importance this House has in making sure that we get AI right.
We need to be innovative, and we need to be leaders in the
world. Instead of being reactionary in our policymaking,
Congress should take a measured approach to AI that seeks first
to understand this rapidly evolving technology while exploring
potential gaps in the law or supervision. We will begin that
work today. I yield back.
Chairman McHenry. The gentleman yields back. The chair
recognizes the ranking member of the Subcommittee on Digital
Assets, Financial Technology, and Inclusion, the gentleman from
Massachusetts, Mr. Lynch, co-chair of the AI Working Group, for
1 minute.
STATEMENT OF HON. STEPHEN F. LYNCH, RANKING MEMBER OF THE
SUBCOMMITTEE ON DIGITAL ASSETS, FINANCIAL TECHNOLOGY, AND
INCLUSION, A U.S. REPRESENTATIVE FROM MASSACHUSETTS
Mr. Lynch. Thank you, Mr. Chairman, and to the ranking
member. Thank you for holding this important hearing. I was
honored, as the gentleman from Arkansas mentioned, to co-lead
the Working Group on Artificial Intelligence alongside my
friend Mr. Hill. In June, we led the Working Group on a visit
to MIT. Through this work, we had thoughtful conversations with
industry leaders, academics, and technologists to learn about
the policy implications of AI.
The explosion of AI adoption across industries has raised
serious questions about implications for financial access,
employment, data privacy, and intellectual property. While AI
has the potential to expand access to financial products, we
need to ensure AI models are explainable, transparent, and do
not lead to bias. Last, it is critical that the U.S. remain a
leader in AI safety and security, and I look forward to
continued collaboration with my colleagues as we further
explore policy actions that are needed. I yield back.
Chairman McHenry. The gentleman yields back. I want to
thank the co-chairs of the Working Group for their leadership.
The day we spent at MIT was fascinating, deeply informative,
and I would commend that to other committee members who were
not able to attend that day. I appreciate Mr. Hill, Mr. Lynch
for your leadership, and also, Mr. Lynch wanted to show off his
native Boston. Neither French nor I, in particular, came away
with an accent. So thank you for that.
Ms. Waters. Excuse me, Mr. Chairman. Did you mention Sylvia
Garcia was there?
Chairman McHenry. Oh yes. Yes. Yes, and Ms. Garcia was
there, but I was just thanking them for that. Let me recognize
the testimony of our distinguished panel. We have a large panel
today, so I am going to introduce everybody at once, then we
will go one by one, and you will have 5 minutes for an oral
presentation of your testimony.
We recognize today Mr. Vijay Karunamurthy, chief technology
officer at Scale AI; Dr. Ondrej Linda, senior director of
personalization AI at Zillow; Ms. Elizabeth Osborne, the chief
operations officer at Great Lakes Credit Union; Mr. Frederick
Reynolds, deputy general counsel for regulation and legal and
chief compliance officer at FIS Global; Ms. Lisa Rice,
president and CEO of the National Fair Housing Alliance; Mr.
John Zecca, executive vice president and global chief legal,
risk, and regulatory officer at Nasdaq. Thank you for taking
time to be here today. Each of you will be recognized for 5
minutes for an oral presentation of your testimony, and without
objection, your written testimony will be made a part of the
record.
I will note that we have House floor votes that are
imminent, and I will consult with the ranking member on the
appropriate time for us to break, but we are going to try to
get through as much as we possibly can so we can get to member
questions.
With that, Mr. Karunamurthy.
STATEMENT OF VIJAY KARUNAMURTHY, CHIEF TECHNOLOGY OFFICER,
SCALE AI
Mr. Karunamurthy. Chairman McHenry, Ranking Member Waters,
and members of the House Financial Services Committee, thank
you for the opportunity to be here today. It is an honor to
discuss the exciting role that AI is playing in the financial
services and housing sectors, industry trends in adopting AI,
and recommendations for the safe deployment of AI. My name is
Vijay Karunamurthy, and I am the field chief technology officer
at Scale AI.
Scale is helping the world's leading technology companies
build, deploy, and evaluate artificial intelligence. Since
2016, Scale's mission has been to build the data foundry for
AI. Every day we work with the leading automotive companies
building high-quality computer vision datasets, nearly every
generative AI company to ensure that their large language
models, or LLMs, have the highest quality data, the world's
most prominent companies applying artificial intelligence, and
the U.S. Government to bring the best-in-class commercial AI
technology to the Department of Defense and other agencies.
Underpinning all of this is our work on AI tests and evaluation
to ensure that our customer's AI is safe to deploy. Our years
of work on the frontlines of nearly every major AI innovation
gives us a unique view on how to build and deploy high quality
and safe AI systems.
AI has the potential to deliver tremendous benefits for
society, but this can only happen if it is deployed safely and
responsibly. The financial services industry has long
recognized the value of AI and has begun to adopt the
technology back in 1982, with the housing sector following
shortly thereafter. Due to the long history of its use, Federal
and State regulators have also incorporated many aspects of AI
into existing regulations for those sectors. In November 2022,
the AI world changed with the release of Chat Generative Pre-
trained Transformer (ChatGPT), and today, nearly every sector
of our economy is actively looking to harness the power of
generative AI. Based on our role in the AI ecosystem, I want to
highlight three important trends that we are seeing to most
successfully deploy of AI in the financial services and housing
sectors.
First, foundational AI elements are critical. At a high
level, AI is only as good as the data that it is trained on.
High-quality data produces high-quality outputs. As the leading
companies in the world deploy AI systems, they should ensure
that their data strategies and budgets are in place so that
they can turn raw data into AI-ready data, which means that is
curated, transformed, and annotated so that AI systems can
leverage it.
Secondly, fine-tuned proprietary data is a game changer.
The leading commercial models today can perform at a master's
degree or a Ph.D. level. However, companies need not only
master's and Ph.D. level capabilities, but also industry-
specific expert level capabilities that only come from years of
institutional knowledge and data within the company. This is
where proprietary data comes in. Training models with the
company's data gives AI the institutional knowledge that allows
the enterprise to derive the most value from it. For example,
JPMorgan reportedly has roughly 150 petabytes of proprietary
data, or 150 times the entirety of ChatGPT's initial training
data. Harnessing the power of that data would create incredible
insights for this bank to better serve its customers because
this data is directly relevant. If companies want to harness
the power of AI for their own applications, they must use the
best-in-class commercial models and leverage their data to
build the highest quality AI systems.
Lastly, deploying AI must be done safely and responsibly.
Test and evaluation is the best way to ensure that AI is safe
to deploy because it entails robust testing of the AI system to
better understand the potential vulnerabilities and
comprehensive evaluations to produce a holistic picture of the
AI system as a whole. For this reason, the U.S. Government has
long recognized the importance of developing the standards,
frameworks, and methodologies that underpin test and
evaluation. As companies leverage AI systems, they need to
consistently test and monitor them to ensure that they remain
safe to deploy.
Given the amount of time that both sectors have used AI,
new regulations may not be necessary. However, a thorough and
comprehensive gap analysis should be conducted to confirm
whether or not these regulations apply. If gaps exist, we must
fill them. We believe that should be done with risk-based
sector specific regulations, but AI safety metrics are still
needed. In practice, this may be as simple as regulators
stating that AI is held to the same standards as humans, but it
must be government's role to establish these metrics.
Lastly, government and industry should collaborate to
ensure that the workforce of the future is ready for this
technology. If we do not train our workforce, we will not reach
the limitless potential of AI.
AI is the most promising technological innovation of our
time, but it will only reach its full potential if it is
deployed in a safe, responsible, and thoughtful manner. This is
skills mission. Thank you again for the opportunity to be here
today, and I look forward to your questions.
[Prepared statement of Mr. Karunamurthy follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman McHenry. Thank you for your testimony. We will now
recognize Dr. Linda for 5 minutes.
STATEMENT OF ONDREJ LINDA, SENIOR DIRECTOR, PERSONALIZATION AI,
ZILLOW
Mr. Linda. Thank you, Chairman Henry, Ranking Member
Waters, Congressman Hill, Congressman Lynch, and members of the
committee for holding this hearing on artificial intelligence
(AI) and for your work over the past month to examine issues
related to AI. I also want to acknowledge and thank the other
witnesses, including Lisa Rice, for her leadership and
dedication to housing fairness and equity.
My name is Ondrej Linda, and I am Zillow's senior director
of AI. My expertise is in developing and deploying large-scale
AI systems in leading AI organizations to create new
experiences to empower homebuyers, renters, and sellers. For
the past several years, I have also been deeply involved in
spearheading responsible AI practices at Zillow. Zillow is the
leading real estate marketplace, and with over 200 million
monthly unique visitors comes a tremendous responsibility to
lead the way in innovating responsible AI technology.
Today I would like to discuss how Zillow uses AI to help
consumers on their housing journey, how we work to identify,
reduce, and mitigate bias and discrimination, and the promise
we see when it comes to AI's ability to address housing equity
and access in this country.
Home searching and buying processes are complicated and
confusing and often fraught with emotions. In fact, according
to 2022 Zillow consumer survey, over half of homebuyers
reported they actually cried during the process. Since that
study was conducted, affordability and supply challenges have
led to homebuyer sentiments sitting at an all-time low. Zillow
is focused on helping consumers navigate this daunting market
and home-buying process by empowering them with information. In
fact, Zillow's first product, Zestimate, was an AI-driven
valuation model that informs people for free of their home's
potential estimated worth. Before Zestimate, homebuyers and
sellers had virtually no public access to pricing estimates.
Traditional online home listings, which show only two-
dimensional images, have not evolved for decades, and we have
changed that. Our AI-powered 3D home tours and immersive floor
plans is increasing transparency and opening up homes to people
who cannot tour in person. Thanks to AI, consumers on Zillow
can get personalized recommendation that expands their search
to a broader set of housing inventory. AI tools could also play
a role in decreasing appraisal bias and unlocking wealth for
people of color by allowing more objective and efficient
approaches to understanding home valuation, including in rural
areas. We are examining how AI can help to modernize existing
industry frameworks with the aim of decreasing costs,
streamlining paperwork, and reducing legacy access burdens in
the housing market.
At Zillow, we believe that AI can be a great tool to help
more people get a home and reduce bias and discrimination, but
only if used responsibly. I am proud to be part of an
organization that has adopted and publicly shared ethical AI
principles to guide our work and elevate industry standards,
fairness, safety, accountability, transparency, and
inclusiveness. To deliver on these promises, we recently open
sourced our Fair Housing Classifier. That means that civil
rights groups, tech companies, and others in the real estate
industry can adopt it for free and collaborate with us to help
improve it.
At Zillow, we are constantly thinking about the next
innovation and how we can have a positive impact on people's
lives. As we speak, Zillow's entire technology and product
organization is at our longstanding Hack Week. They are
evaluating and experimenting with new technologies, including
AI, to unlock new opportunities. Just down the street at FHFA,
Zillow is participating in a tech sprint on generative AI in
housing finance. We truly believe that these public-private
partnership and cross sector collaborations will be key in
harnessing the power of AI to make housing more accessible to
all.
Again, I want to thank the committee for holding this
hearing and your focus on this topic. We look forward to
working with all elected officials to identify ways to create a
pathway to AI innovation in housing that help inform, protect,
and empower customers. I look forward to your questions.
[Prepared statement of Mr. Linda follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman McHenry. Thank you for your testimony. After this
testimony of Ms. Osborne, we will recess the committee for
floor votes. For the panel, that should take until about 11:15
till we get back.
So with that, I will recognize Ms. Osborne for 5 minutes.
STATEMENT OF ELIZABETH OSBORNE, CHIEF OPERATIONS OFFICER, GREAT
LAKES CREDIT UNION
Ms. Osborne. Thank you. Good morning, Chairman McHenry,
Ranking Member Waters, and members of the committee. I am
Elizabeth Osborne, the chief operations officer for Great Lakes
Credit Union. We were founded in 1938 and are a low-income
designated, Housing & Urban Development (HUD)-certified
financial cooperative in the Chicago land market and
surrounding areas. We serve over 90,000 members in that market.
At the heart of our mission is the credit union philosophy of
people helping people. At Great Lakes Credit Union (GLCU), that
includes state-of-the-art technology and a range of financial
solutions in place for one goal: to enable our members to live
life on their terms.
Credit unions like mine are committed to using AI safely,
securely, and with the goal of helping members meet their
financial needs. AI enables credit unions to compete more
efficiently with larger banks and non-bank companies in the
financial services arena. AI has not altered credit unions'
historical role as relationship lenders committed to
maintaining a close bond with the communities they serve.
AI usage at credit unions generally falls into three
categories: underwriting support, combating fraud, and customer
service, which is the area we have focused on at GLCU. As part
of our commitment to innovation to our members, GLCU entered
into a partnership with interface.ai and launched Olive, our
virtual conversational AI assistant, with the goal to provide
our members with an enhanced, easy-to-use, and voice-enabled
virtual system. Olive is able to handle a wide range of member
inquiries and requests such as account balance, transaction
history, transfers between accounts, and more. Olive is
available 24/7 through multiple channels, including phone and
chat.
Since the introduction of Olive, GLCU has realized
remarkable results in terms of member satisfaction, member call
center performance, and employee engagement. For example, Olive
consistently handles over 60 percent of total inbound calls
during business hours and over 75 percent of all calls after
business hours, this compared to less than 25 percent handling
rate with our previous telephone banking solution which did not
include AI technology. Olive has increased the number of calls
fully serviced by the virtual assistant by over 200 percent
since launch. Olive provides recommended responses and advice
to call center employees when a member requires special
assistance or prefers to speak with an employee. An inherent
benefit of Olive is that it has enabled the credit union to
elevate the job description and level for call center agents as
their role moved from providing transactional support for
members to providing more consultative advisory support. This
shift has resulted in a higher pay grade for staff and a better
career path for our call center employees.
GLCU management is continuously working to improve and
expand Olive's capabilities and features. One of our main goals
this year is to upgrade Olive to speak Spanish to better serve
our Spanish-speaking members, which represent a large
percentage of the GLCU membership base. In addition, GLCU is
also exploring additional methods to leverage AI to improve
operations and services. One tool is Microsoft Power Automate,
which enables the credit union to automate repetitive and
manual tasks. By using Power Automate, staff can save time and
resources, reduce errors and delays, and increase productivity
and accuracy.
As policymakers grapple to legislate and regulate in this
emerging environment, it is important to recognize many
existing laws are technology agnostic and still apply. As with
other technologies, consumer financial protections and
antidiscrimination rules continue to have broad applicability,
whether the decisionmaking is human or AI. Existing regulation
requires credit unions to adopt robust practices to ensure that
the use of new technology does not jeopardize safety or the
rights of individual members. It is encouraging to see
recognition of the broad applicability of existing fair lending
and consumer financial protection laws in the bipartisan
Working Group report.
In conclusion, AI is an important tool for credit unions,
helps us to enhance service to our members, and provides
services that economies of scale may not otherwise allow us to
do. At GLCU, Olive is a strategic partner that helps us fulfill
the underlying mission of a credit union. Our use of Olive is a
prime example of how credit unions can effectively deploy the
use of AI to improve the lives of our members.
I thank you for this opportunity to appear here today and
welcome any questions you may have.
[Prepared statement of Ms. Osborne follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman McHenry. Thank you for your testimony. Pursuant to
the chair's previous announcement, the committee stands in
recess. I encourage members to return immediately after votes.
With that, we recess.
[Recess.]
Chairman McHenry. The committee will come to order. We will
continue with the testimony from our panel.
Next up in the order is Mr. Reynolds. You are now
recognized for 5 minutes. I feel like it is icing the kicker
for like the final three. The first three, you got off easy.
Sorry. So Mr. Reynolds, you are recognized for 5 minutes.
STATEMENT OF FREDERICK REYNOLDS, DEPUTY GENERAL COUNSEL FOR
REGULATORY LEGAL AND CHIEF COMPLIANCE OFFICER, FIS GLOBAL
Mr. Reynolds. Thank you, Chairman McHenry, Ranking Member
Waters, Congressman Hill and Congressman Lynch, and members of
the committee. My name is Frederick Reynolds, and I am deputy
general counsel and chief compliance officer for FIS. During my
time as a Federal prosecutor, as the deputy director for
Financial Crimes Enforcement Network (FinCEN), and now in the
private sector, I have had the opportunity to see firsthand how
new and emerging technologies impacted the financial sector and
national security, the two of which are inextricably linked.
Artificial intelligence, if responsibly utilized and regulated,
can have a significant and positive impact in modernizing
products and services, increasing financial access and
inclusion, and enhancing ethical standards and customer
outcomes. However, it also presents, in particular, generative
AI, substantial risks.
FIS is a financial technology company that delivers
technology solutions to major financial institutions around the
globe. Our operations power the global economy, managing money
at rest, money in motion, and money at work. Given the vital
role that we play in financial services, we are acutely
conscious of our significant responsibility to ensure that our
products and services are both safe and leading edge.
One area that FIS is currently focused on is the impact of
technologies like GenAI and Web3 on fraud prevention, identity
verification, and existing customer due diligence procedures.
As these technologies become more widely adopted, our
traditional means of assessing trust, i.e., identity, are less
effective through anonymous or pseudo-anonymous online
identities and deepfakes. This, coupled with an increased
ability to avoid the formal financial system, puts the overall
customer due diligence (CDD) and anti-money laundering (AML)
programs at risk, recognizing that the current AML controls,
which require full knowledge of a user's legal identity as a
proxy for trust, may be at risk of becoming ineffective. It is
imperative that the industry look to next-generation CDD
identity controls to take advantage of GenAI and Web3, but also
protect against their misuse. This requires us to rethink how
to define identity and create solutions, such as digital
identity or zero-knowledge identity proofs that focus on
behavior and trust within the financial ecosystem rather than
easily fake data elements.
As chief compliance officer, I have a responsibility to
safeguard FIS and its customers by ensuring adherence to laws,
regulations, and ethical standards. This commitment extends to
FIS' advancements within AI. Our priority is clear: to deliver
products that our clients can trust, avoiding high-risk and
uncertain solutions that do not meet legal, regulatory, and
ethical standards.
FIS is continuously exploring ways to enhance what we can
offer to our clients, and we deeply understand the unique
challenges that our clients, both large and small, face and the
opportunities that AI can bring to these businesses. We are
embracing innovation but taking measured steps toward it to
deliver real value to those we serve. We pride ourselves on
robust risk management programs to ensure that we are
effectively managing risk, while, at the same time, allowing us
to be agile and innovate. As we further integrate AI and GenAI
into our operations, our approach remains grounded in six core
principles--transparency, explainability and auditability,
accuracy and accountability, privacy and data protection,
security--all of which must be rooted in fairness and human
values.
We believe the current regulatory regime governing
financial services are robust enough to support the responsible
adoption of AI technologies. Consumer protection,
antidiscrimination, and financial market regulations provide a
solid foundation for using AI in financial services. These
regulations ensure that AI technologies are used ethically and
responsibly. Still, considering we are so early in the
innovation curve of GenAI, it is imperative that we continually
reassess regulations to ensure that they allow for innovation
in a risk-thoughtful manner. The one area, however, we do
believe that congressional action is critical is updating our
current customer identification program (CIP) and CDD regime
and the development of a digital identity system.
The AI regulatory landscape is, more broadly, not as stable
and not as defined as it could be. Globally, the regulatory
environment for AI is uncertain, with each country adopting a
unique approach based on varying risk considerations. While
consistent global standards are ideal, it is imperative to our
ability to innovate that we have at a minimum a consistent
Federal approach to AI regulation. Fragmented regulation across
States will introduce complexity, burden innovation, and impede
the positive impact that can come with AI technology adoption.
We urge policymakers to adopt a cohesive regulatory framework
that not only promotes innovation, but also ensures robust
compliance with regulatory requirements.
Thank you once again for having myself and FIS at this
hearing and look forward to further discussions and questions.
Thank you.
