[House Hearing, 116 Congress]
[From the U.S. Government Publishing Office]





 
                   ROBOTS ON WALL STREET: THE IMPACT

                   OF AI ON CAPITAL MARKETS AND JOBS

                   IN THE FINANCIAL SERVICES INDUSTRY

=======================================================================

                                HEARING

                               BEFORE THE

                 TASK FORCE ON ARTIFICIAL INTELLIGENCE

                                 OF THE

                    COMMITTEE ON FINANCIAL SERVICES

                     U.S. HOUSE OF REPRESENTATIVES

                     ONE HUNDRED SIXTEENTH CONGRESS

                             FIRST SESSION

                               __________

                            DECEMBER 6, 2019

                               __________

       Printed for the use of the Committee on Financial Services

                           Serial No. 116-73
                           
                           
                           
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]          




                            ______                      


             U.S. GOVERNMENT PUBLISHING OFFICE 
42-632 PDF             WASHINGTON : 2020 

                           
                           

                 HOUSE COMMITTEE ON FINANCIAL SERVICES

                 MAXINE WATERS, California, Chairwoman

CAROLYN B. MALONEY, New York         PATRICK McHENRY, North Carolina, 
NYDIA M. VELAZQUEZ, New York             Ranking Member
BRAD SHERMAN, California             ANN WAGNER, Missouri
GREGORY W. MEEKS, New York           PETER T. KING, New York
WM. LACY CLAY, Missouri              FRANK D. LUCAS, Oklahoma
DAVID SCOTT, Georgia                 BILL POSEY, Florida
AL GREEN, Texas                      BLAINE LUETKEMEYER, Missouri
EMANUEL CLEAVER, Missouri            BILL HUIZENGA, Michigan
ED PERLMUTTER, Colorado              SEAN P. DUFFY, Wisconsin
JIM A. HIMES, Connecticut            STEVE STIVERS, Ohio
BILL FOSTER, Illinois                ANDY BARR, Kentucky
JOYCE BEATTY, Ohio                   SCOTT TIPTON, Colorado
DENNY HECK, Washington               ROGER WILLIAMS, Texas
JUAN VARGAS, California              FRENCH HILL, Arkansas
JOSH GOTTHEIMER, New Jersey          TOM EMMER, Minnesota
VICENTE GONZALEZ, Texas              LEE M. ZELDIN, New York
AL LAWSON, Florida                   BARRY LOUDERMILK, Georgia
MICHAEL SAN NICOLAS, Guam            ALEXANDER X. MOONEY, West Virginia
RASHIDA TLAIB, Michigan              WARREN DAVIDSON, Ohio
KATIE PORTER, California             TED BUDD, North Carolina
CINDY AXNE, Iowa                     DAVID KUSTOFF, Tennessee
SEAN CASTEN, Illinois                TREY HOLLINGSWORTH, Indiana
AYANNA PRESSLEY, Massachusetts       ANTHONY GONZALEZ, Ohio
BEN McADAMS, Utah                    JOHN ROSE, Tennessee
ALEXANDRIA OCASIO-CORTEZ, New York   BRYAN STEIL, Wisconsin
JENNIFER WEXTON, Virginia            LANCE GOODEN, Texas
STEPHEN F. LYNCH, Massachusetts      DENVER RIGGLEMAN, Virginia
TULSI GABBARD, Hawaii                WILLIAM TIMMONS, South Carolina
ALMA ADAMS, North Carolina
MADELEINE DEAN, Pennsylvania
JESUS ``CHUY'' GARCIA, Illinois
SYLVIA GARCIA, Texas
DEAN PHILLIPS, Minnesota

                   Charla Ouertatani, Staff Director
                 TASK FORCE ON ARTIFICIAL INTELLIGENCE