[Prepared statement of Mr. Reynolds follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman McHenry. Thank you. Ms. Rice, you are now
recognized for 5 minutes.
STATEMENT OF LISA RICE, PRESIDENT AND CEO, NATIONAL FAIR
HOUSING ALLIANCE (NFHA)
Ms. Rice. Chairman McHenry, Ranking Member Waters, and
distinguished members of the House Financial Services
Committee, thank you for the opportunity to share information
about NFHA's work regarding the application of artificial
intelligence in the housing and financial services sectors.
NFHA is the Nation's only national civil rights agency solely
dedicated to eliminating all forms of discrimination. As a
trade association for more than 170 fair housing agencies
located throughout the country, NFHA works to support survivors
of discrimination and build resilient, inclusive, well-
resourced communities where everyone can thrive.
AI holds great promise for improving systems, democratizing
housing and credit opportunities, lowering costs, and
increasing productivity. It also holds great dangers for
perpetuating bias, spreading misinformation, and generating
other harms. For NFHA, AI and technology are the new civil and
human rights frontier. How we develop and use these powerful
systems will dictate whether people will have fair access to
critical opportunities in a manner that is safe, or whether
people will continue to be presented with prohibitive systems
that can generate inaccurate decisions that consumers do not
understand.
Multiple enforcement actions, including those brought by
NFHA, as well as research conducted by many stakeholders, show
there are multiple systems in use today that produce harmful
impacts. Credit scoring and insurance scoring systems that rely
on outdated, unrepresentative, or inaccurate data result in
consumers being denied for the credit and other services they
need to purchase a home, obtain insurance, or secure rental
housing. Bias systems also result in consumers paying higher
costs unnecessarily. For example, one study revealed that Black
and Latino borrowers are being overcharged by risk-based
pricing systems to the tune of $765 million per year. The
Casualty Actuarial Society acknowledged that algorithmic bias
can manifest in systems used in the insurance sector, including
underwriting, pricing, and claims models. Tenant screening
selection and dynamic pricing systems can also perpetuate bias
and generate decisions that are inexplicable.
Automated valuation models, or AVMs, can also generate
bias. With support from Zillow, NFHA hosted a hackathon just
last year designed to explore these challenges. The teams
participating in our hackathon identified several structural
barriers. In fact, because of these systemic flaws, in some
cases, AVMs had only a 15 percent rate of accuracy in
predominantly Black neighborhoods.
While AI can pose perils that we must not ignore, it can
also hold great benefits. We can use AI to increase data
accuracy, provide guidance to consumers on how to appropriately
access risk and credit, or enable lenders to use rental housing
and cash-flow data to underwrite consumers. We can also use AI
to increase fairness. Just a few months ago, NFHA and FairPlay
AI released important research explaining how distribution
matching, a technique that allows multifaceted optimization,
can increase fairness in mortgage loan originations for Black
and Latino applicants by up to 15 percent and reduce pricing
disparities for the same groups by up to 20 percent with no
tradeoff in model accuracy.
To ensure AI will be used to benefit society, NFHA urges
Congress to pass comprehensive legislation that includes the
following: shoring up civil and human rights protections;
providing full internet access for all communities; providing
the public access to data, particularly for important research;
ensuring consumers have agency over their own data; supporting
the use of privacy-enhancing technologies; assuring systems are
safe, accurate, and effective; requiring entities to provide
accurate notices to consumers; providing funding to promote
important research; ensuring Federal employees are well trained
and agencies have sufficient resources to provide oversight and
enforce the Nation's laws; and providing for human
alternatives. The recent global information technology (IT)
outage that halted airline services and crippled banking
institutions provides a clear example why there must be a human
or manual alternative.
Thank you so much for the opportunity to testify, and I
look forward to answering your questions.
[Prepared statement of Ms. Rice follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman McHenry. Thank you, Ms. Rice. Mr. Zecca, 5
minutes.
STATEMENT OF JOHN ZECCA, EXECUTIVE VICE PRESIDENT, GLOBAL CHIEF
LEGAL, RISK AND REGULATORY OFFICER, NASDAQ
Mr. Zecca. Thank you, Chairman McHenry and Ranking Member
Waters, and all the members of the committee. On behalf of
Nasdaq, I appreciate the opportunity to testify on Nasdaq's use
and governance of artificial intelligence technology. From our
founding 50 years ago, when we became the world's first
electronic exchange, Nasdaq has continued our role as a
technology leader in the capital markets. Our technology powers
more than 130 markets around the world. We help the banking
community fight financial crime. At Nasdaq, our vision is to be
the trusted fabric of the world's financial system, and our
purpose is to advance economic progress for all. Our solutions
enhance liquidity by powering robust markets, ensure
transparency by leveraging data and analytics to fuel
decisionmaking by corporates and investors, and protect
integrity by managing risks, surveilling markets, enhancing
compliance, and fighting crime. Right now, AI applications are
enabling a fairer and more efficient and resilient financial
system.
Today, I wanted to share a few examples of how we are
harnessing AI's potential at NASDAQ. Our anti-financial crime
business is leveraging AI to more effectively than ever before
fight criminal organizations engaged in smuggling, drugs,
terrorism, and money laundering. An estimated $3.1 trillion in
illicit funds flowed through the world's financial system in
2023, and last year alone, losses from fraud totaled $485
billion globally. To address this massive problem, our cloud-
based platform is used by thousands of financial institutions
of all asset sizes to dramatically improve the effectiveness of
their financial crime risk management programs. After all,
there is a money trail behind every act of terrorism or human
trafficking. Financial crimes are not victimless crimes. We
must be more innovative than the criminals who are also using
AI.
To further protect the integrity of our markets and our
market surveillance business, we use machine learning to detect
insider trading, market manipulation, and fraud. AI will also
enhance effectiveness as we analyze billions of transactions
and events every day. Meanwhile, innovation continues as we
recently introduced Dynamic Midpoint Extended Life Order (M-
ELO), the first exchange AI-powered order type, and we are also
deploying AI in our market data, corporate governance, and
index businesses. These products are a reflection of a robust
governance framework centered around our responsible AI
principles. Governance is centrally administered, eliminating
the AI sprawl risk of more decentralized approaches. Leading
this effort is an executive steering committee that includes
our CEO and other senior members of our leadership team. The
program is aligned with a U.S. National Institute of Standards
and Technology AI risk management framework.
As Congress and agencies look to develop AI-related laws
and regulations, we ask them to consider the following:
Existing regulations and regulatory structures should be
leveraged where possible. Like prior technological innovations,
the adoption of AI technology does not necessarily demand
sweeping regulatory changes. New regulations should avoid
focusing on specific technology itself and, instead, be
proportionate to the benefits, risks, and harms of the AI
application. For example, the regulation of an AI tool to
detect drug trafficking should not be the same as the
regulation of the identical AI tool if it is used to approve
apartment rental applications.
The regulatory environment should endeavor to be flexible
and allow for innovation. We support sandboxes to promote
innovation and believe the regulatory environment should be
transparent with a feedback loop between government, academia,
industry, and the public, and, most importantly, the
development of industry standards is critical. AI-specific
regulation should be consistent and harmonized. Regulators
should coordinate across borders. While we oppose the creation
of a central regulator, we support leveraging National
Institute of Standards & Technology (NIST) to coordinate across
the government. It is also important to avoid a patchwork of
differing State AI laws that could stifle innovation and harm
the competitiveness of the United States globally. We want to
enable the use of AI, not hinder it.
Nasdaq is committed to using AI in a responsible, ethical,
and transparent manner. We look forward to working with the
committee and regulators to foster a supportive and balanced
regulatory environment for AI. We believe that AI is a
promising technology that can bring significant benefit to our
industry and the wider economy. I would be happy to answer any
questions you may have.
[Prepared statement of Mr. Zecca follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Chairman McHenry. Well, thank you for your testimony. This
has been a great panel, and I will now recognize myself for 5
minutes for purposes of questions.
As I said in my opening, this committee has been at the
forefront of examining how AI is impacting the financial
services and housing industries, both during my chairmanship
and the chairmanship of Ms. Waters, and I want to use today's
hearing to continue examining the range of benefits and
potential risks of AI, specifically GenAI and its challenges
through some use cases. Mr. Zecca, how do you envision AI
evolving over the next decades to not only enhance market
efficiency and risk management, but also to foster innovation
and accessibility, accessibility in particular of our capital
markets?
Mr. Zecca. Well, I think that is an important question, how
does it democratize the markets, and I do think there are a
couple of opportunities. For one, I think through the uses, as
I mentioned, of analytics and data, there is the opportunity to
develop products, whether it is exchange-traded fund (ETF)
products, unique products that could be available at lower cost
to Main Street investors, and I think that is a true
opportunity for them to diversify. Similarly, I think you will
see more of the analytics being used for investment advice, and
again, targeting communities that have not been able to pay for
individual tailored financial advice. So I think there are
great opportunities there.
In the market, I think you will see unique order types
coming, but the part that I am always interested in, you know,
being the head of regulation is the surveillance end, and I do
think AI can be a force multiplier for us in monitoring
markets, looking for aberrant activity. Just think about the
billions of trades that are going through the market. If you
can have AI sort of do the triage, then the analysts can focus
on a much more developed case.
Chairman McHenry. Excellent. Excellent. Well, Mr.
Karunamurthy, I do not think we can overstate your testimony,
and particularly, you talk about the need for high-quality
data, and I do not think we can overstate how vital that is. So
how does Scale AI ensure the quality and accuracy of data,
particularly as it relates to input data, as well as model
structures?
Mr. Karunamurthy. Thank you, Chairman McHenry. This is an
important part of our work, working with industry partners, and
so we are happy to spend some time talking about it. The work
that goes into making data AI ready is really fundamentally
looking at the quality of the data that is going into these
models. Do we have diverse geographic data, do we have data
from multiple different perspectives on a topic, and also
helping the models to understand what are authoritative sources
of data. So where these models have been trained on a range of
different perspectives about current events or financial
markets, helping these models understand what are
authoritative, sometimes proprietary internal sources of data
that can better address customer questions. By paying attention
to the data and how that data is shifting over time, you get a
better sense of how these AI systems will perform. It is a
constant process of observation about where new data is coming
in from, how the models can use more authoritative sources of
data and then how that data can be improved over time.
Chairman McHenry. Now, Mr. Reynolds, I know that FIS is
focused on the impact technologies like GenAI, blockchain, the
deployment of blockchain technology, on fraud prevention,
identity verification, existing customer due diligence
procedures. Would you highlight the opportunities that GenAI
and blockchain technology, the challenges that it poses, and
the opportunities it has also presented?
Mr. Reynolds. Yes, thank you, Chairman. When you look at
GenAI, obviously one of the foundational points about GenAI is
it is incredibly creative, and it really levels the playing
field, both in terms of opportunities to enhance the system,
but then attacks on the system. If you look at attacks on the
system, one of the risks of GenAI is it can very creatively
come up with alternative identities. It can fake data, and it
can fake data elements. So one of the ways that I think we have
to look at GenAI and the potential blockchain is to understand
is there an opportunity to create, for instance, a digital
identity that a consumer can have, if it is on the blockchain,
it becomes immutable. So it becomes a digital identity, which,
over time, as it exists within the system, it can gain
permissions based on what it does in the system where it is.
That can sit on a blockchain, and it could be something that
can allow consumers much greater access to financial services
because that sort of passport, so to speak, allows them into
the financial system.
I think, though, again, we have to be very careful with
generative AI because in our current system that is really
reliant on basic identity elements like date of birth, address,
social security number, it is very easy to fake now with GenAI,
and I do think it puts our current CIP system somewhat at risk.
Chairman McHenry. Okay. Mr. Zecca and Mr. Reynolds, you are
talking about reducing the cost structure, which gains a huge
advantage for market participants. I think that is an
interesting outcome, and thank you for mentioning it. We will
now recognize the ranking member, Ms. Waters, for 5 minutes.
Ms. Waters. Well, thank you very much, Mr. Chairman. I
would like to address this question to Ms. Rice. We know that
third-party tenant screenings and rent-setting companies and
the data they use are often the primary basis for landlord
decisions about who has access to housing and how much renters
will pay. Numerous lawsuits have highlighted how technologies
have used inaccurate or discriminatory data that may not even
be connected to the applicant or use data that artificially
inflates rents. For example, the company, RealPage, which
offers rent-setting algorithms to landlords, is currently being
investigated by the Justice Department for potentially
anticompetitive practices and sued by several States' attorneys
general for allegedly using private data to artificially
inflate rents and grow profits. What kind of rules or
restrictions on data and transparency should be established for
these and other types of third-party companies that affect
housing access and affordability?
Ms. Rice. Ranking Member Waters, thank you for the
question. As it relates to tenant screening selection systems,
which can often use data that can result in biased outcomes,
data like criminal records history, credit scores, data like
whether or not an eviction has ever been filed against a
person, that kind of data that is highly correlated to race can
often result in discriminatory outcomes and can generate
decisions that are, well, that data may not necessarily be
predictive as to whether or not a person is going to be a good
tenant or pay their rent on time.
So systems that utilize data that is highly correlated to
race, like criminal records, that type of data should really be
constrained and limited in models. We think that when a model
is using data that serves as a proxy for race or gender, that
there should be higher protocols and standards placed around
those models, and they should be regularly monitored and tested
for discriminatory outcomes. They also need to be subject to
our antidiscrimination laws.
When it comes to dynamic pricing models, those also can
present prohibitive barriers for consumers. For example,
consumers who are using housing choice vouchers can often be
denied housing opportunities because the price-setting
algorithms are not predictable. You can go to an apartment
complex in the morning and get one rental rate for a unit and
go to the apartment complex in the afternoon and get a
completely different rental rate that can exceed HUD's fair
market values. So we have to have tight controls on these types
of technologies that can, as I said, disproportionately
discriminate against survivors of domestic violence, who are
disproportionately women, people of color, people who have
disabilities, and other protected classes.
Ms. Waters. So when you say, ``tight control,'' you mean
legislation?
Ms. Rice. Yes, I do.
Ms. Waters. That would absolutely have the guardrails that
are needed in order to protect the consumers?
Ms. Rice. Yes.
Ms. Waters. All right. Thank you very much. I have another
question that I would like to ask Mr. Karunamurthy. The
complexity within AI models is not easily explainable or
observable, can result in unfair bias, discriminatory outcomes
that impact customers' lives. I think transparency of the
underlying data of an AI system can help address challenges in
explainability, bias, discrimination, and even compliance with
regulatory standards.
In fact, I led a delegation to Silicon Valley and met with
several University of California (UC), Berkeley scientists who
warned that not being able to understand the outcomes of
particular AI models is one of the most alarming outcomes and
something that must be addressed. With increased availability
of generative AI more financial services companies rely on
third-party vendors for high-quality training, data for AI
models. Your company, Scale AI, which provides data labeling
services to companies has a unique and important role in this
ecosystem.
This is a long question, and I have overrun my time. So I
think I am going to have to bring it to your attention to get
an answer following this committee.
Ms. Waters. Thank you, and I yield back.
Chairman McHenry. Thank you, Ranking Member. We will now go
to the gentleman from Arkansas, co-lead of this taskforce, Mr.
Hill, for 5 minutes.
Mr. Hill. Thank you, Chairman. Thank you, panel. Thanks for
your indulgence this morning as we went and voted.
Over the course of our many early morning roundtables, the
committee heard from companies about how they are using
artificial intelligence to lower compliance costs, better
detect fraud, and enhance customer service, just to name a few.
We also heard from the supervisors from HUD to the Public
Company Accounting Oversight Board (PCAOB) to Securities and
Exchange Commission (SEC) and the bank supervisors that they
were all being principles-based in their looking at how to
supervise AI. Taking this sort of technology-first approach to
understanding AI use cases in the financial sector, in my view,
has been central to our Working Group's approach. The problem
with policy-first approach previously taken by the committee is
that technology changes so quickly that we are stuck talking
about yesterday's policy issues while the actual technologies
and the use cases move right ahead, and that is sort of the
challenge I have seen in other jurisdictions like in Europe's
approach to AI.
So let me start out and turn to that European approach.
Foreign jurisdictions have taken and seek remarkably different
approaches to regulating AI. Mr. Reynolds, you and Mr. Zecca
both work at companies with a global presence. Mr. Reynolds,
how would you assess the different approaches being
contemplated, such as the comprehensive good housekeeping seal
of approval approach that Europe is taking? What do you think
about that?
Mr. Reynolds. Thank you, Congressman, for the question. It
is very challenging as a global company to deal with such
differing approaches. In the U.S., we have taken a very open
approach, and I think that has allowed a significant amount of
innovation. In my remarks, I noted that we strongly believe in
good regulation and good regulatory behavior, and that is a
strength of our company. So certainly I am a fan of following
the rules as the chief compliance officer, and I think it is
important.
I do think with the European approach, it tends to be
fairly restrictive, and so I do think from the European
approach, it can stifle innovation and make it harder, I think,
for companies to adopt this new technology and I think it
creates an unlevel playing field across different geographies
because the technology that will work well in the United States
may not work very well in Europe----
Mr. Hill. Thanks.
Mr. Reynolds [continuing]. and it may be very different in
Asia.
Mr. Hill. Mr. Zecca, a comment along that line?
Mr. Zecca. I agree with that. I think when you look at the
AI regulation in Europe, which is just going into effect now,
it has been a 2-year process, and if you look at where they
started to where they are now, it has changed a lot, and it is
not only because there were comments. It is because the
underlying technology has changed in the last few years.
Mr. Hill. Yes. Kind of hard----
Mr. Zecca. So the question is----
Mr. Hill [continuing]. It is kind of hard to decide what
you are----
Mr. Zecca. Right. So now they have frozen it in time----
Mr. Hill. Yes.
Mr. Zecca [continuing]. in a way, and that is----
Mr. Hill. So that also speaks to another key point I think
we heard in our roundtables, which is that there is this
presumption I picked up in the roundtables of
``noncompliance.'' In other words, there is this assumption
that if you are deploying AI internally for internal compliance
purposes or marketing purposes, that somehow you are oblivious
to your existing supervisory, regulatory, or legal obligations
under Federal law, and, Lisa, we have talked about this. They
suspect that you are not in compliance, which, generally, I am
going to stipulate that is probably not true but how do we have
people show their homework effectively?
Mr. Reynolds, let me go back to you just because you work
with so many community banks, and then maybe we will get a
credit union comment there but people are trying to comply with
fair credit reporting, fair lending, right, when they use AI.
They do not miss that obligation. They still have the
obligation.
Mr. Reynolds. Yes, absolutely, Congressman. I think
compliance is really at the forefront of certainly all of our
customers' minds. It is in the forefront of our mind. The first
thing we did in AI actually is to set up a governance council
to ensure that we were following the laws, following the
regulations, and really following ethical standards for AI, and
making sure all of our use cases followed all the regulatory
standards and ethical standards. So I think that is critical,
and I do not think anybody loses sight of that. I think the key
is really auditability.
Mr. Hill. Yes.
Mr. Reynolds. We make sure that we are very clear as to how
we are using it, why we are using it, what controls we have in
place, and we openly share that with all our members.
Mr. Hill. I think that is good. I mean, that is what we
heard from supervisors: show us your homework on how you got
there. Let us hear it from a credit union perspective on that.
Ms. Osborne. Yes. Thank you, Congressman Hill. I agree with
Mr. Reynolds' testimony and what he stated here. As a financial
institution, we are subject to a myriad of regulations, so when
we look at technologies, AI or not, and providers, the same
rules apply. So any expectations we have of staff to ensure
that we are adhering to compliance standards, any rules and
regulations, that would need to be built into AI solutions as
well and then the expectation with third parties, which we
leverage for services like this at an asset size like Great
Lakes Credit Union, we then would develop with them reporting
and monitoring.
Mr. Hill. Thank you. I appreciate that response, and I
yield back the balance of my time. Mr. Chairman, thank you.
Chairman McHenry. We will recognize the gentleman from
California, Mr. Sherman, for 5 minutes.
Mr. Sherman. It is natural that we focus on the AI issues
that confront us now and in the next few years, but I want to
take a minute to say that we need to research and develop and
deploy systems that will monitor for and prevent self-
awareness, volition, survival instinct, and ambition in
superhuman computers and artificial intelligence. The last time
there was a higher level of intelligence introduced on this
planet is when our ancestors said hello to Neanderthal. We then
said goodbye to Neanderthal, and I do not look forward to being
the Neanderthal of this century. We are not actually spending
any money on a problem, and this problem will not affect us
until the latter half of this century on preventing self-
awareness and ambition in artificial intelligence.
Now to the issues of today. When I first looked at AI, I
said, great, decisions will be made without bias and without
racism because I could not imagine a machine being racist and
then I looked at a study that came out of an elite university,
and they looked at all the factors that causes a student to be
successful at the university, and they concluded that the
number one factor was being named ``Jared,'' that being named
``Jared'' apparently correlates to success at that university.
It does not correlate with, hopefully, reducing and eliminating
bias in our society. This is not entirely new. We had
statistical analysis. Then we had algorithms. We had computers.
Now we had artificial intelligence. At each stage, we face new
decisionmaking mechanisms that deserve our attention, but it is
not new in-kind. We have made this jump before.
I am very excited about our ability to monitor for money
laundering and monitor transactions, and particularly in the
capital market signs of insider trading, and I am particularly
concerned that crypto will allow us to see everything and
understand nothing. We will know all the transactions and we
will not know who is making them.
Let us see. The one other thing that we have is whether it
is hard and fast rules imposed on an employee or whether it is
a computer or AI. Clearly a prior criminal record is not a
positive, a prior eviction is not a positive, but if you are
dealing face-to-face the old way, you can overcome that
perhaps, whereas, if you have AI and it gives so many bad
points for a criminal record and so many bad points for an
eviction, then the computer makes the decision without the
intervening human appeal.
Ms. Osborne, if AI will first be deployed by the giant
financial institutions, how critical is it for a community
financial institution or, like, a credit union to have access
to affordable technology solutions and to have that in a way
that is consistent with your compliance applications, all the
while competing with much larger institutions?
Ms. Osborne. Yes. So thank you for your question,
Congressman. In terms of our usage of technologies and partners
to aid us in providing loans and decisioning to members, we,
through our vendor or third-party management requirements,
these are regulatory requirements that are across the board,
the same regardless of the size of the financial institution.
Those standards are followed to ensure that we have a rigorous
approach when looking at third parties. So when thinking about
the different types of factors that would ensure there is not
bias or decisioning for that, that would be unfair to one, we
would then ensure that our technology is applying that same
rules to that as well, and that we have the appropriate checks
and balances just like we would for a human.
Mr. Sherman. I yield back the remaining second of my time.
Chairman McHenry. The gentleman from Oklahoma, Mr. Lucas,
is recognized for 5 minutes.
Mr. Lucas. Thank you, Mr. Chair.
Chairman McHenry. Also the chair of the Science Committee.
Mr. Lucas. Thank you, Mr. Chairman. Advances in AI systems
have led to a wide range of use cases with the potential to
benefit nearly all aspects of our society. With the rapid pace
of technological progress in artificial intelligence, it is
essential that we encourage this innovation here in the United
States. The best way to reduce potential risks associated with
AI is to ensure that advancements in this field are made with
American values. We should not rush to regulate or put up
roadblocks that will impede American leadership and have a
detrimental impact on safety, security, and transparency. I
appreciate Chairman McHenry for putting this hearing together
so that we can hear from experts on the use cases of this
technology.
Dr. Linda, you discussed how Zillow is in the early stages
of integrating AI tools into real estate transactions. How do
you see this technology simplifying the house-buying process?
Mr. Linda. Thank you for the question, Congressman. We do
ask our real estate partner agents how do they spend their
days, how effective they are at their work, and what they told
us is that they spend most of their time doing mundane,
repetitive tasks. That is where we see that AI has a very big
role to play, functioning as a trusted co-pilot to really
automating tasks. We recently experimented with AI co-
summarization, which is a great way that we can help agents
spend their time with the customer and really putting the
humans at the center of the real estate transaction. The more
we do that, the more we can enable customers, and understand
the process and eventually buy a home.
Mr. Lucas. Most agencies have existing authority to oversee
the outcomes of AI systems within their perspective
jurisdictions. Dr. Linda, could you discuss how AI tools are
deployed within current regulatory structures and how the
policymakers can avoid impeding novel use cases for these
systems?
Mr. Linda. Thank you for the question, Congressman. I
think, you know, Zillow has been at AI development on behalf of
our customers since 2006, and the whole time, Fair Housing and
Equal Credit Opportunity Act has been our guiding principles.
We constantly look at how this regulation can be applied the
same way we have been applying to our business and our human
operators as well to AI. We do not see any difference in there.
As a specific example, we recently open sourced some of our
work that enables you to use recent large language models and
make them fair housing compliant and aware of the regulation.
So from that standpoint, we feel that exploring how existing
regulation can be applied to AI is the way that we can make
progress.
Mr. Lucas. Mr. Zecca, you discussed how AI allows for an
extraordinary amount of data to be analyzed. Could you
elaborate on how Nasdaq leverages this technology to track
fraud and manipulation and how this information is communicated
to market regulators?
Mr. Zecca. Absolutely. So if you think about in the
financial crime area, you know, the massive data, you are
looking at activity that can go across accounts, people who
have common names, a trail that may end up in a manual Google
search. So it is really about tying the threads together to
build a case, and AI can do a lot of that work for the
investigator. My experience with bank compliance groups is they
want to comply, but they are strapped and there is a lot of
information being thrown at them, so we have seen 30, 35
percent increases in productivity because of the use of AI. So
then those cases, when they are built, could be a report to the
government. Alternatively, in the market surveillance, where we
do that, we would either prosecute the case ourselves or refer
it to the SEC.
Mr. Lucas. Continuing with you, Mr. Zecca, how
transformative do you view AI to be in our capital markets, and
how can our handling of this technology influence our status of
having the most robust and liquid capital markets in the world?
Mr. Zecca. Well, I think that is the core question,
Congressman, how do we maintain our lead in the capital market
is one of the great American advantages, and we certainly would
not want to do anything that hinders that. I think it is
important to remember that the financial sector is heavily
regulated and the regulators who oversee them are very
knowledgeable about the market, and I think that piece is
important when you marry that with AI knowledge for the
oversight.
I think where it is going to make the most difference, I do
think it is going to democratize products because you are going
to be able to offer things at a cheaper price. I think the
information over the long run is going to be more accurate and
more digestible, and we will see unique trading opportunities
that do not exist now that we cannot even imagine. As I
mentioned, we introduced the first AI order type, which had a
fill rate improvement. We went through the SEC process, but I
think there will be many more.
Mr. Lucas. With that, Mr. Chairman, I yield back the
balance of my time.
Chairman McHenry. We will now recognize the gentleman from
Massachusetts, Mr. Lynch, co-chair of the Working Group on AI.
Mr. Lynch. Thank you, Mr. Chairman. Mr. Zecca, I know you
are suggesting that democratization is going to develop. So Sam
Altman, I think he reported that GPT-3 cost $100 million, then
GPT-4 was $400 million. Now they are saying GPT-5 is going to
be somewhere between $2 and $2-and-a-half billion. For these
large language models, it does look like because of the costs
and the energy required, that there will just be a few big
firms that will control that market. You are talking about
democratization, but it does not look that way. It looks like a
few big players that have sort of organic access to massive
data sets, that those might dominate and that would not be
truly democratic, you know. Push back on that, if you will.
Mr. Zecca. Sure. Well, I think it is always something we
have to consider, how the concentration is happening. When you
break down the large language models, there are open-source
ones, there are non-open source, so there are quite a few
providers.
Mr. Lynch. Yes.
Mr. Zecca. Then in many cases, larger companies, in
particular, are using their own data sources. Of course,
remember we are talking about the large language model piece,
but there is also the code writing side of it, which is
different, which is----
Mr. Lynch. Sure. That is right.
Mr. Zecca. Yes.
Mr. Lynch. And training.
Mr. Zecca. So I think there are a lot of use cases where it
is not going to be quite so concentrated. Even where it is
concentrated--remember, the data lake, the model you are
training it on--but then there is a lot of customization in
what you are looking for. Let us say in the trading side, if
you are a trading firm, you are looking for some advantage that
makes your algorithm unique.
Mr. Lynch. I guess what I was getting at is just the
exponential increase in costs, and there will be fewer and
fewer people that can do that if the costs continue to escalate
at the current rate. It is just exponential, but I appreciate
your opinion on that. By the way, Mr. Chairman and Ranking
Member, I am very grateful for all the experts here. I know a
bunch of you are frequent flyers to our roundtables, and you
have appeared at these hearings. So I am very, very grateful at
the diversity of perspective here and the value that you bring
to these hearings. I am grateful because we got a lot to learn.
Mr. Reynolds, I know you were at FinCEN for a while. These
smaller firms they are acquiring, they are not building their
own data sets or models. They are acquiring them AI-for-
service. On the one hand, it is great because with the help of
AI, they might be able to compete with some of these big firms.
On the other hand, they are going to be relying on a third
party for monitoring, for compliance, and for monitoring
possible bias in their AI systems. How do we reconcile that, or
do you think most of the firms now that provide AI as a service
are able to properly monitor what is going on in these small
banks or smaller financial firms?
Mr. Reynolds. Yes. Thank you, Congressman. It is a great
question and one that we think about quite a bit. As I
mentioned, we have a very strong governance process around our
models and around our AI. So when we deploy that technology to
customers of ours, which are predominantly banks or brokerages,
capital markets players, we obviously are very conscious that
we are sort of the first level of defense. So we think of
ourselves, I guess by analogy, as the belt and the bank is sort
of the suspenders in that you have both the bank itself having
an obligation to look at those tools and we provide them
information so they can make their assessment to ensure that it
is operating as designed without bias. We also are very focused
on that ourselves to ensure that it is. One of the ways that we
really do that is to make sure that a human is either in the
process or on the process, and I think that is critical for the
technology at this stage.
One of the great things is AI can learn, GenAI can learn,
so hopefully, as it learns about mistakes it makes, you can
correct those and it will not do that in the future, but to
really do that, you have to make sure that you have human
beings in the process to ensure those basic concepts of human
fairness that go into these decisions.
Mr. Lynch. Thank you very much, Mr. Chairman. I yield back.
Chairman McHenry. Thank you. Thank you for your leadership
as well. The gentleman from Missouri, Mr. Luetkemeyer, is
recognized for 5 minutes.
Mr. Luetkemeyer. Thank you, Mr. Chairman. Ms. Osborne, at
our National Security and Illicit Finance roundtable, a
panelist shared that his firm has seen a 450-percent increase
year-over-year in AI-powered deepfake attacks against financial
institutions. As generative AI compounds in its ability to
manipulate video and voice are better trained, this problem
will only get worse. I am concerned that the small community
institutions, such as Mr. Lynch referenced here a minute ago,
like yours are going to bear the brunt of these attacks. Do you
share these concerns?
Ms. Osborne. Thank you for your question, Congressman. At
Great Lakes Credit Union and across the credit union industry,
we prioritize safety and security for our members. Deepfakes
are absolutely a concern. It is something that is top of mind.
What we have done at Great Lakes Credit Union with our virtual
assistants, because that is one risk that we have to monitor
and manage, too, is we recently introduced voice
authentication. Voice authentication is one additional method
of authentication for our members to use. We were very careful
and considerate in the rollout of that method of authentication
because we needed to ensure that it provides additional layers
of controls for our members.
The way voice authentication works with Olive, our virtual
agent, is it is two-factor, right? So it is multi-factor
authentication (MFA) in partnership with voice recognition. The
member has to enroll in the program. That is the first piece.
Then that technology, once they enroll and they use the
platform, it combines anti-spoofing technology and compares it
against 25 to 30 different factors, such as the phone number
the member is calling from. In addition to that, the type of
transaction that is being requested is risk rated, and varying
levels of additional authentication are offered depending on
the transaction risk.
Mr. Luetkemeyer. What are the risks that you face with
generative AI that we are not thinking about?
Ms. Osborne. That is a very good question. I would say
generative AI is constantly changing. So for our institution,
for Great Lakes Credit Union, the challenge we have and the
risks we have is, how do we continue to ensure that we are
meeting the best of both worlds, providing the service to our
members, while also ensuring the appropriate controls and
guardrails are in place. So it is truly in partnership with our
third parties that we work through that, constant discussions
with those third parties and check-ins help to ensure and
mitigate that risk as much as possible.
Mr. Luetkemeyer. You know, I sit on the China Select
Committee as well. I had a briefing this morning, and AI is
used to surveil the Chinese people. They spend $300 billion a
year to surveil, detain, and build detention camps on their
people, and AI is a big part of that. So now in my seat here in
the Financial Services Committee, we look at this from the lens
of, on my gosh, what can the Chinese do to us here using AI,
which you guys have all talked about this morning all the
benefits of it. There are some risks to it as well and some
real concerns, and those are the ones that I think that we as a
committee have to think in terms of and figure out how we can
protect the industries that we oversee, the individuals, the
people of this country. So I am very concerned.
I think we had a situation last spring with the run on
Silicon Valley and those other banks where within 10 hours, $42
billion went out the door. I can see a scenario where the
Chinese are looking at artificial intelligence in many
different ways. You figure out which one you want to take. They
could have an artificial intelligence-generated commercial with
Chairman Powell up there or with Bloomberg or with Dave Ramsey
or somebody who is a recognized financial expert saying, uh-oh,
we have 100 banks today that are in trouble, and all of a
sudden everybody runs out, and we have FedNow, which is, as you
know, almost instantaneous. Instead of taking 10 hours to do
$42 billion, they are doing in 10 minutes with FedNow, a year
or so from now, perhaps.
So it really strikes me as being a wonderful tool. It also
has some really big challenges, and I am very concerned about
it. Do you think I have a right to be concerned? Mr. Zecca, you
deal with the Nasdaq folks. I mean, this has to be concerning
to them to have short selling going on with this sort of
situation.
Mr. Zecca. Well, thank you for the question. I think it is
very important. We obviously had situations before where people
have tried to put out fake information, but there is the risk
that deepfakes have a level of legitimacy or apparent
legitimacy that others do not. So we monitor social media
trending to try to see situations that we may have missed where
something came up.
Mr. Luetkemeyer. So it begs the question then, what do we
as legislators need to do to come up with regulations or
provide oversight in a way that monitors this or controls this
yet does not stifle the markets? We have to be thinking about
this, and I am sure you guys have already thought about it as
well. Mr. Zecca? I will have to ask the question later. I am
out of time, but thank you, Mr. Chairman.
Chairman McHenry. The gentleman from Illinois, Mr. Foster,
is recognized for 5 minutes.
Mr. Foster. Thank you, Mr. Chairman and to our witnesses.
Mr. Reynolds, on page 5 of your testimony, you identified
one area where you believe that congressional action is
critical, which is updating customer identification programs
and development of a digital identity system. Can you say a
little bit more about that?
Mr. Reynolds. Yes, thank you, Congressman. This is an area
that I think for the last couple of years we have noticed
growing weakness, I think, in the CIP and the CDD system. I
think it started with Web3, where you have users who are able
to exist on the blockchain have anonymous or pseudo-anonymous
identities and are able to divert away from the legitimate
financial system and really avoid some of the controls that we
have. One of the things that I think is critical for us is, as
these controls start to weaken as deepfakes become more
prevalent as the ability to fake identities become more
prevalent, we need to think is there a different way to think
about identity.
Identity, essentially, is in our system a proxy for trust,
how much do I trust you, and I think based on how much I trust
you is, do I know who you are. So I think if we start instead
from the basis of trust and think how can I instead understand
how much do I trust you without necessarily regard to identity
purely, that will be the basis for a new digital identity
system that will help solidify our existing CIP and CDD----
Mr. Foster. Now, in other countries, including, I guess,
the entirety of the EU, they are going to an approach where you
basically have the national government present something that
is very similar to the digital driver's licenses, a Real-ID-
compliant digital driver's license, where if you want to open a
bank account, you say, okay, get out your phone, do your
biometric login, show your face shape, and take advantage of
the trusted electronics in modern cellphones to present a
federally issued certificate that pretty much says here is my
proof that I am a single legally traceable person. This will
allow a single step know your customer (KYC) in 30 seconds,
which will really remove one of the big barriers to entry for
like fintechs setting up their KYC operations or small banks.
It is a real burden on them that could be avoided if you do
what actually a lot of the States in the United States are
doing, is just provide digital driver's licenses as a means of
proving you are who you say you are in an online environment.
Is that part of the solution we should be looking at for
dealing with deepfakes and other----
Mr. Reynolds. Absolutely. I think that concept of the
digital ID will make sort of banking portability a reality, and
I think it will allow consumers to move very freely throughout
the financial system. As you note, it will significantly reduce
costs, but significantly increase trust.
Mr. Foster. Yes. Ms. Osborne, did you have any comments on
it?
Ms. Osborne. Yes. So at GLCU, we do not currently use or
accept digital ID. However, it is an area that we are actively
researching and hope to in the future pursue. Our biggest
challenge at this point is really consumer adoption. You do not
see the use of it and nor are there many providers that we
could leverage to integrate into our platforms for this. They
exist, but they are still new and so we are doing our due
diligence and thoroughly vetting partners and solutions out
there to ensure we make the right decision for our members when
that time comes.
Mr. Foster. Yes. I think many people are also worried about
sort of splintering a set of commercial identity ecosystems so
that, you know, maybe log in with Google or log in with
Facebook becomes a de facto standard and that makes a lot of
Americans uncomfortable, that if there has to be a standard way
to authenticate you are who you say you are, then that maybe is
an essential government role to provide a way to assert your
Real ID, which is something you are going to need to get on an
airplane soon enough. If you can present that in a digital
online format, you know, essentially, the heart of the mobile
ID that was developed by NIST.
Let us see. This also relates, I think, to privacy in ways
that I think people underappreciate. Mr. Karunamurthy, you have
to deal with the various privacy bills that are moving through,
and it is a real burden if all of a sudden a customer has the
right to retract the data that your model has been trained on
and if, you know, spent millions of dollars. How do you
anticipate having to deal with that?
Mr. Karunamurthy. Thank you, Congressman. We absolutely at
Scale believe privacy is one of the important considerations to
make when considering the data used to train AI models and the
data that is used to observe and improve these models in the
future. While we do not train our own models, and so that
specific case is not something that we have to particularly
consider, we do take into account privacy when we are
considering different perspectives we can gain on AI models.
There are certain situations where you can use proxies for
actual customer data without using customer data itself and
still gain a lot of insight into how an AI system will perform.
Increasingly, we are using those methods to try to get good
observability into how these systems operate without needing
oversight of individual customer data.
Mr. Foster. Thank you, and I yield back.
Chairman McHenry. We will now recognize the gentleman from
Michigan, Mr. Huizenga, for 5 minutes.
Mr. Huizenga. Thank you, Mr. Chairman. Given the rapid
advancements in artificial intelligence and potential on
various sectors, what specific measures should Congress
consider to ensure that AI technologies are developed and
deployed ethically with adequate protections for privacy,
security, and fairness? This was what ChatGTP came up with as a
question when it was loaded in write a question for a
congressional hearing on AI. So that was my team's idea to
check in with the experts apparently, but just to kind of make
the point that this, well, I will let others decide whether
that is a more intelligent question than has been asked in this
room previously by actual humans, but it is certainly the
reality. This goes back to the chairman's point right at his
opening, which is, this is the reality of our lives today.
So I am going to see if we can get back to that, but I have
a couple of other quick things I need to do, and, Ms. Osborne,
I am going to start with you very quickly. Credit unions are
seeking to adopt more effective fraud management and the use of
AI through that. The impracticality of manually monitoring
transaction patterns is certainly something that we have heard
about for a number of years but what are the barriers to
incorporating AI-powered fraud analytics that you are
experiencing?
Ms. Osborne. Thank you for your question, Congressman
Huizenga. So fraud is one of the three areas that we are
currently looking to expand AI use in at GLCU, as I included in
my written statement. We do not currently use AI for fraud.
However, there are great benefits to incorporating that
technology to improve accuracy. Really, the goal here is to not
just detect, but to prevent those type of fraudulent
occurrences from happening.
Mr. Huizenga. Yes. Are there some barriers, though, that
are keeping you from implementing that?
Ms. Osborne. Yes. It is really about finding the right
partner. At our asset size, a 1.6-billion-dollar institution,
we are very dependent on third-parties. We go through a
rigorous process to ensure we find the right solution,
integration with that is also key. There are quite a few
players out there and it is constantly emerging and changing.
Mr. Huizenga. Okay. All right. Thank you. Mr. Zecca, in
your testimony, you noted an estimated $3.1 trillion in illicit
funds flowing through global financial system in 2023. Last
year alone, losses from fraud totaled $484-plus billion dollars
globally. How has Nasdaq used machine learning to address
illicit activity and then specifically criminals are also using
AI to develop new and more sophisticated financial crime
strategies, what are you doing to out-innovate those folks?
Mr. Zecca. Well, we have a whole team that tries to
deconstruct and analyze the information we get from our bank.
Under the Patriot Act, there is an opportunity for U.S. banks
to share information. So through a trusted third-party like us.
We are able to see the connections across all the
organizations. It helps us build more valuable patterns that we
can then implement.
Mr. Huizenga. That has been somewhat controversial, that
third-party, in allowing insight. I know that has been an
ongoing conversation. Sticking with you, sir, I want to discuss
the AI Act. According to a recent article in Financial Times,
European regulators admit the costs and compliance of the EU's
new AI regulation could hit six figures for a company as small
as 50 employees. Essentially, in my mind, being a small
business owner myself, this amounts to a tax on small
businesses, startups, entrepreneurs. It is another barrier that
is being put in the way, and we are not talking about massive
companies with massive compliance offices. In your view, what
lessons can the U.S. learn from the EU's approach to date? Now
I will let you answer.
Mr. Zecca. Well, yes, we are still digesting it in fairness
because it just came out, but I think that the two areas that I
would focus on, one is one act that tries to describe
everything that is going on is a very complicated thing to do,
and it does stick things entirely.
Mr. Huizenga. Because, in fact, I think you had written it
down. I think you wanted to avoid specific regulation, that was
just going to be a follow up on that, but you also said you
wanted consistent and harmonized regulation. I am curious how
that fits into flexibility because I agree. I mean, I think
Europe sort of locked itself in a box here in a way that does
not seem manageable.
Mr. Zecca. Right. It is true, and so I think I am talking
about the world we would like to see, not perhaps the world we
live in. I think organizationally, having a general system of
principles-based regulation, we are concerned about a central
regulator, in part, because they are not going to have the
expertise about the underlying subject matter even if they have
the AI expertise and see application that matters. So that is
not the direction that Europe is going. They do have a central
AI office. So I think that is another lesson we can learn here
and then there is a concern about State regulation just going
in different directions and so it will be very hard for
companies to comply with different AI regulation.
Mr. Huizenga. I appreciate that. I yield back.
Chairman McHenry. I will now recognize the gentlewoman from
Ohio, Mrs. Beatty, for 5 minutes.
Mrs. Beatty. Thank you, Mr. Chairman and Ranking Member,
and thank you to all the witnesses being here today, and let me
give a special, Mr. Chairman, if I may, shout out to someone I
call a friend, Lisa Rice. More importantly, a few years ago,
she received one of the highest awards that we gave for fair
housing, so it is good to see you again today.
I will start with you, Mr. Linda and Ms. Rice. I would like
to kind of follow up on some of the questions we have heard,
and especially from Ranking Member Waters. What legislative
solution should we be pursuing to ensure responsible use of AI?
Mr. Linda?
Mr. Linda. Thank you for the question, Congresswoman. We
are really spending a lot of time thinking about how our
existing regulation, such as Fair Housing Act, apply to
everything we do at Zillow, including the use of AI. We see AI
as the same or a solution that is similar to how we would
instruct and train great real estate agents. We believe in
proportional risk-based and flexible regulation and to the
extent that there are new gaps, we welcome the chance to be
part of dialog like this, one where private and public sectors
comes together, and really putting the customer at the center
of it because, ultimately, we want to put the customer first,
find ways to help them get home and find a home in this really
challenging market. That is the approach that we would like to
put forward.
Mrs. Beatty. Ms. Rice, would you like to add anything?
Ms. Rice. Congresswoman Beatty, thank you for the question
and for your support for fair housing over the years. So I do
agree with some of the other speakers on today's panel that we
should take a principles-based approach to legislation, but we
also need to incorporate an outcomes-based approach. One of my
members in Los Angeles just lost a case, in fact. They
conducted systemic testing of an apartment complex that was
using dynamic pricing systems. Every time they send a tester of
color to the apartment complex, the apartment rental rate was
higher for the same unit, every time they sent a white tester,
the apartment rental rate, again, for the same unit was lower,
and this was over and over and over again. When they brought
the lawsuit, the landlord said, look, it is not our fault. We
were using a dynamic pricing model. That model set the rental
rate, and we are not liable because it is a third party that
did it.
If we make sure that we have comprehensive legislation that
is both principles based but also outcomes based, we can close
the loophole on those kinds of discriminatory practices we are
seeing in the marketplace.
Mrs. Beatty. Let me go to another part since we are talking
about housing. Let me stick with you. What is the role of AI in
the underwriting process for mortgages?
Ms. Rice. We are seeing, particularly amongst our larger
financial institution partners as well as fintech partners,
that AI is playing a critically important role. So,
Congresswoman Beatty, I will lift up a positive. One of the
positive uses that we are seeing with AI is the ability to use
rental housing payment data and cashflow data in the
underwriting process and the utilization of those two
nontraditional characteristics is actually expanding access to
credit in a more responsible way. We know that credit scores,
loan-to-values (LTVs), debt-to-incomes (DTIs), highly
correlated to race, and they also result in lesser or lower
access to responsible credit.
Mrs. Beatty. You mentioned race, so let me, while the clock
the ticking, in the latest funding bill draft, House
Republicans have eliminated funding for fair housing
enforcement and blocked HUD's ability to enforce affirmatively,
furthering fair housing provision of the Fair Housing Act. Can
you discuss the importance of Federal fair housing enforcement
to combat automated discrimination in housing lending through
online housing platforms with AI technology? I have about 30
seconds.
Ms. Rice. It is extremely important, and the thing that I
will say is that many of our business partners have recognized
that if we eliminate discrimination, we will actually increase
profits, will increase productivity. Citigroup released a
report that showed if we eliminate racial inequality, we would
increase the U.S. Gross Domestic Product (GDP) by $5 trillion
over a 5-year period.
Mrs. Beatty. You are getting some nods, so.
Ms. Rice. Right. That is one of the reasons why the
Business Roundtable in its platform supported affirmatively
furthering fair housing and disparate impact so that those
businesses can responsibly increase their footprint and profit.
Mrs. Beatty. Thank you. My time is up. Thank you very much
for that.
Chairman McHenry. I will now recognize the gentleman from
Minnesota, the majority whip, Mr. Emmer, for 5 minutes.
Mr. Emmer. Thank you, Mr. Chairman. Thank you for holding
this important hearing today, and like the others, I want to
thank all the witnesses for your testimony.
Mr. Karunamurthy, if I said it right, and hopefully I did,
you have had a long and very impressive career in machine
learning and artificial intelligence. Just to get us started, I
have a quick ``yes'' or ``no'' question. Have you contemplated
any potential intersections between AI and digital assets?
Mr. Karunamurthy. Yes.
Mr. Emmer. All right. I figured that was the answer. The
next is between AI and digital assets, it seems necessary and
inevitable to meet. In fact, there could realistically be a
future symbiotic relationship between the two. In many ways, AI
has opened their eyes to the reality of digital inauthenticity
and crypto ensures digital authenticity. In other words, the
immutable nature of digital assets transactions offers digital
assets and blockchain, it offers them up as at least one tool
for authenticating and if not potentially being the key tool,
which is becoming increasingly unclear with the rise of AI. Of
course it goes without saying that AI offers significant
promises and benefits to our economy. AI, along with digital
assets, are leading us toward a powerful global digital
economy. We just need to ensure this technology is crafted by
Americans and with American values.
Scale AI is at the forefront of AI development, given AI's
need for reliable data. When I think of what makes data
reliable, I am not only concentrated on the data inputs
themselves, but also the processes by which data is gathered
and verified. Last month, there was an article in Cointelegraph
that discussed the critical role of blockchain technology, a
critical role that technology can play to ensure that data used
in AI models is authentic, immutable, traceable, and
transparent. This would therefore give certainty. Sir, as you
consider how data labeling will evolve over the medium and long
term, do you think blockchain technology can be used as a tool
to ensure data authenticity in the future? If so, how?
Mr. Karunamurthy. Thank you, Congressman, and I appreciate
the opportunity to talk about the important role AI plays for
U.S. economy and American competitiveness. We consider this the
most transformative technology we faced in the last 20 years,
and so it is important to get ahead of these opportunities and
also highlight safety risks. The provenance of the data that we
use to train these models and that these models leverage in
order to come up with decisions is increasingly important,
paramount for us to observe, monitor, and to find authoritative
sources of information. So whether we are talking about the
blockchain or digital identity solutions, those all have an
important role to play in ensuring that data is accurate and
up-to-date.
We also have developed a variety of techniques that
understand how these AI models leverage that data. One
important one that we have let in our written testimony was red
teaming to understand that those models can be tricked or
fooled in certain ways to believe that someone is who they are
not or to reveal private information and so we are developing a
lot of those techniques that will come at the intersection of
blockchain and other solutions and helping establish the
providence of the data use of the models and how that drives
decisionmaking.
Mr. Emmer. Thank you. I believe the convergence of
blockchain and AI cannot only improve the trustworthiness of
data, but the decentralization of artificial intelligence data
can mitigate single point of failure security issues and
additionally as AI systems communicate with each other and need
to transact with each other to obtain information, digital
assets and AI can have that symbiotic relationship. Sir, what
do you envision is the primary way that digital asset
technology can engage and interact with artificial intelligence
systems in the future?
Mr. Karunamurthy. Thank you. An important way is to make
open and transparent what information is being used by these AI
models so that as a public we can evaluate where these models
are giving safe and responsible answers or where these models
might have been misled. Having an open and transparent system
to understand where data is coming from or where data has
evolved over time can be incrementally important to the U.S.
economy and it is important for the evaluation of these
systems.
Mr. Emmer. Well, thank you. Finally, Mr. Karunamurthy--
sorry, I was going to blow it at some point--to develop and
lead in the next iteration of the internet powered by AI and
digital assets, we need this technology to be designed and
built by Americans and with American values. How can government
support entrepreneurs and businesses like yours in the global
talent world to not only attract, but to retain high skilled
tech talent right here in the United States?
Mr. Karunamurthy. Yes. We are absolutely in a global talent
world, and we have seen increasing evidence that countries like
China have really powerful AI talents, and they are putting in
decisions of authority where they can train future AI models.
We believe here in the U.S., it is important to highlight the
role that developing talent and upskilling workers plays in
developing the future of this technology, which goes beyond
just researchers. It also means American workers needs to
understand how AI can be used to assist them in their job
functions and understand where these models are capable or
where they might not be capable today. So we are spending a lot
of effort in helping upskill and retrain workers in these
technologies.
Mr. Emmer. Thank you, and thank you, Mr. Chair.
Chairman McHenry. The gentleman from Illinois, Mr. Casten,
is recognized for 5 minutes.
Mr. Casten. Thank you, Mr. Chairman, and thank you, also,
Mr. Chairman, for putting together the AI Working Group. I
think we have had really a very productive set of hearings.
Thank you all because I know a lot of you have come out many
times for this.
So in a lot of our Working Group hearings, I have raised
this concern about disparate impact and the idea that you can
have discriminatory outcomes independent of discriminatory
intent, and there is a long history of companies being held
accountable one way or the other for that. I have this growing
concern that you see a lot of people saying, well, the AI did
it, I did not do it, I did not program it to do this. Before
digging into the questions, I am not a lawyer, and, Ms. Rice,
you are way smarter than this, and I thought you had in your
written testimony, like, an interesting block on this. I wonder
if you would just share with us briefly the legal distinction
between disparate treatment, disparate impact, the relative
enforceability, and some of the things you have shared
previously about what to think about before I get to the
questions I want to go through.
Ms. Rice. Congressman, thank you. Sure. I would be happy to
do that. So an example of disparate treatment is where
technology firms treat people differently based on their
protected class characteristics. A case example of that is in
our lawsuit against Facebook in which Facebook, it designed a
model so that if you are advertising for an employment, a
credit, or a housing opportunity, you could turn off African-
American audiences. You could turn off Asian-American
audiences. You could turn off Hispanic audiences. You could
never turn off white audiences. So that is an example of
differing treatment. You are treating people differently based
on their protected class status.
An example of disparate impact is where you have a rule or
a system that is not race based, but it is generating
discriminatory outcomes. An example of that is our lawsuit
against Prudential in which we challenged the use of their
insurance scoring system. If you looked at the underwriting
algorithm, there were no variables that were akin to things
like race or gender, but it was generating discriminatory
outcomes. In that case, Prudential had to prove that it had a
business necessity for using the insurance scoring model.
Mr. Casten. I am sorry because I am just sensitive of the
time, but is it safe to say that it is a higher legal bar to
get enforcement done to disparate outcomes than disparate
treatment?
Ms. Rice. Yes. Under disparate impact, yes.
Mr. Casten. Yes, lacking the smoking gun. Okay. So for
those of you who are building and running AI algorithms, do any
of you currently have in that code blocks that test for
disparate treatment or for disparate impact? Just show of
hands. Do any of you include those blocks of code? Mr. Zecca,
you do include those blocks of code? Okay.
Mr. Zecca. We do. We are not in banking, but, yes, for our
order types to ensure that all customers are treated equally.
Mr. Casten. Okay. Ms. Osborne, you did not raise your hand.
Is there any particular reason?
Ms. Osborne. Yes. So at our sites, we do not develop in-
house AI.
Mr. Casten. Okay. So it is not a relevant question for you.
Fair enough. Okay. So I am glad that you do, and do you test
for disparate treatment or disparate outcome, right? Like in
the example, like all of you who raised your hand, given that
it is a higher legal bar, do you also test for it that you are
getting different outcomes?
[Nonverbal response.]
Mr. Casten. Yes, across the bar. Okay. Part of the reason I
ask all of that is because that is a legal bar, right? But
there is this thing, Project 2025, that my colleagues across
the aisle and the Republican candidate for President are making
a big push on, when one of their provisions on page 538 is to
remove disparate impact as a valid theory of discrimination for
race and other claims. By the way, Justice Thomas, Justice
Alito have indicated in some of their dissenting opinions that
they also do not think those are valid claims.
If those laws were to be taken away, can you commit that
you would continue in your algorithms to test for disparate
impact, or is this only something that we are doing because it
has currently got some legal protection against it? Would all
of you maintain those models even if the government were to say
discrimination is now fine because thanks to Project 2025, we
are all good with discriminating?
[A chorus of ayes.]
Mr. Casten. We have yeses? We are going to hold you to
that. Thank you. I yield back.
Chairman McHenry. Now we will go to the gentleman from
Pennsylvania, Mr. Meuser, for 5 minutes.
Mr. Meuser. Thank you, Mr. Chairman. Thank you all very
much for being here. It is interesting, fascinating, good to
hear. So the EU's AI Act, which the Financial Times reported as
rushed and vague, serves as a clear example where regulations
are very often the problem. They specifically excluded
technologies like generative AI, as we have been discussing,
focuses instead on existing tools, highlighting the limitations
of broad regulatory approaches. Considering AI's impact, we
need to recognize the disproportionate burden that regulation
places on small businesses. Small businesses typically bear
regulatory costs more intensely as evidenced by many things
that this administration put forth, so just the Consumer
Financial Protection Bureau's (CFPB's) 1071 rule, FinCEN's
beneficial ownership requirements, and Basel III Endgame.
Mr. Zecca, I would like to just start with you. The EU's AI
Act is currently being implemented. How would you grade the
European approach so far?
Mr. Zecca. Well, I think the core question is how do you
prepare for the future, and the AI Act is sort of a point in
time. My main worry is as the technology develops, if you look
back over 2 years, it has already moved from where they started
because the technology has moved. How do you prepare for the
future? I also worry about the context of a single AI regulator
that is responsible for the enforcement because I do believe
that, like, for example, we operate markets in Europe. There
are prudential regulators for the markets. What happens if they
disagree with the AI office? So I think there are a number of
areas there that are unclear on how enforcement is going to
work and how technology is going to innovate.
Mr. Meuser. How would the EU's AI Act affect Nasdaq?
Mr. Zecca. For our operations in Europe, it will have an
impact. It will have an impact on a lot of U.S. companies that
operate in Europe.
Mr. Meuser. Okay. Do you believe these standards should be
calibrated for the particular type of use case, or should it
focus on regulating the technology broadly?
Mr. Zecca. Well, thanks for that question. I think that is
the critical idea. I think it is risky to regulate the
technology because the same technology can be used for
different use cases. You should regulate the risks and
opportunities of the use case. For example, anti-financial
crime (AFC), financial crime, one of the biggest things we want
to ensure is that the financial sector can use AI to fight
financial crime even if there is not pure explainability
because we are trying to keep up with criminals.
Mr. Meuser. All right. Thank you. Ms. Osborne, your credit
union is similar in assets to many community banks but larger
than many of mine. Can you describe how AI has helped you scale
your business and better serve your customers?
Ms. Osborne. Absolutely. Thank you for your question, Mr.
Congressman. At our asset size, $1.6 billion, currently in
terms of AI, we are using it primarily for customer service by
introducing our virtual assistant, Olive. We have been able to
streamline and really focus on the right level of support for
our members for those that require a human to interact with
them. We provide that ability. I see that as a great case study
across all financial institutions any time we can automate. We
can improve the outcome of decisioning, and any type of data
input today that is very manual, that is a great benefit,
especially for a not-for-profit organization like Great Lakes
Credit Union.
Mr. Meuser. Is it expensive to take it on board?
Ms. Osborne. It was not, no. It was in line with what my
expectations are for a technology of that size.
Mr. Meuser. Are your competitors engaged in equipping
themselves with AI as well?
Ms. Osborne. They are starting to. So our current provider,
Interface AI, they serve both banks and credit unions, and they
have currently over a 100 clients. There are several providers
like them, so you are starting to see that adoption ramp up.
Mr. Meuser. All right. When I was in business, particularly
in the early 2000s, we had a saying that if you do not use IT
as a weapon, somebody is going to use it against you, so I
would imagine your outlook on AI is pretty much the same. So if
it were regulated in line with the EU Act, for instance, where
cost prohibitions came from the regulations that were
excessive, how would that affect your ability to serve your
customers in your day-to-day operation?
Ms. Osborne. Yes. So today's regulations are very
technology agnostic, so that is benefit for us. If additional
regulations were proposed, that would increase the cost for our
institution to serve our members. Those costs would have to
then be passed down, and that would be a burden to our
consumers.
Mr. Meuser. Okay. Running out of time, but I would love to
also hear from perhaps all of you on some of the downside that
you think about. We are talking about the positive aspects.
What are some of the downside? Somehow we can get that
afterwards.
Mr. Meuser. I yield back, Mr. Chairman.
Chairman McHenry. The gentlewoman from Massachusetts, Ms.
Pressley, is recognized for 5 minutes.
Ms. Pressley. Thank you, Chair McHenry, and to all our
witnesses for joining us today.
As we examine the role of artificial intelligence in
financial services and housing, we must be honest about the
truth. The algorithms that value our homes, filter our job
applications, and approve or deny our loans exist in a larger
context of biases that have plagued our communities for
centuries. This is evident in the home appraisal industry where
systemic racism robs families of color of generational wealth
and economic opportunity simply because of the color of our
skin. The data is clear and the data is damning. Homes in Black
neighborhoods are consistently undervalued by at least 21
percent compared to similar homes in predominantly white
neighborhoods. This translates to more than $162 billion in
loss equity for Black homeowners, opportunities denied, dreams
deferred, and the perpetuation of the vast racial wealth gap
that exist in our country. Now, some may argue that AI and
automated valuation models offer an objective solution to bias
and discrimination, but I caution against blind faith in these
systems. These systems are only as biased as the data used to
train them and the human beings who design them.
Mr. Karunamurthy, is that right? I have been practicing all
day. Names matter and I want to get it right, okay? Is that
right?
[Nonverbal response.]
Ms. Pressley. Okay. You are the chief technology officer at
one of the most financially valuable AI startups in the
country. Are you taking a race-conscious approach to
eliminating bias while developing and deploying AI technology?
Mr. Karunamurthy. Thank you, Congresswoman Pressley. We do
believe these are really important issues to consider in the
future of this technology. So from the forefront or from the
beginning of our approach to evaluating AI models, we have
considered what different sort of perspectives are important to
ensure that we are not just putting blind faith in this
technology, that we understand how decisions are made by AI,
what sources of data they are using, and that data avoids bias
or the course of action.
An example of that would be ensuring that when these models
use geographic data, they are not biased by prior sources of
data and maybe information on regions or districts that may be
out of date or that may not give the full picture of what may
be present in a given area. Then, we try to ensure that these
models explain their decisionmaking to human operators so that
there can be human oversight of how these models came to the
decision. Thank you.
Ms. Pressley. Thank you. Dr. Linda, Zillow has been a major
player in the home appraisal industry by making home valuation
data readily accessible on the internet. How is Zillow ensuring
that its AI algorithm does not perpetuate financial services
industry's long legacy of redlining and housing discrimination?
Mr. Linda. Thank you for the question, Congresswoman. It is
absolutely a problem as you pointed out, and we work tirelessly
to measure, understand the bias. As an example, our Zestimate,
our main tool that we developed, started a company to empower
our customer with information so that they can often challenge
some of the bias that you pointed out. Every change to the
algorithm is not only evaluated on does it improve the
accuracy, but also does it improve fairness, especially with
respect to the dominant race of the demographic of the people
living in the neighborhood. We did make some changes. As you
noted, the neighborhood is an important aspect. We make sure
our algorithm does not look just at the small neighborhood, but
also look at a much bigger area, which allows us to be much
more comprehensive and resulted in more objective, accurate,
and fair assessment of a value.
Ms. Pressley. Thank you. Ms. Rice, you heard these two
responses. Do you agree that technology companies like Scale AI
and Zillow are doing everything they can to ensure AI tools and
housing appraisals do not become vehicles for discrimination?
Yes or no.
Ms. Rice. Yes, regarding Zillow. I do not know for Scale.
Ms. Pressley. What steps should these companies be taking
additionally in your opinion?
Ms. Rice. I think Zillow has a very great model, and we
partner with Zillow a lot on responsible AI and fairness
principles. So testing the underlying data to make sure that it
is representative, make sure that it is accurate and make sure
that it is not missing critical information, testing the
underlying data used to train the models to make sure that
there are not proxies that are highly associated with race. So
that is a pre-processing phase. In processing, making sure that
you are testing those systems for their outcomes and you
understand what the outcomes are, and if you are seeing
discriminatory outcomes, recalibrating.
Ms. Pressley. I have to reclaim my time. Thank you so much.
It is clear that we need more transparency, accountability, and
equity for both human and AI-driven home appraisals. We need
robust, independent bias testing of the AI systems, diverse
representation in the appraisal industry, and a commitment to
advancing racial equity from our regulators and industry
stakeholders if we already use AI responsibly and safely in our
communities. I still believe in the promise of homeownership as
a path to financial security and generational wealth.
Chairman McHenry. The gentleman from Texas, Mr. Williams of
Texas, is now recognized for 5 minutes.
Mr. Williams of Texas. Thank you, Mr. Chairman, and thank
you all of you for being here, and I am a great supporter of
innovation. I am a car dealer in Texas, and artificial
intelligence has been at the forefront of innovation. Across
all industries, AI has assisted in making business operations
easier and less time intensive, including in the financial
services industry, and AI provides increased workstream
efficiency and improved client engagement. It is important for
us to understand this evolution so that we may continue to make
the practice of running a business more efficient and foster
greater competition within industries. Ms. Osborne, could you
elaborate on how artificial intelligence helps you and your
credit union make lending decisions?
Ms. Osborne. Thank you for your question, Congressman
Williams. At GLCU, we do not currently use AI for lending.
However, it is an area we are actively pursuing. I think it is
important to note that 2 weeks ago, Karen Harbin, CEO for
Commonwealth Credit Union, testified and spoke of her use of
Zest AI and how that has helped her credit union be innovative
as well as expand lending to areas, demographics, disadvantaged
individuals that in the past, because she is lending outside of
the credit score, she was unable to do so beforehand.
Mr. Williams of Texas. Okay. Thank you. Mr. Linda, you
mentioned in your testimony that Zillow has been using
artificial intelligence in some capacity since the company was
founded, and so you are no stranger to the expansion and ever-
changing landscape of artificial intelligence. Some view
artificial intelligence as dangerous and confusing, but you and
your company have used it to turn something also confusing,
like finding a house and obtaining a mortgage into a more
simplified consumer-friendly process. Mr. Linda, how has Zillow
used artificial intelligence to make the often opaque and
stressful process of finding a home easier for consumers?
Mr. Linda. Thank you for the question, Congressman. As you
pointed out, from the beginning of the company, we started by
putting customer and transparency at the forefront of what we
do and using AI to do that. Empowering customer information,
how much a particular home might be worth was the start, and
then we continued by using AI to help you find all the homes on
the market. It is not easy to understand all the inventory out
there, and AI, again, is a great tool to do it. During the
pandemic, it became extremely useful to have an AI tool that
allows you to virtually and remotely tour a home. These are
just some of the examples where AI makes the process more
transparent, accessible, and we are looking forward to how we
can further use the technology to further empower our customers
to get them home.
Mr. Williams of Texas. Okay. Thank you. One of the biggest
challenges in running a successful small business is the number
of roles owners have to play, from CEO to CFOs to CMO to
managing inventory and fulfilling or delivering orders. By
leveraging AI tools, small businesses can access the same
technology that big business use to their advantage, making
them more efficient and giving them the ability to make more
informed financial decisions. Operating a small business is
becoming increasingly harder due to challenges like high
inflation, supply chain disruptions, and worker shortages.
Artificial intelligence tools give small business the
possibility to run smoothly and be able to tackle other issues
that they face. All of this shows why small businesses should
not be overlooked in discussions surrounding AI because they
understand the competitive benefits and risk perspective, so,
and I will say I am chairman of the Small Business Committee,
so this means a lot to us. Mr. Reynolds, as Congress looks to
potentially regulate emerging technologies like AI, how should
we try to ensure that small businesses can access and use AI
applications that allow them to compete with a larger market
participant?
Mr. Reynolds. Thank you, Congressman. I think critical to
that is that the regulation has to be designed to allow rapid
innovation. I think, obviously, to the extent that we can
innovate and lower the cost of the technology that allows
different market participants to be able to embrace the
technology and use it. If you look, for instance, at FIS, we
have some very large clients. We have some very small clients
and it is critical to us that we provide AI services and allow
access to AI, both for large institutions, but also for small
institutions because that really does level the playing field.
Mr. Williams of Texas. I have a limited amount of time. I
am just going to ask you another short question, Mr. Reynolds.
Can you expand on the use of artificial intelligence by
financial institutions to assist in risk management and
preventing unlawful activity?
Mr. Reynolds. Yes, sir. We think, actually, that artificial
intelligence, in many ways, is going to be critical to the
evolution of combating financing of terrorism, combating money
laundering, combating fraud. One of the most significant
problems in that is the false positive rate. We all have seen
that for years. It runs in the high 90's percent, sometimes
even more than that. We really think that the use of generative
AI, especially in fraud, is going to dramatically lower the
false positive rate, which will, again, allow more transactions
throughout the economy, but allow better transactions. We have
seen some very positive results in early testing on fraud
systems where we have dramatically reduced false positives and
increased true transactions.
Mr. Williams of Texas. Okay. I yield my time back. Thank
you very much.
Chairman McHenry. The gentleman from Texas, Mr. Green, is
recognized for 5 minutes.
Mr. Green. Thank you, Mr. Chairman. I would like to
associate myself with the remarks, to a great extent, of the
chairman and also the ranking member. The ranking member has
touched on some issues that are very important to me, and I
would like to look a little bit further into these issues.
Ms. Rice, you indicate that artificial intelligence, and I
am reading from your message that you have codified, artificial
intelligence is the new civil rights/human rights frontier.
That is a strong statement, ``new civil rights/human rights
frontier.'' You go on to explain about data going in and how
the data, over the years, has been a part of corruption, and
when you put something in that is discriminatory, then you will
probably get information that is going to reflect what you put
in. I will be quite honest. It saddens me to know that we have
suffered all of these years of actual discrimination--slavery,
convict leasing, invidious discrimination in the form of
segregation--and now artificial intelligence may not only
perpetuate, but do it in a more horrific way. Could you please
elaborate more on why you would conclude that this is the civil
rights/human rights movement of our time, perhaps, you say the
next frontier?
Ms. Rice. Certainly. Much like the industrial revolution,
the industrial frontier changed the landscape of our Nation, of
our economy, of our neighborhoods, and societies. AI will be
doing the same thing, we believe. It is going to drive
innovation. It is going to be the main driver of every single
market in our Nation. From that perspective, we believe that AI
and technology are the new civil and human rights frontier.
You are absolutely right, bad data in means bad data out,
which is why at the National Fair Housing Alliance, we have
been innovating on new systems and constructs that compel
legacy systems and new systems that are being developed to be
fairer. Just a few months ago, we released a groundbreaking
report on a new strategy that we have developed with one of our
partners, FairPlay AI. That strategy utilizes distribution
matching. It is a way of sort of globally optimizing for
systems for both profit and fairness to make sure that they are
not generating biased outcomes.
Mr. Green. Mr. Linda, you seem to have a similar position.
Reading from your report, you indicate that we place a great
emphasis on transparency, meaning your organization, and
fairness in the housing industry, which has a history of
discrimination and bias against people of color and other
historically marginalized groups. You go into the LLM, which is
the large language model, and you give some explanations, but
your views are very similar. Would you kindly explain to us why
you have concluded that this is something that can be of harm
to people of color and other marginalized groups?
Mr. Linda. Thank you for the really important question. I
think any kind of progress is to start from a point of
realizing what we are starting with, and any data we use when
building AI system will have imprinted past history of
segregation, and we have to be accountable to it. We have to
understand it. To your point, AI can pose a lot of risk, but AI
can also shed light on some of these things. We recently shared
some of our research where we used AI to understand at a
national scale what are the common phrases that are being used
by listing agents when they put listings online, and we saw
stark differences. Just from starting from a point of
understanding, that helps us to put that early on in our
development cycle and build better products and better AI model
because we are aware of this.
Second, we want to make sure that auditing models for
fairness is a part of our everyday work. That is what we do at
Zillow, and that is what we try and do, share it with the
community and publish our methodologies. It is a vital part of
developing responsible AI system for the benefit of our
customers.
Mr. Green. Thank you. My time is up. Thank you, Mr.
Chairman. I trust we will use it for good. Thank you.
Chairman McHenry. The gentleman from Tennessee, Mr. Rose,
is recognized for 5 minutes.
Mr. Rose. Thank you, Chairman McHenry, and I also thank
Ranking Member Waters for holding this hearing, and thank you
to our witnesses for being here today and sharing your time
with us and your expertise.
I am proud to say that in my home State of Tennessee, there
are already financial institutions that are leveraging
artificial intelligence to bring great benefits. One such
institution is Oak Ridge National Laboratory Federal Credit
Union, or ORNL--I will try to say it--based out of Oak Ridge,
Tennessee. Thanks to artificial intelligence models, ORNL will
be able to serve more customers and provide customers with a
better experience overall. I am proud to say that Tennessee is
home to financial institutions such as ORNL FCU and hope that
as we discuss and debate artificial intelligence regulation, we
do not hamper innovation.
Ms. Osborne, one area that interests me regarding the
utilization of artificial intelligence is expanding access to
financial services to those who live in rural communities
versus high population areas. How does leveraging AI help to
expand access to rural communities?
Ms. Osborne. Thank you for your question, Congressman Rose,
and also for your story. That is a great story to hear. At
Great Lakes Credit Union, we have a similar story. We have
introduced AI and seen great results from that. In terms of
expanding access to more rural areas and just broadly overall,
I can speak from Great Lakes Credit Union's perspective. By use
of Olive, our virtual AI agent, we are able to serve members
where we are able to provide support to them 24/7. So
regardless of where they live, what time zone, what their needs
are, they are able to have a conversation with Olive and get
the support they need. That also plays, in fact, with other
areas of AI that we are exploring that would allow us to lend
to more members, members beyond the credit score that we were
not able to in the past that maybe are newer or in
disadvantaged communities.
Mr. Rose. Thank you. I appreciate that insight. There have
been some concerns that consumers are limited in recourse
against unfavorable decisions made by artificial intelligence.
However, the Fair Credit Reporting Act and the Equal Credit
Opportunity Act regulate adverse action when a creditor acts
against a consumer based on their credit report. Mr. Reynolds,
is it not true that adverse action regulation applies equally
to human-made decisions and system's rules-based decisions or
AI decisions?
Mr. Reynolds. Yes, it does.
Mr. Rose. Thank you. I think that is pretty plain and
simple. Ms. Osborne, as a former bank director, I know
firsthand how burdensome and difficult it can be to comply Bank
Secrecy Act information and share it with the appropriate
agency. How can artificial intelligence be leveraged to help
satisfy regulatory requirements?
Ms. Osborne. Yes. Thank you for that question. Yes. So
Suspicious Activity Report (SAR) is a perfect example. Any
other type of reporting, if we have the ability to leverage
technology like AI to automate, to make it more accurate and
timely, there is tremendous benefit. That would reduce resource
times, would improve our response times to meet regulatory
requirements as well, and because it is more accurate, less
review is needed. Then we can utilize resources in better ways,
such as improving or providing additional support for our
members.
Mr. Rose. I agree with that. Mr. Reynolds, as financial
institutions increasingly leverage artificial intelligence to
better identify suspicious activity reports, will it not be
imperative that they receive strong feedback from Financial
Crimes Enforcement, or FinCEN, regarding the suspicious
activity reports submitted so that their models are
appropriately calibrated?
Mr. Reynolds. Yes, I absolutely agree with that and I think
that this is something which AI and generative AI, in
particular, really should enable the agency to much more
quickly and much more thoroughly provide feedback real time, to
your point, to make sure that models continue to learn the
right lessons.
Mr. Rose. I could not agree more. Frankly, I believe that
the Bank Secrecy Act needs to be reformed in a wholesale
fashion but until then, FinCEN simply must provide productive
feedback to financial institutions so that they can reduce the
regulatory cost, the cost of regulation through tools such as
artificial intelligence. I can say firsthand as a bank
director, I saw this repeatedly. I suspect everybody up here
who has had to deal with this directly, the lack of effective
feedback from FinCEN is a major problem for institutions in
knowing what to report. Thank you, Mr. Chairman. I yield back.
Mr. Steil [presiding]. Mr. Horsford is recognized for 5
minutes.
Mr. Horsford. I want to thank the chairman and the ranking
member for holding this very important hearing, and to thank
you to our witnesses for appearing to discuss the ongoing
adoption of artificial intelligence in our financial sector.
AI has the capacity to change the way we work, the way we
interact as consumers, and the very way that we live. As we
examine the rapid proliferation of new AI development, we owe
it to the constituents that we serve to remain forward thinking
on these trends, which will undoubtedly have a significant
impact on their livelihoods. We have an ethical, moral, and
societal responsibility as leaders to establish safeguards to
ensure that the advancement of AI protects all of our
communities, including communities of color. Everyone deserves
equal opportunity to benefit from the transformational change
of what I believe will 1 day be a powerful public good.
The constituents I represent in Nevada and across the
country as chairman of the Congressional Black Caucus are
counting on us to be forward thinking and to establish a
responsible AI policy framework today so that everyone can
enjoy an equitable deployment in the future. Recognizing the
challenges ahead, the Congressional Black Caucus (CBC) launched
the AI policy series to bring together leaders from government,
industry, academia, and civil society to provide an overview of
best practices to help shape these policies.
Under the thoughtful leadership of vice chair,
Congresswoman Yvette Clarke, Congresswoman Barbara Lee, and
Ranking Member Cleaver, we have explored the importance of
design, the implications of job equity, and the potential
impacts on the racial wealth gap. I want to thank the ranking
member, Congresswoman Maxine Waters, for her leadership in
bringing us to Silicon Valley to meet with industry leaders
specifically. Similarly to the work done by this committee,
these briefings have been focused on equipping Members of
Congress with the expertise that is necessary to apply existing
laws to the utilization of AI and the understanding to identify
regulatory gaps that need to be fixed before they lead to
consumer harm. However, our work distinguishes itself by
centering marginalized voices and examining the dangerous
implications of bias that exist within AI. We must ensure that
not only do our communities have the opportunity to participate
fully in the transition to this technology, but to also develop
the protections and the guardrails that can prevent the worst
impacts of inherent bias in a proactive way. Both in lending
decisions and automation of certain careers, unchecked reliance
on AI can certainly lead to algorithmic redlining.
I wanted to ask Mr. Karunamurthy, how do you believe AI
developers can proactively address the social and political
values, including bias around race, that are appearing in
generative AI today?
Mr. Karunamurthy. Thank you, Congressman Horsford. These
are really important issues for developers to consider, but
also folks adopting AI technology in industry and in financial
services and housing companies.
Two key parts of this process that I want to highlight. The
first is considering the data that goes into training these
models and that these models leverage to make decisions.
Ensuring that data is authoritative and up-to-date is
incredibly important, also observing that you have robust and
diverse perspectives in that data can be very important to
ensure these AI models give objective and fair answers to given
questions. The second is the importance of testing these models
and these AI systems as they are built, observing situations
where they may give a biased answer, and understanding how the
decisionmaking process of these AI systems led to bias. This is
an important first step in mitigating potential harms of these
systems and it is also important for human oversight of these
systems going forward.
Mr. Horsford. Thank you. These are the conversations that
must be had if we are going to prevent discrimination within
our financial system from becoming baked in the very start as
we continue to train new models. It is so much more that can be
accomplished with these technologies if we are able to harness
their capabilities to create, not hinder, economic advancement.
Again, I think there is a lot more that we need to explore
around the Fair Housing Act and the protections that are
afforded and the dangerous positions that are proposed by
Trump's Project 2025 that impact this proposal under that
agenda. So I look forward to having further discussion with
those of you on the panel. I yield back.
Mr. Steil. The gentleman yields back. I now recognize
myself for 5 minutes.
I will start with you, if I can, Mr. Ondrej. Where are we?
I am looking down here. Sorry about that. Thank you for being
here.
One of the challenges we face, in particular, is home
prices. The median mortgage has doubled in the past 3-and-a-
half years from $1,200 to $2,400 a year. Huge challenges with
where we have been placed from the Biden Administration as it
relates to inflation. That will ultimately, God willing, come
down as we get reckless spending under control. Other areas I
think we have room for improvement is on the transaction costs
on buying and selling homes. How has Zillow utilized the AI
application to reduce the cost and complexity associated with
selling a home?
Mr. Linda. Thank you so much for the question. We
definitely recognize that affordability and the constrained
supply is a big challenge for our customer and makes the
already daunting process even more complicated. Where we see
that we, and especially AI, can play a big role is in improving
the ability of our customers to understand all the supply on
the market.
Imagine a customer who searches with a specific zip code,
and they might only be shown that specific result. We use AI to
augment their search results to make sure they are aware of the
other options they have and perhaps lead them toward a home
they might otherwise not consider. Similarly, as already was
mentioned, the cost associated with the process. The more we
can automate mundane, repetitive tasks for the lending
officers, for real estate agents, the more efficient we can
make them and potentially some of these costs can be
transferred to the customer.
Mr. Steil. That is ultimately the goal, Mr. Linda. In
particular, as we look at this, what are the concerns you have
that an impediment might be placed from a regulatory
standpoint, maybe with good intentions, but that would
ultimately have a negative impact on the end user and the
consumer.
Mr. Linda. Yes. We truly believe that the existing
regulations, such as the Fair Housing Act, should be applied
equally to the business processes, the real estate
professional, as well as to AI. So what we are looking for is
exploring how that could be applied, and, of course, we are
supportive on exploring additional risk. We fully support
proportional risk-based and flexible regulation that further
enables a company like us and everybody else to innovate for
the customer without stifling innovation.
Mr. Steil. Thank you very much. Let me jump over to you,
Mr. John Zecca, if I can. Technology, specifically AI, can
really drive down costs in particular and be a big benefit for
markets. Can you briefly touch on how Nasdaq's identifying
applications of AI that can improve the way the exchange works?
Mr. Zecca. Sure. So I put them in two buckets. One is the
trading activity itself, and we recently had approved the first
AI-powered order type, and that was a process with the SEC to
get them comfortable with the changes. It gets a better fill
rate. I think you will start to see more order types that take
market conditions into account, so that is one bucket. The
second is in surveillance. We have an obligation to monitor our
markets. We have huge surveillance systems that take, ingest
billions of transactions a day, and so AI will help us get
those cases together and our investigators get a better case.
Mr. Steil. As that is coming online, what has your
interaction been like with regulators?
Mr. Zecca. The SEC has been very pragmatic and very
interested. I find often when technology is being adopted, not
only by us, but by the government, that there is just a
different knowledge level. I think AI, cloud before it, cyber,
of course, are areas where the government is very focused as
well. So that has helped the discussion because they do come
with a growing knowledge base, and they have been very
constructive in their questions.
Mr. Steil. So then let us dive in. You are talking about
the use cases for AI. In your testimony, you mentioned $3
trillion illicit funds flowed through the financial system last
year, we are all working to reduce that. AI could be a
significant tool. One example that is attracting a lot of
attention is synthetic identity fraud where criminals are using
AI to present themselves as if they were legitimate customers.
How is Nasdaq working to stay one step ahead of the bad actors?
Mr. Zacca. Yes. We are working with our customers. So we
are not a bank ourselves, but on the AFC side, we are working
with community banks, large banks. So those organizations in
the check fraud area, which is a huge and has been a growing
problem, in the money laundering area. They are coming with
problems that they feel need to be addressed. So we are using
those as the case studies as we try to develop our technology.
Mr. Steil. Are these criminals utilizing AI to try to
defraud or avoid Bank Secrecy Act (BSA)/AML filters?
Mr. Zecca. I think there is some of that. I mean, they are
using technology to try to, in the check fraud space, have more
realistic-looking checks that they are stealing in an AML to
move their money more transparently. If I could make one point,
I do think it is important that regulators recognize that there
are different use cases and the regulation that should apply to
them should be different. So if we are fighting financial
crime, we probably should have a different standard than using
the same technology in another space, and so I think that is
something that regulators need to consider.
Mr. Steil. Thank you very much. Thank you all for being
here.
I will now recognize Ms. Garcia for 5 minutes for the
purpose of asking questions.
Ms. Garcia. Thank you, Mr. Chairman. It is great to see
some familiar faces that we have visited either before here or
in the task force. So I first want to thank Ranking Member
Waters for her leadership on this very important issue in
addition, of course, both to the chair and the ranking member
for creation of the Working Group. It has been a pleasure to
work with my colleagues as we explored AI and how to ensure the
United States proceeds in a safe, competitive, fair, and
efficient manner as more and more people learn to work and live
with AI.
We all know how expensive buying models can be. For
example, our local institutions cannot afford to invest and
create their own AI models. Ms. Osborne, as of right now, are
smaller institutions like your credit union able to compete
with larger entities?
Ms. Osborne. Thank you for your question, Congresswoman
Garcia. So as a $1.6-billion institution, if we compare that in
size to all banks and all credit unions, we are considered
smaller.
Ms. Garcia. Smaller. So are you able to compete?
Ms. Osborne. We are able to compete, and that is due to
third-party relationships that we have available to us. We are
too small to develop AI in-house.
Ms. Garcia. Okay. So what can we do to better support our
community-based organizations as AI becomes more and more of a
commonplace?
Ms. Osborne. Thank you for that question. I would say our
existing regulation is technology agnostic, so that right there
is a great starting point. We have the guardrails in place to
enable financial institutions of all sizes to proceed with
entering into relationships like our current relationship with
interface.ai. One area that would be helpful is if our
regulators provided a statement or clarity in terms of where AI
fits in that regulation. My assumption, and in partnership with
vice chairman of the NCUA, I joined an AI panel with him last
March, and we had great conversations about the needed clarity
in regulation in which AI is considered technology. We consider
AI technology, thus regulation exists for that.
Ms. Garcia. What about any security concerns taking place
when your financial institutions or other small entities
explore the adoption of third-party service?
Ms. Osborne. So we have a very thorough third-party
management policy and program due to regulation. That applies
across the board, no matter which partner you are looking at,
so all technology partners, including AI. So one factor that we
spend quite a bit of time looking into is the security of that
particular partner. We look at any type of data breaches they
may have had, what their reputation in the market is. We review
their financials. We spend a great deal of time vetting that
partner before we make a decision. That long, lengthy process
is important to ensure we provide safety and security to our
members. It also is why sometimes it takes a bit longer to make
a decision on a partner, but it is the right reason to do so.
Ms. Garcia. Well, I was encouraged about your comments with
regard to addressing some language access issues. So is there
not enough language access, AI third-party providers out there
already that you all can easily adapt?
Ms. Osborne. So I can speak to our current relationship
with interface.ai. That solution provides the ability to speak
80 different languages to our members. We are just starting,
just scratching the surface by offering Spanish very soon.
After that, we plan to introduce Polish. AI is constantly
evolving and moving, and so I see the adoption of language
somewhat newer and be made available to us. So I expect that
will continue to grow over time.
Ms. Garcia. Since your English version is Olive, will the
Spanish version be Popeye?
Ms. Osborne. It will be Olivia.
Ms. Garcia. Olivia? Okay. Well, we will take Olivia. Okay.
Well, thank you so much. Mr. Karunamurthy, in your opinion,
could AI impact these jobs or similar jobs to those in my area,
which are like longshoremen, construction workers, and pipeline
workers? So many workers are just scared of losing their jobs.
How can we reassure them?
Mr. Karunamurthy. Thank you, Congresswoman Garcia. This is
a really important factor for us to consider as this technology
gets adopted across the U.S. economy. American workers are some
of the most capable, intelligent workers on the planet, and
they are immensely capable of leveraging this technology.
However, we have to make a broad adoption in upskilling workers
and helping them understand where AI can be trusted and where
AI needs additional oversight in order to be used successfully.
We consider that the important work that we do here at scale,
and hopefully, we can be more competitive as an economy as we
leverage this successfully.
Ms. Garcia. How can we reassure them not to be scared?
Mr. Karunamurthy. I think it starts by helping workers
understand where we are with this technology, what the
limitations of it are, and where the real opportunities are so
that folks feel that this is something that assists human
workers rather than something that is just going to replace
jobs.
Ms. Garcia. All right. Thank you. I yield back.
Mr. Fitzgerald [presiding]. The gentlelady yields back. The
gentleman from Kentucky, Mr. Barr, is recognized for 5 minutes.
Mr. Barr. Thank you, Mr. Chairman, and I want to first say
how much I appreciate the good work of the Working Group on AI
in this Congress, and it is important work because AI and
related technological progress in financial products hold a lot
of promise for innovation. A responsible approach to
developments in financial technology is to strike the right
balance between innovation and regulation to ensure that AI
generates net benefits and enhances the well-being of
Americans. We must recognize that government regulators already
have regulations, rules, guidance, and laws in place that put
restrictions on uses of technologies by financial institutions.
Congress and any future administration must be exact and
definitive in identifying AI use cases for which those tools
are thought to be somehow insufficient. Congress should not
defer solely to often opaque regulatory agencies to effectively
legislate through regulation on AI and related technological
innovations. We should not rely on simple fears and speculation
on potential adversities of innovation. As well, I believe that
as regulatory agencies themselves adopt AI or utilize their own
regulatory algorithmic black boxes, they must adhere to the
same principles and regulatory treatment that they require of
the private sector.
Mr. Zecca, let me ask you a question first. Do you agree
that Federal regulators of financial institutions, including
the Federal Reserve System (Fed), FDIC, and OCC, already have
regulations, rules, guidance, and laws that place restrictions
on use of technologies, including AI, generative AI, and the
like, by financial institutions and third-party vendors with
whom they contract?
Mr. Zecca. Well, thank you for the question. Yes, it is one
of the most heavily regulated industries in the world.
Mr. Barr. One of the pieces of feedback that I am getting,
especially from smaller community banks and community financial
institutions, is that when they rely on a third-party vendor
that provides AI services, they are concerned that regulators
would hold them accountable for the artificial intelligence and
the algorithms developed by these third-party vendors. Should
the law provide a safe harbor for especially small community
financial institutions and allocate the liability properly?
Mr. Zecca. Well, I think it is an important question. I
would look at it from the perspective, I think any organization
is going to want to know what is going on, so there is going to
be a natural interest for the organization to understand what
is happening. At what point does liability come in? I think
that is a point for discussion. We are a vendor to a lot of
community banks, and we work collaboratively with them to
ensure that we are meeting their needs and that they understand
the processes that we are using.
Mr. Barr. Well, let us direct that question to Ms. Osborne.
What is your view on the liability that could attach to a
credit union, for example, when using a third-party vendor?
Ms. Osborne. Yes. Thank you for your question, Congressman
Barr. From my perspective, an AI third-party provider is a
technology provider, and so the same rules apply for all. I do
not see a benefit of assigning liability for a specific AI
where if we think of the flip side, the arguments that we have
been making is that AI is a technology provider, so we should,
with technology agnostic regulation, treat them equally. From
our perspective, AI providers, we do treat more as a critical
vendor, so they are vetted more thoroughly. So if that vetting
and review is done properly, we should have awareness of
technology providers that may not be meeting the needs of the
industry.
Mr. Barr. Final question for any of our witnesses. Can
anyone comment on the activities at the CFPB? These days, the
Bureau has shown itself to be more inclined to move the market
by intimidation and supposition rather than by evidence and
administrative procedure. Industry players responsible for
developing the innovative AI solutions that can create
beneficial change for consumers need an informed regulator that
establishes and maintains reliable, consistent rules of the
road. Are you all observing overreach by the Bureau and fear of
speculative regulatory actions that would stifle innovation?
Anyone worried about CFPB stifling innovation in artificial
intelligence?
[No response.]
Mr. Barr. No, maybe? Okay. Crickets. Okay. Well, the reason
why I asked that question--my time has expired--we have heard
from some fintechs that this is actually going to hamper bank
fintech partnerships. With that, I yield back.
Mr. Fitzgerald. The gentleman yields back. We now recognize
the gentleman from North Carolina, Mr. Nickel, for 5 minutes.
Mr. Nickel. Thank you so much. Thanks to our witnesses for
being here with us today. I am very glad we are holding today's
hearing on artificial intelligence, or AI. It is important that
Congress, and particularly the Financial Services Committee,
does not fall behind as new technology advances. Also, thanks
to Chair McHenry and Ranking Member Waters for their work on
this in a bipartisan way, and of course, thanks to our current
chair, Fitzgerald, for traveling around the world on a
bipartisan Congressional Delegation (CODEL) where we made AI a
significant part of the work of our group as we traveled the
world.
First question, Mr. Zecca, to you. The financial services
industry seems to be ahead of many other sectors when it comes
to AI. Not only have market participants been using it for many
years, but we already have a strong legal framework to protect
consumers, prevent discrimination, and more. We need to
determine what tools we already have to address AI, assess
regulatory gaps, and continue to encourage innovation here in
the U.S. Do you agree with that, and what tools does the
Federal Government already have to address AI?
Mr. Zecca. I do agree with the premise. I think that the
Federal Government, I will talk about the agency that I work
with most, which is the SEC. Not only are they in the process
of adopting AI for their own operations, but they also have a
great familiarity with the markets, how technology changes have
come, and they have looked at changes in the past, so deepfakes
or other things that might impact market stability.
Over the years, the SEC has come out with circuit breakers,
limit up/limit down bans, which stopped trading around stocks.
So there are a lot of changes that have been made over the
years that would apply to AI just like any other issue. So I
think there is a lot of work that has already been done.
Mr. Nickel. Can you talk a little bit about regulatory
gaps? What do you see there and what should Congress be doing?
Mr. Zecca. Well, I do think that it is always important
with new technology to see where the gaps are. I think the
technology is still evolving in many areas. So where a gap may
develop may not exist yet, so I think in areas we want to look
at market stability, use of deep fakes, certainty about
individual identities are all things that I think are
important. Over time, gaps will develop, and I think that is
the beauty of the regulatory process. Aside from legislation,
there is the opportunity for specialty agencies that are
experts in the area to take action.
Mr. Nickel. Thank you so much. Ms. Rice, next question to
you. In their latest draft funding bills, House Republicans
have cut funding for fair housing enforcement and blocked HUD's
ability to enforce the affirmatively furthering fair housing
provision of the Fair Housing Act. Can you discuss the
importance of Federal fair housing enforcement to combat
automated discrimination in housing and lending through online
housing platforms and AI technologies?
Ms. Rice. Congressman Nickel, thank you for the question.
Yes, what we know from decades of research is that when we
remove discriminatory barriers from the marketplace, we
actually increase profitability and productivity. That enables
companies to expand their footprints, right, and have a broader
market reach. So that is one of the reasons why some of the
largest lending institutions in the United States of America
joined with us in asking the previous administration to not
place undue restrictions on the disparate impact rule that
would disable entities from being able to appropriately use
disparate impact to analyze their systems for fairness. It is
also one of the reasons why the Business Roundtable, of course,
which is comprised of the Nation's largest corporations in
America, adopted a platform calling for the adoption of a
strong, affirmatively furthering fair housing rule. When we
level the playing field for everyone, we increase productivity
and opportunity.
Mr. Nickel. Thank you so much. Mr. Reynolds, to you. AI-
generated audio and visual content, also known as deepfakes,
has armed criminals with a new tool, contributing to a
significant increase in the frequency and sophistication of
attacks against or through the financial services sector. In
addition to attacking financial institutions, deepfakes have
the potential to create widespread panic and confusion,
especially when combined with social media, leading to consumer
fraud, scams, and significant risk to the U.S. financial
stability. How is your organization and industry responding to
these emerging threats, especially combating malicious AI-
generated audio and visual content?
Mr. Reynolds. Yes. Thank you, Congressman, for the
question.
Mr. Nickel. I realize now we have 5 seconds.
Mr. Reynolds. Very briefly. We are looking at two things.
One is reinforcing existing controls that help stop that;
second, using generative AI to actually combat generative AI.
Mr. Nickel. Thanks. I yield back.
Mr. Fitzgerald. The gentleman yields back. We now go to the
gentleman from Georgia, Mr. Loudermilk, for 5 minutes.
Mr. Loudermilk. Well, thank you, Mr. Chairman. Thank you
all for being on this panel. It is a very important subject,
something I have been very interested in since I have been in
Congress. I have spent 30 years in the IT industry, and I think
back from when I first got involved, how quickly technology has
grown since then over the previous 30 years. In fact, if you go
back and you look at how the internet has grown, it was the
internet that drove most of the technology advances. If you
look at these devices here, they have grown so much over the
past few years, it is interesting to understand that the least
used part of this is the telephone, right, because that was the
most heavily regulated. The reason we saw such advancement in
information technology is because of our entire structure, our
economy here, technology has been the least regulated. It has
been free from a lot of excessive government regulation.
Now, a couple of Congresses ago, I was ranking member of
the first Artificial Intelligence Task Force, and what we were
talking about then, as far as AI, is different than what we are
talking about today with generative AI. I am thinking 3 to 5
years from now, we are going to be having an entirely different
conversation about AI. I agree with some of the testimony is,
if we can stop government from stamping out innovation and
working to allow, with constraints, with guardrails, to protect
individuals from AI. I understand a lot of people are scared by
what they do not understand, but we can see a lot of
advancements that are good for the industry there.
With that, Mr. Reynolds, in previous hearings about AI and
financial services, I had not had much of an opportunity to
discuss the technology's application in payments and financial
messaging. We talked about fraud detection and prevention, but
I would like to talk about how AI is used to streamline and
improve payments messaging. As you know, by early 2025, Fedwire
will have completed a transition into the ISO 20022 financial
messaging standard. The Society for Worldwide Interbank
Financial Telecommunication (SWIFT) network will have fully
phased out the SWIFT MT format, and transition to the SWIFT MX
format, which is also aligned with that same standard, by
November of next year. The ISO 20022 provides for much more
information-rich messaging than the older messaging standard.
Combined with artificial intelligence, this standard could
vastly improve efficiency in payments, messaging, and create
positive feedback loop and improving messaging algorithms.
All of that leads up to this question: what benefits could
a more information-rich financial messaging standard have in
terms of integration with artificial intelligence?
Mr. Reynolds. Thank you for the question, Congressman. I
think, as you noted, as we move to a more information-rich
architecture, it certainly allows for greater speed of
payments. It allows for greater accuracy of payments. What it
also allows is, on the regulatory side, it allows us to better
understand how particular companies, how particular individuals
are using payments, which should also increase our ability to
deter and detect fraud. It should increase our ability, for
instance, to protect vulnerable populations. We look
consistently at, for instance, elderly account holders and how
can generative AI systems perhaps look at behavior of their
payments, notice anomalies, and potentially intervene to
protect more vulnerable account holders.
So I think when you see those more information-rich
environments, there is certainly, I think, economic benefit for
the system, but I think there is also a benefit to the
consumers as well.
Mr. Loudermilk. All right. Thank you. Follow up question on
that. For the banks and businesses that FIS works with, what
benefits could the new ISO 20022 standard combined with AI
algorithms have in terms of payment messaging, if any?
Mr. Reynolds. It will certainly make them more secure. I
think, to start with that, safety and security is always one of
the primary concerns that we have at FIS, and to ensure that
for our clients. I think, additionally, again, as I noted
before, we work with very large institutions. We work with very
small institutions and everything in between. I think that
moving to this standard and moving to that technology platform
will allow us to really deploy technology that all firms of
varying sizes can use, which levels the playing field.
Mr. Loudermilk. Mr. Zecca, regarding trading, how is this
messaging standard apply in securities, and is Nasdaq using any
type of machine learning to process or analyze trading and
settlement messages?
Mr. Zecca. So we are. The surveillance piece is where we
are the most active with machine learning and moving into
generative AI, so trying to assess all of the data in the
market and determine patterns of bad behavior, insider trading,
and things like that. So over time, I can see we did have the
first order type recently approved by the SEC that has AI in
it, but I can see other use cases for it down the road.
Mr. Loudermilk. All right. Thank you. I yield back.
Mr. Fitzgerald. The gentleman yields back. We now recognize
the gentlewoman from Michigan, Ms. Tlaib.
Ms. Tlaib. Thank you, Mr. Chair. I know you have heard from
a lot of my colleagues. One of the areas, though, I think is
really critical to talk about is, while AI may offer a number
of it, it can also promote what we call market concentration or
manipulation and deepen existing power imbalances that we
experience in the 12th Congressional District in Michigan, and
also lead to discriminatory outcomes.
AI, as you all know, played a large role in boosting
investor ownership in single-family rentals, for example, has
been documented. These institutional investors are often highly
concentrated in specific markets, purchasing more than half of
single-family rentals in Atlanta recently, and controlling 75
percent of rental homes, I believe, right outside of Nashville,
Tennessee and because such investors use LLCs and special-
purpose vehicles, it is often impossible to know true market
share that they enjoy. We have also seen how RealPage is being
used by many landlords to set rental prices now, which is so
incredibly scary. For example, it has been reported that 90
percent of rental units in large buildings in Metro DC, right
here in the area, are priced using that software.
President Rice, what challenges does AI pose in terms of
facilitating market concentration or manipulation? Can you give
us a sense what legislative gaps exist and what we could be
working on, such as, I have heard folks talk about a National
Rental Housing Registry to get a clearer picture. I do not know
really what the answer is and would love to hear from you.
Ms. Rice. Thank you, Congresswoman Tlaib. Yes, you are
absolutely right. So AI is powered by Big Data, and so your
ability to effectively build AI models and use them is going to
be associated with your ability to sort of harness and amass
large amounts of data. You have to be able to get the data, you
have to be able to control it and use it, and we know that
larger institutions have a much better ability to acquire the
huge data sets that are necessary for building AI systems.
As you noted, we also know that large institutional
investors have been able to use AI to strategically purchase
single-family homes and take them off the market. We believe
that there has to be both a principles-based and an outcomes-
based system of legislation so that we can accurately govern AI
systems.
Ms. Tlaib. Just to kind of lead into the next kind of
issue, what my team and I have noticed that AI has regularly
tipped the scales toward landlords, in favor of landlords
versus tenants. I know a host of technologies have been
deployed by specific landlords to screen tenants, to collect
payments, to evict families, and more. These tools raise a
significant concern of the agro, whatever, the bias--I can
never say that word--digital surveillance, and data collection,
and security. Data collection should scare all of us. Today,
tenants might actually have to grapple with racially biased
facial recognition technology, which I have been pushing my
colleagues to abandon in our country, or be forced to use apps
to pay rent or communicate with their landlord.
President Rice, can you touch base on some of the issues
being raised on what we call now, I think in the nonprofit or
folks or people that work in housing, the landlord tech
movement?
Ms. Rice. Certainly. One of the primary concerns related to
proptech is the inability of consumers to get accurate
information about the decisions that those technologies are
making. So for example, if you are denied because of a tenant
screening selection system, you do not currently have the right
to know why that system denied you or even how it denied you or
perhaps what data was used to deny you. If a dynamic pricing
system is setting a rental price at a certain level, you do not
currently have the right as a consumer to know, to see the
transparency about how those rental prices were set.
Ms. Tlaib. I ran out of time.
Mr. Fitzgerald. The gentlewoman's time has expired. The
gentlewoman from Missouri, Congresswoman Wagner.
Mrs. Wagner. I thank you, Mr. Chairman. Mr. Zecca, I
appreciated Nasdaq's participation in the AI Working Group
roundtable earlier this year, and as a global technology
provider and operator of exchanges, could you please describe
the ways in which Nasdaq utilizes AI in its products and across
markets? Briefly.
Mr. Zecca. Sure. Well, I think you put it into the three
categories. One is potential order types, in other words,
adding liquidity to the market. The second is using both data
and analytics to allow corporates and investors to make better
decisions, so giving them higher quality data collated in
different ways. Finally, it is ensuring that the markets are
well surveilled. We also have solutions that work on compliance
reporting and regulatory reporting, and then perhaps the most
exciting is always we fight financial crime. We help eliminate
money laundering and other financial crimes.
Mrs. Wagner. Yes, that is key. What guardrails, sir, are in
place to ensure there is proper security, oversight, and
governance in AI?
Mr. Zecca. Yes. So the corporate governance piece is
important at each company. What we tried to do was start with
the governance process. So we basically blocked off all access
and we made a single path for approval, and that created a
single inventory of all the models. So we know that all of the
models are going through the same rigor. We assign a data
scientist to each project and then the cross-functional group
that approves all models includes lawyers, risk managers, and
they look with our ethics policies how we are implementing and
where the human interaction is and ensuring stability and all
the other accountability and all of the factors that you need.
Mrs. Wagner. I care deeply about retail and what I will
call mainstream, everyday investors saving for their retirement
and for the future. How does AI benefit the millions of retail
investors participating in our capital markets, and could you
please provide a few use cases?
Mr. Zecca. Sure. I do think there is a huge opportunity
here. Again, it is in its early days, but what we see are the
main opportunities with that data and the analytics that we
talked about. You can create products, unique products that
allow mainstream investors to diversify their portfolios at a
reasonable cost, types of tools and access to various
investment vehicles they have never had before. You can also
see a world where they can get financial advice provided
through AI with human intervention but in a way that they
probably could not have afforded if they had to get individual
advisors. I think those are two great examples.
Mrs. Wagner. Yes. I serve and I am proud to serve as the
chairman of the Capital Markets Subcommittee, and as you
highlighted in your written testimony, participants in our
capital markets have been using AI for decades. How has Nasdaq
approached this technology over the years and how has it
changed with the, I would say, the development of GenAI?
Mr. Zecca. Well, Congresswoman, I think you hit on the
important point. A lot of the guardrails that we developed in
the past, I think, apply with AI. So the markets over the years
have built up a lot of contingencies for aberrant trading
activity, for fast movements in the market. There are circuit
breakers and other things. We also monitor social media to try
to discover attempts to manipulate stocks. So all of those help
us now, but I would argue we have the potential to enhance our
monitoring with the AI as well, so AI will help us detect
fraud.
Mrs. Wagner. Would you describe how AI applicants are
enabling a fairer and more efficient financial system?
Mr. Zecca. I do think, as we said, the access is one. I
also think that the financial crime is another one. There is a
huge amount of financial crime that is an impact, it is a tax,
but it also supports a lot of the crimes that we are most
concerned about, human trafficking and terrorism. If AI can
allow investigators to zero in much more quickly on potential
crime, I think society benefits and that is a huge equalizer.
Mrs. Wagner. Mr. Zecca, last year the SEC approved Nasdaq's
filing to introduce the first exchange AI-powered order type.
What is this new order type and which market participants
benefit from this and what kind of engagement have you had with
the SEC?
Mr. Zecca. Sure. Well, the order type, basically, it is a
midpoint order type, and so it is people who are not as focused
on speed. They just want clarity of execution, so it kind of
freezes the market for them. Normally that was a static period,
and what we have done with AI is take market conditions into
account, so it is a variable time horizon and that has
increased and improved the fill rates. So that was something we
had to explain to the SEC when they came with very intelligent
questions.
Mrs. Wagner. Thank you. My time has expired. I yield back.
Mr. Fitzgerald. The gentlewoman yields back. We now
recognize the gentleman from Ohio, Mr. Davidson, for 5 minutes.
Mr. Davidson. I thank the chairman. It kind of out of order
how I add my questions, but to stay on the capital markets
theme there, Mr. Zecca. Nasdaq has a unique perspective on the
market, obviously straddling some of the biggest tech companies
in the world. One of the challenges that we see with the
standard trading desk might have an analyst, kind of a math
geek, and a programmer and they work together in tandem. You
could see how AI could do a lot of the work of analytics, could
do a lot of the work of the code, and might even be able to
accelerate or lower the bar for the skill needed in the math,
so it could change that function. Then you have core things
that seem to be a challenge in the market right now. I mean, in
theory, put call parity is a reality, but we all know there are
shorts that are oversold at times.
How do you see AI bringing reconciliation to markets like
that, problems that are not getting solved today. You can see
how it takes out labor and it is something that really works
pretty efficiently already. How are we going to use AI to solve
problems that are not getting adequately solved yet today?
Mr. Zecca. Well, I think the first thing to remember is
that the industry is heavily regulated. There are a lot of
organizations, including the SEC, Financial Industry Regulatory
Authority (FINRA), and us, watching the market. To the extent
our tools are enhanced by AI, our capacity to catch activity
that actually does violate the law increases, so I think that
is a net positive.
Mr. Davidson. Okay. Well, I hope that put call parity will
be a reality using blockchain or AI or something, aside from
just regulators that seem to miss a lot, but, hey, sports have
refs that miss calls every now and then, too. When we review
them, maybe we can make them right somehow. We will see how
that goes as we continue to talk about regulating our capital
markets.
Dr. Linda, real estate is front and center. As people are
dealing with inflation and the cost of living today, the cost
of homeownership or even residence for people that are renters
is one of the biggest challenges for households. If you look at
Zillow, you guys have really created a hobby for lots of
people. In some free time, they will go out and check. If I am
traveling, I will say, oh, what is the market price for a house
here versus where I live or whatever. So you collect a massive
amount of data, but on that consumer end we have talked a lot.
Several other people have asked questions on that.
I am curious if you have any insight into the rental market
because the Justice Department (DOJ) just did an investigation
into a software company in Atlanta that was analyzing rents.
They were basically helping the biggest private equity groups
and hedge funds and real estate investment trusts (REITs) buy
the software and rig the rental market to inflate prices and,
frankly, raise the barrier. What is out there to offer hope for
the consumer that we could find a way to lower the cost of home
ownership?
Mr. Linda. Thank you for the question, Congressman. I think
you are getting to the key of how Zillow is trying to benefit
the customer using AI, and that is with transparency. Similar
to how we have Zestimate, our key model that puts a price on
every roof in a wholesale market, we also do that in the rental
market. On every home, customers can see what that rent might
be, and that empowers them with information that they can use
potentially to have a more informed discussion with that
landlord and challenge them or ask them, why is the price you
are asking me that much higher? That is another application
where AI is used for the benefit of a customer by providing
more transparency.
Mr. Davidson. Yes, and this particular software company
would target the biggest, wealthiest funds and create a barrier
to entry so that they had a market there, and as we have all
experienced, somehow, Zillow makes money. It is all free out
there. Maybe we will save that for another scenario, but the
tool is available for consumers and it does give people some
availability.
Finally, Ms. Osborne, if you look at credit unions, Ohio is
certainly home to many of them and it works for our
communities. You look at the costs, you are generally looking
at third parties that you have to partner with in small
business. As a small business owner myself, we could not build
everything we wanted. You had to find trusted vendors. What is
the regulatory burden like for you guys? On the other hand, you
have regulators that are pushing you guys to collect massive
amounts of data and engage in operation Choke Point 2.0, 3.0,
whatever version of regulatory things there. What promise does
this offer for you to run a more efficient operation and serve
your community better?
Ms. Osborne. Thank you for your question, Congressman
Davidson. So from a financial institution perspective and as a
smaller credit union, there are significant benefits of using
AI. I found pricing from third parties to be comparable and as
expected for the services they offer, weighing that against
regulation, so our existing regulation is technology----
Mr. Davidson. Thank you. I am sorry, I did not leave you
with much time with the question, and please feel free to reach
out to my office.
Mr. Davidson. My time has expired, and I yield back.
Mr. Fitzgerald. The gentleman yields back. The gentleman
from Iowa, Mr. Nunn, is now recognized for 5 minutes.
Mr. Nunn. Well, thank you very much, Chairman Fitzgerald,
and thank you for the panel being here today. Together, we have
spent more than six sessions going over an engagement on the
future of artificial intelligence, its impact here, and the
technology that comes with it. Nowhere is this more clear than
in the financial sector and its opportunity to revolutionize
where we are going. We recognize already today the human to AI
ratio output is roughly one-to-one, but I think we all concur,
in less than 5 years, that output is going to look more like a
ratio of 1 to 1,000. It is a bold new space we are in, and with
this comes both opportunity and clearly challenges as we
discussed on this.
In the course of certainly my time here in Congress, we are
going to see AI advance to an over $4 trillion industry and has
the opportunity to revolutionize economies around the globe. At
the same point, it is enfranchising some of our greatest
adversaries, both in the Nation, State, and at the hostile
actor level. Generative AI, as we noted, is going to touch
every aspect of this and it is going to unleash not only the
economic productivity, but the empower communities like mine
back in Iowa. So I want to start here with a conversation on
how to implement this.
We have a number of regulatory challenges. We have
innovation being sparked, and as we learned from you, is that
there are a lack of lanes in the road in where operators are
going to be. In short, I believe we need a plan, which is why
working with this team, we have introduced H.R. 7881, the AI
Plan Act. 7781 discusses the opportunity for private sector to
help work with key leads in the government sector--Treasury,
Commerce, Homeland Security--to both help inform this, but
shape something that is meaningful.
So I want to get started with you, Mr. Reynolds. Do you
believe artificial intelligence requires an entirely new
legislative system and a regulatory regime that comes with
that, or can we start to apply some of these existing laws and
get us on the same sheet of music or on the same level of code,
if you will?
Mr. Reynolds. I think certainly in financial services, the
existing regulatory regime is sufficient to deal with AI at
this point. I think if we were looking at something more broad
for AI, I think a principles-based sense that really is sort of
technology agnostic. I think the challenge is with AI and
generative AI, the different use cases will demand very
different regulatory solutions, and so I think a one-size-fits-
all will be very difficult for AI but specifically in financial
services where I obviously specialize, I think the existing
regulatory regime right now is sufficient.
Mr. Nunn. I would agree with you. In fact, I think by the
time we write laws, the level of AI that we have is probably
going to advance beyond that. The principles-based seems like a
much more prudent approach. I appreciate the innovation coming
out of the private sector on this.
Ms. Osborne, you shared with us, from your perspective,
both at the credit union level, something that is hugely
important for my main streets, my hometowns. The challenge here
that we saw from the national security side, we heard from the
CEO of Pindrop that there has been as much as a 450-percent
increase in deepfake voice attacks using artificial
intelligence. In fact, in as little as 2 seconds, a deepfake
voice can be changed, fundamentally changing how you do
consumer protection. Who do you see as your ally on the
government level to be able to address when a small credit
union like yours gets attacked? Who do you pick up the phone
and call?
Ms. Osborne. Thank you for your question, Congressman Nunn.
So we are regulated by the National Credit Union Administration
(NCUA), so in the event we had a security event or risk, we
would reach out to both the Federal Bureau of Investigation
(FBI) as well as the NCUA.
Mr. Nunn. So not unlike our cyberattacks here, has that
been flushed out because the challenge is when you get
attacked, it is the same type of AI generative attack that can
happen instantly to be able to hit a ton of credit unions,
banks, lenders, all at the same time. Who in the Federal
Government do you think is on point to deal with this?
Ms. Osborne. I am not sure of the answer to that, so I
would have to get back to you on it.
Mr. Nunn. You are not alone.
Ms. Osborne. Okay.
Mr. Nunn. I am not sure, after our conversation, Congress
is fully aware, and I would offer the Federal Government. All
the more reason I think we need to have a collective effort
when we are still at this one-to-one ratio versus when it is 1
to 1,000.
With my few seconds remaining here, Mr. Karunamurthy, you
highlight here some of the future threats. What do you see that
we should be dealing with today in Congress to help better the
AI space, and I want to talk specifically about what threats
should we be looking at now.
Mr. Karunamurthy. Absolutely. These threats are real. We
know that State actors like Iran and North Korea are using
generative AI technologies to help spread misinformation, to
probe our cybersecurity defenses. It is important to leverage
techniques such as red teaming, which can help understand where
AI systems are vulnerable and find ways to build more resilient
systems. We have a great partnership with the Chief Digital and
Artificial Intelligence Office (CDAO) within the Defense
Department, and we hope to build more of those collaborations
to spread more resilient systems in the future.
Mr. Nunn. I could not agree with you more. I think there
are good best practices out there. We need to incorporate more
of them from recommendations on the private sector panels like
this into what we are doing in the Federal Government. I
appreciate you being here. I look forward to the future. Thank
you, Mr. Chair. I yield back the remainder of my time.
Mr. Fitzgerald. The gentleman yields back. The gentleman
from South Carolina, Mr. Timmons, is recognized for 5 minutes.
Mr. Timmons. Thank you, Mr. Chairman. I want to thank all
the witnesses for being here today.
Before the late 20th century, evaluating creditworthiness
in the homebuying process was poorly standardized. Essentially,
a homebuyer's ability to repay a line of credit was based on
prior relationships to lenders and was highly subject to class,
race, and gender biases. As a result, homebuying was largely
unattainable to many Americans. However, in 1989, Fair Isaac
Corporation (FICO) invented the first standardized consumer
credit score. Upon its adoption, determining one's
creditworthiness became easier for consumers and lenders to
determine, and a standard was created that limited bias in the
homebuying process. The results speak for themselves. As our
technology has progressed, so have the factors that go into
determining creditworthiness. This has led to the development
of standardized credit scores, which provide an even more
objective snapshot of an individual's risk profile. By
innovating how we viewed creditworthiness, we allowed the
American Dream to be accessed by a never-before-seen number of
Americans. Despite the complex quantitative analysis being
conducted throughout the lending process, the financial
services industry continually has to fight back against claims
of systemic racism in preferential lending.
AI technology is no stranger to the financial services
industry, as machine learning has been leveraged for a variety
of use cases over the course of the last decade plus. The
technology's continued development means it can potentially
serve as a powerful tool to help financial institutions
mitigate claims of discrimination. By implementing unbiased
algorithms and rigorous oversight, AI can enhance objectiveness
and transparency in the decisionmaking process. It has become
clear that the technology's ability to process large tranches
of data will make it yet another tool to determine risk and
open up access to credit for many more Americans by removing
human biases from lending decisions.
As my colleagues across the aisle are seeking to stifle the
technology's implementation amid new claims of bias and
discrimination being contained within AI datasets, I believe
that with appropriate oversight and intentional implementation,
that is something that can be easily avoided. With
transparency, we can all come to agree that is not a problem.
Dr. Linda, would you agree that with proper data inputs, AI
has the potential to expand access to lending opportunities and
enable more Americans to access the credit they need to buy a
home?
Mr. Linda. Absolutely, I believe in that, and I can give
you one example. At Zillow, if you are renting homes through
the Zillow platform and the landlord works through Zillow for
the payment, you can opt in to have your rental payments
submitted to the credit reporting agencies, which actually we
have seen, and as was pointed out before, can expand the access
to credit. So we truly believe that AI has the potential as we
look forward with the existing regulations.
Mr. Timmons. Essentially using additional data inputs to
make it more fair and more transparent?
Mr. Linda. Yes.
Mr. Timmons. Thank you. Ms. Osborne, what role does AI play
in the lending process, and how has the industry begun to adopt
AI technology, and has it expanded access to credit for more
individuals?
Ms. Osborne. Thank you for your question, Congressman
Timmons. So currently at GLCU, we do not use AI for lending.
However, it is an area we are exploring because of the benefits
it introduces to be able to lend to individuals that we are
unable to determine creditworthiness for today. There was a
testimony 2 weeks ago by the CEO of Commonwealth Credit Union,
Karen Harbin. She spoke of a relationship with Zest AI. Zest AI
enables just that. So she is able to use that platform to lend
to disadvantaged individuals by going beyond the credit score
and using transaction data like frequent bill pays, you have
paid your utility company 12 months in a row, things like that.
That is a great benefit to AI that really broadens the
capabilities of financial institutions to lend to those in
need.
Mr. Timmons. You would do that by accessing their checking
account and seeing the monthly payments and that they are made
on time? Is that how you do that?
Ms. Osborne. Absolutely, yes. We have access to a wealth of
data that we should be using for those type of decisions.
Mr. Timmons. Thank you. Dr. Linda, you kind of touched on
this, but how else has Zillow utilized AI to expand home
ownership opportunities?
Mr. Linda. We do that by informing the customers of the
process. We are exploring how they can find what it takes to
buy a home by empowering them with information of, for
instance, what a home is worth, by making it easier through AI,
through things like natural language search to find homes on
the market and understand what home they might be purchasing
and by empowering the real estate operators to spend more time
with a customer, touring homes in person, walking them through
the process, and really getting them to the home they
eventually want to move into.
Mr. Timmons. Thank you for that. AI is here to stay, and we
need to use it to the best of our ability while mitigating any
adverse impacts it can have. Appreciate you all being here, and
with that, Mr. Chairman, I yield back. Thanks.
Mr. Fitzgerald. The gentleman yields back. I now recognize
the gentleman from Tennessee, Mr. Ogles, for 5 minutes.
Mr. Ogles. Thank you, Mr. Chairman, and thank you to the
witnesses for being here today.
AI is certainly on everyone's mind these days. It seems
like it comes up in almost every conversation that I have with
constituents or stakeholders. Since my district includes part
of Nashville Music City, I have learned a lot about the music
industry's hopes for and concerns about the technology. On one
hand, as you might suspect, they have concerns about how
intellectual property will be protected as generative AI comes
into wider use. However, many in the industry are also quick to
point out that AI-based tools are incredibly useful in things
like music production and strongly caution against taking
reckless actions that could hinder the use or development.
I bring up that example to highlight a principle that also
applies in the financial services sector. While everyone,
including companies in the industry, lawmakers, and
policymakers, consumers alike need to be attentive to ensure
that AI tools are used in a way that upholds the rights of all
parties, protects rather than diminishes their security, and
ensures fairness, we also have to ensure that we are not
stifling beneficial innovation. Let us be honest. When it comes
to AI, that is a little bit of a challenge, right? Where is the
gives and takes there? There is a lot of innovation using AI,
so we do not know all of the ways it could be useful or
empowering, driving down costs, improving access to financial
products, or empowering the creation or expansion of
businesses, but with that, there is also some downsides that we
may not quite realize yet.
Ms. Osborne, in your written testimony, you described the
NCUA's risk management rules as robust. You note that their
previous requests for information have indicated that they
understand the credit unions face considerable and wide-ranging
competitive pressures where AI can help. In your experience, at
least so far, is the NCUA's regulatory approach to credit union
use of AI striking a good balance between ensuring compliance
with the law and welcoming beneficial innovation where there is
room for improvement, ma'am?
Ms. Osborne. Thank you for your question, Congressman, and
I do. So the regulations that exist today provide the
appropriate guardrails to make decisions on use of AI. AI is a
technology, and our regulation today, as it stands, is
technology agnostic, so it currently applies. AI fits nicely
into that. Where I do think there is an opportunity for further
clarity is a statement or some type of comment from our
regulator, the NCUA, that AI is a technology, and then,
therefore, it would fall within our existing guardrails.
Mr. Ogles. Well, you literally jumped to my next question.
So you note that the adoption rate of AI programs by credit
unions is still relatively low. Do you believe that it would be
helpful to increase adoption if the government would give you
some sort of statement or some clarity on the issue and how
they can and cannot be utilized because, again, when you look
at the regulatory State, especially in the financial services
sector, there is always that hedge of getting too far out in
front of the regulators because if you do, then they snap you
back, and they penalize you in the process.
Ms. Osborne. Yes, that is an excellent point, and it is a
concern that has been shared with me from other credit unions
as well. It creates hesitation to engage in a partnership with
an AI solution where we all know the benefits of AI technology
are so ginormous where we still cannot measure it. So by
providing that clarity, it would help credit unions and other
financial institutions to make better decisions on that.
Mr. Ogles. This is somewhat subjective, of course, but when
you look at AI, and if it can truly be implemented in such a
way to, from your perspective, further vet someone who is
coming in for credit because you are having lower risk or you
have lowered your risk, that is where you have lower defaults,
there is a savings there, there is a profit factor, but a
savings there for the consumer, right? That also increases
access at a time where you have greater inflation, where rents
have increased, and, quite frankly, for a lot of individuals,
the pursuit of the American Dream and that picket fence is
slipping away. You now have the opportunity to get some of
those people back into the equation.
So I do think it is imperative, Mr. Chairman, that as we
look at AI, as we look at, especially into the financial
services sector, that we encourage, if not force, the CFPB, the
other NCUAs, et cetera, to provide clarity so that our
financial institutions can be bold leaders in this space, both
from the perspective of protecting consumer data and the
consumer themselves, but also creating better and greater
opportunity for those that otherwise may not qualify.
With that, I had other questions for most of you, and I
apologize for not getting to them. I really appreciate all of
you being here and sitting there and enlightening us on this
issue. With that, Mr. Chairman, I yield back.
Mr. Fitzgerald. The gentleman yields back. I am now going
to recognize myself for 5 minutes.
I just had a couple of questions. I am a member of
Financial Services but also a member of Judiciary. I happen to
be the only member on both committees right now, so I did have
a couple of questions. I thought maybe we could just go right
down the line. Mr. Karunamurthy, let me direct it to you first.
What safeguards do we have in place to avoid copywritten
material, and then the second part would be on anything you
would want to hold a patent on. Do you have any comments on
that? We can go right down the row and everybody can kind of
give us a short answer on that.
Mr. Karunamurthy. Thank you. It is important to understand
the role that intellectual property (IP) rights and copyright
plays in the development of this technology. We take this
approach of testing and evaluating how models perform in really
interesting scenarios to understand when the models might have
been trained on material that might be problematic and
situations where that model might help with that material. That
is an important first step in understanding what the impact of
violations of IP rights may play in development of this
technology. We hope that will inform future discussions over
how to understand the data that is used to train the models and
what data can be used in the future.
Mr. Linda. Thank you for the question. We certainly comply
with all the existing regulation and our contractual obligation
to our partners, who do publish data on a platform like Zillow.
To your question about patenting, we lead with transparency. We
are trying to share a lot of our methodology, including open
sourcing, some of it. On occasions, we do have a regular
process, and we explore what from our technology we want to
patent. We use our legal patent expertise to guide us through
the process.
Ms. Osborne. At our current asset size, we do not develop
in-house AI. We use third-party, so I cannot really comment.
Mr. Fitzgerald. Thank you.
Mr. Reynolds. Thank you, Congressman. As a technology
company, obviously our product is intellectual property, so
this is a particular concern for us with GenAI. One of the
issues right now that exists in the industry where I think
there could be clarity is the ability to either copyright or
patent products that involve GenAI or have used GenAI in their
creation. So that is obviously something we are very focused on
and looking for increased clarity in the legal regime around
that.
Mr. Fitzgerald. Very good.
Ms. Rice. I would agree with the comments of Mr. Reynolds.
We find the process of trying to obtain a patent extremely
difficult and very costly. We are a nonprofit organization, so
our ability to hire very expensive attorneys is very limited.
There are not a lot of attorneys that have expertise in this
area and they charge very, very high rates, so anything that
can help democratize that patent process would be beneficial to
nonprofit groups.
Mr. Zecca. Finally, I would just note we are spending a lot
of time right now on data ownership and control. So we are
rewriting a lot of contracts to ensure that we have mutually
agreed ownership rights and understand what our rights going
forward are, and just to think more broadly on intellectual
property, including patent ownership as AI becomes more
involved in creation of patents, you know, what are the patent
rights I think will be important.
Mr. Fitzgerald. Very good. I will yield, and we will now
recognize the gentlewoman from California, Mrs. Kim, for 5
minutes.
Mrs. Kim. Thank you very much, and I would thank all the
witnesses for joining us today. It really was an honor for me
to personally serve on the Working Committee on AI, and through
many discussions that we have had through the roundtable
discussions, I learned a lot. I am glad that now we are having
a full committee hearing at this setting to display the AI
capabilities and the new opportunities that the technology may
bring into the financial services sector.
As you know, advancements in AI have created a lot of
potential, but there are also a lot of fears about how AI will
change how we interact with one another and conduct our work.
History tells us that skepticism of new technologies is often
unfounded and misplaced, as was the case with the railroad 200
years ago and most recently personal computers. So, Mr.
Reynolds, let me ask you a question first. One of the issues
raised in our Working Group discussion was the issue of AI
explainability. In your testimony, you highlighted how
explainability is part of your core principles as you integrate
AI and GenAI. So can you touch on how you are meeting those
core principles on explainability and addressing the black box
issue?
Mr. Reynolds. Yes. Thank you, Congresswoman. I think,
fundamentally, we start from the premise that I think you
touched on, that for GenAI to be successful, it really has to
be trusted. One of the ways that we think it is imperative for
it to be trusted is explainability. If people see it as a black
box, if they see it as something where they do not understand
why the result is coming out of it, then they are never going
to fully trust it. That is not to suggest GenAI will not make
mistakes. Human beings do all the time. GenAI will also, and it
will hopefully learn from those mistakes as you educate the
models.
Fundamentally, what we tried to do is we really start from
the first principles of adhering to explainability as well as
the other principle-based areas that we talked about. What we
do is we actually have rigorous regulatory guidance within the
company where both in terms of intake and then in ongoing
supervision of the models, we look at them very carefully to
ensure that they constantly adhere to those principles.
Mrs. Kim. That is how you are enabling AI models to be
audited, right?
Mr. Reynolds. Yes, ma'am.
Mrs. Kim. Yes. Okay. Thanks. How is AI being deployed in
the payments industry to detect fraudulent payments or any
other nefarious activities?
Mr. Reynolds. I think if you look at AI more generally, it
has been deployed across payments and in the industry for quite
a long time. It originally started, I think, as a very basic AI
that was more of a rules-based system and has grown over time
to machine learning. I think what we are starting to see now,
though--and we have started to prototype some of this and
started to test it--we have seen very significant results of
much lower false positives, but also, importantly for
consumers, much more payments that are allowed because they are
not flagged inappropriately as potentially fraudulent. So we
think actually what we will see is consumers will have much
less false positives, which makes the system safer and faster,
but we will also see better payment flows.
Mrs. Kim. Thank you. Let me ask a question to Mr. Zecca. I
want to continue on that track on how AI can be leveraged to
fight criminal activity. So can you talk about how hard it
would be to identify millions of data points that identify
criminal activity without the use of AI?
Mr. Zecca. Well, thank you for the question. I think it is
critical. We are on the frontier of a new opportunity to
connect the dots. When you look at any financial crime, you are
looking for a needle in a haystack. The people that we are up
against are trying to obfuscate, they are trying to hide their
activity, and sometimes it is a question of different bank
accounts, different leads, different countries. AI can perform
a lot of that work so that the investigators get a much more
developed case, and they get fewer false positives, as Mr.
Reynolds was saying. I think that is a potential game changer,
particularly as crime gets more sophisticated.
Mrs. Kim. Considering you are leveraging AI to improve the
effectiveness of financial risk management programs, can you
speak about the accuracy of AI in identifying illicit activity,
and are there any real case studies you can share with the
committee?
Mr. Zecca. Well, let me come back with that because I think
it is an interesting question on the case studies, and I do not
have them handy.
What we have tried to do, first of all, we do validate it
through our own work and, of course, our customers because we
are selling these to banks and institutions that are also doing
their own validation. We are seeing about a 30 percent increase
in efficiency, and that means a reduction of false positives.
Now, that is not the same thing as saying everything is going
to develop into a case, but we are definitely getting people on
the road to getting to those cases faster.
Mrs. Kim. Thank you very much. My time is up. Appreciate
it.
Mr. Lawler [presiding]. The gentlelady's time has expired.
The gentleman from Nebraska, Mr. Flood, is now recognized for 5
minutes.
Mr. Flood. Thank you, Mr. Chairman. I had the great
privilege of serving on the AI Working Group, which I really
enjoyed. Before I begin my questions, I would like to quickly
recall some of the main takeaways I had from those roundtables.
I think it is worth starting with the simple explanation of
terms. Artificial intelligence covers a lot of ground,
including some technology that is not really all that new. I
personally am a little less interested in the use of things
like algorithms, which have been around for a long time, than
the use of technology that has shocked the world since the
debut of ChatGPT, for example. Well, we identified a number of
interesting use cases with the financial service industry for
artificial intelligence. We found less use cases for generative
AI, specifically. A lot of the use cases for generative AI that
we did hear about, for example, it is used in chatbots, do not
necessarily raise the types of red flags that I worry about as
much from a regulatory standpoint.
I know that some of my colleagues are very concerned about
the potential for things like AI discriminating against people
based on race or ethnicity. That is a hugely important issue.
There is no doubt that if a company discriminates against
protected classes in a violation of law, they need to be held
liable regardless of whether it is an algorithm or a human that
did the discriminating.
To the extent that there are gaps, I think it is
appropriate to have a conversation about legislation to fill
them. That being said, I did not necessarily see lots of
evidence of that in our meetings, and when you think about it,
that may be partially because companies have used AI in things
like underwriting for many years at this point. While the term
AI sounds new, some of what we are talking about today is not
new or groundbreaking.
That leads me to one of my main takeaways from these
panels. I think that some of the greatest threats from AI in
financial services are not necessarily going to come from how
regulated entities use or misuse the technology, but how
nefarious actors could use generative AI. On that note, I want
to mention that I worked on a bill with my colleague,
Representative Pettersen, that is noticed for this hearing. It
is called the Stop Deep Fake Scams Act. It is a good bill and a
bipartisan opportunity to highlight some of the problems that
we are having in the face and age of generative AI.
Mr. Linda, I want to briefly pivot to generative AI use in
the housing market. You indicated in your testimony that Zillow
is interested in using generative AI tools in real estate
transactions through your platform. Can you expound on that?
What specific ways would you use generative AI in real estate
transactions?
Mr. Linda. Thank you for the question, Congressman.
Focusing on the customer, we do see generative AI being a great
tool in terms of informing them of the process. We did ask many
of our customers would you use generative AI and how would you
feel about it, and they share with us that often, especially
first-time buyers, feel that it is difficult to ask a landlord
a question, you know, what does it take. They feel like they
might be judged and that they cannot afford a home. They
actually express that having something like a chatbot, having
an experience that is always available, patient, can answer
them with the factual information would be actually useful to
get them the information they need.
On the professional side, we see big use for generative AI
to save loan officers and agents time from mundane repetitive
tasks, things like AI-powered call summarization. If you get on
the phone with a customer, you have your notes, so you know
what the follow up steps are, you can move to the next
customer, so you can spend more time in person with them. These
are just some of the use cases we are thinking about that we
feel generative AI can be very good at.
Mr. Flood. Mr. Karunamurthy--now I practiced this for the
better part of 5 minutes in my head and I just did a poor job
there--Karunamurthy. In your testimony, you identified some of
the work your firm does for financial services clients.
Nominally you describe that is applying AI to ``harness the
power of their proprietary data.'' More specifically, what kind
of work are you referring to there? Is this an effort to
improve these clients' customer chatbots or other applications
of generative AI that we should note?
Mr. Karunamurthy. Thank you, Congressman. A lot of the
initial use cases we have seen excitement about are areas where
you need to synthesize and understand the huge range of data
and see how that data is evolving over time, for example, for
reviewing legal or compliance documents and identifying redline
changes or other changes and the impact those can have, and
then helping human operators understand what is happening in
that data.
We also see a range of use cases where these ALMs can
generate code and that code can go and discover new insights
out of the data automatically, and so assisting data analysts
in doing their job more efficiently. We have always seen that
as an assistive technology that helps data analysts, wealth
advisors, other job functions do their job better rather than
replacing jobs at this point.
Mr. Flood. Thank you, Mr. Karunamurthy. With that, I yield
back.
Mr. Lawler. The gentleman yields back. I now recognize
myself for 5 minutes.
Mr. Linda, purchasing a home is probably the most
significant investment the vast majority of Americans will ever
make. Realtors play a vital role overseeing and facilitating
this process, especially for first-time homebuyers. Do you
believe that humans will continue to retain an important role
within the entire transaction, or do you see AI playing a
larger role as we move forward?
Mr. Linda. Thank you for the important question. I shared
in my opening remarks that from our research, we surveyed
buyers, and they shared with us that over half of them actually
cried during the process. That is how emotional and distraught
and difficult the process is, and that is why we see the human
as the centerpiece of the whole entire transaction. We are
actually trying to use AI to enable the humans to spend more
time together, enable the agent to spend more time touring
homes in-person with a customer as opposed to, again, doing
mundane repetitive tasks. So we truly believe that the human
aspect is a key to real estate transaction, and we see AI as an
enabler of humans spending more time together as opposed to
less or replacing anyone.
Mr. Lawler. Now, we understand that AI can process vast
amounts of data, including property records, sales history,
economic indicators, and market trends within the real estate
market. Using AI to process and analyze this information can
help identify patterns and correlations. Do you believe that
this means that AI will make buying a home a reality for more
people within the market?
Mr. Linda. Absolutely, by increasing transparency. As an
example, the Zestimate empowers more people with information,
by empowering people to find homes on the market more easily
through things like natural language. These are some of the
strategies that Zillow has been deploying to really enable more
people to find a home, make more informed decision, and, as I
mentioned already, spend more time with the agent in person so
they can get the advantage in the market and actually end up
with purchasing a home.
Mr. Lawler. So to that end, how is Zillow ensuring that AI-
powered tools are developed and deployed responsibly to make
searching for renting, buying, or financing a home more
equitable and transparent?
Mr. Linda. Thank you for your question. We have several
levels of things that we do. I will start by really putting
fair housing and the existing regulation at the forefront of
what we do, and kind of infusing it with what is needed to
deploy AI responsibly. We started doing that by putting
publicly our principles and then following them in what we do.
We have a cross-functional working group at Zillow that
proactively intakes algorithms early on as the ideas come up.
We look at them from the lens of fair housing and responsible
AI. We look for inputs that we should not be using. We think
about the impact on the customer, define fairness metrics. We
go through the audit. Every time a major new update to
algorithm happens, we do it again because every time algorithm
changes, algorithm that was fair yesterday might be not be fair
tomorrow. So this is a repetitive process, and we also come to,
you know, places like this to share our experience, and we
continue to work as we go forward.
Mr. Lawler. Mr. Zecca, according to your testimony, an
estimated $3.1 trillion in illicit funds flowed through the
global financial system in 2023, and last year alone, losses
from fraud totaled $485.6 billion globally. How has Nasdaq used
machine learning to address illicit activity?
Mr. Zecca. So our Verafin product provides AFC surveillance
software for more than 2,000 institutions, so all the way from
credit unions to tier 1 banks. In that context, what they are
trying to do is put together the various pieces to develop
cases, and some of the data will be within their own systems.
Some, frankly, will be on the internet when you are trying to
figure out whether they have a connection with another
individual.
AI is a game changer in trying to connect those dots and
having the compliance departments, who are stretched with a lot
of different responsibilities, have them focus on the most
likely cases and avoid the false positives. So we are seeing a
significant increase in efficiency, and I think over time, that
will even increase.
Mr. Lawler. As is often the case when new technologies
emerge, criminals are also using them----
Mr. Zecca. True.
Mr. Lawler [continuing]. to develop new and more
sophisticated financial crime strategies. So from that end of
it, what is Nasdaq doing to out innovate the bad actors, if you
will?
Mr. Zecca. Well, we are working with our clients on their
use cases. I do think your question raises two important
points. One is I think most institutions would say we would
like a stronger feedback loop with the government. When we do
our suspicious activity reports, which ones are the most
useful? While there is feedback, I think that is an area where
people would want more. The second is clarity on the
expectations of the regulatory requirements. You know, if you
are fighting crime, the level of explainability that you may be
able to accomplish maybe that is less important than trying to
stay up with the criminal. So I think that is another piece
that regulators will need to weigh in on.
Mr. Lawler. Thank you. I would like to thank our witnesses
for their testimony today and for your patience this morning.
Without objection, all members will have 5 legislative days
within which to submit additional written questions for the
witnesses to the chair. I will forward those questions to the
witnesses for their response. I ask our witnesses to please
respond as promptly as you can.
[The information referred to can be found in the appendix.]
Mr. Lawler. With that, this hearing is adjourned.
[Whereupon, at 2:30 p.m., the committee was adjourned.]
A P P E N D I X
July 23, 2024
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
[all]