                    BILL FOSTER, Illinois, Chairman

EMANUEL CLEAVER, Missouri            BARRY LOUDERMILK, Georgia, Ranking 
KATIE PORTER, California                 Member
SEAN CASTEN, Illinois                TED BUDD, North Carolina
ALMA ADAMS, North Carolina           TREY HOLLINGSWORTH, Indiana
SYLVIA GARCIA, Texas                 ANTHONY GONZALEZ, Ohio
DEAN PHILLIPS, Minnesota             DENVER RIGGLEMAN, Virginia

                            C O N T E N T S

                              ----------                              
                                                                   Page
Hearing held on:
    December 6, 2019.............................................     1
Appendix:
    December 6, 2019.............................................    33

                               WITNESSES
                        Friday, December 6, 2019

Fender, Rebecca, CFA, Senior Director, Future of Finance 
  Initiative, Chartered Financial Analyst (CFA) Institute........     8
Lopez de Prado, Marcos, Professor of Practice, Engineering 
  School, Cornell University, and Chief Investment Officer, True 
  Positive Technologies..........................................     6
McIlwain, Charlton, Vice Provost, Faculty Engagement and 
  Development, and Professor of Media, Culture, and 
  Communication, New York University (NYU).......................     4
Rejsjo, Martina, Vice President, Nasdaq MarketWatch..............    11
Wegner, Kirsten, Chief Executive Officer, Modern Markets 
  Initiative (MMI)...............................................    10

                                APPENDIX

Prepared statements:
    Fender, Rebecca..............................................    34
    Lopez de Prado, Marcos.......................................    53
    McIlwain, Charlton...........................................    66
    Rejsjo, Martina..............................................    80
    Wegner, Kirsten..............................................    89

              Additional Material Submitted for the Record

Foster, Hon. Bill:
    Written statement of Public Citizen..........................    98


                         ROBOTS ON WALL STREET:

                          THE IMPACT OF AI ON

                          CAPITAL MARKETS AND

                         JOBS IN THE FINANCIAL

                           SERVICES INDUSTRY

                              ----------                              


                        Friday, December 6, 2019

             U.S. House of Representatives,
             Task Force on Artificial Intelligence,
                           Committee on Financial Services,
                                                   Washington, D.C.
    The task force met, pursuant to notice, at 9:42 a.m., in 
room 2128, Rayburn House Office Building, Hon. Bill Foster 
[chairman of the task force] presiding.
    Members present: Representatives Foster, Cleaver, Casten, 
Adams, Garcia of Texas; Loudermilk, Hollingsworth, and 
Riggleman.
    Also present: Representative Himes.
    Chairman Foster. The Task Force on Artificial Intelligence 
will now come to order.
    Without objection, the Chair is authorized to declare a 
recess of the task force at any time. Also, without objection, 
members of the full Financial Services Committee who are not 
members of this task force are authorized to participate in 
today's hearing.
    Today's hearing is entitled, ``Robots on Wall Street: The 
Impact of AI on Capital Markets and Jobs in the Financial 
Services Industry.''
    The Chair will now recognize himself for 5 minutes for an 
opening statement.
    First off, thank you all for joining us today for what 
should be a very interesting hearing of this task force. Today, 
we are looking at exploring how artificial intelligence (AI) is 
being deployed in capital markets, from automated trading, to 
portfolio allocation, to investment management decisions.
    We are also going to consider how the use of this 
technology is changing the nature of work in financial 
services, rendering some jobs obsolete and changing the skill 
sets needed to excel in others.
    It would not be much of an exaggeration today to say that 
Wall Street, quite literally, is run by computers. Long gone 
are the days where traders would be screaming orders on the 
floor of the New York Stock Exchange and financial analysts 
would use TI calculators and pore over the ticker tape and 
financial statements to glean insights into a company's value.
    I actually hear about those days from the limo driver who 
takes me back, who used to be a floor trader on the Merc.
    Today, trades are automated and orders are executed in 
milliseconds or microseconds. Passive exchange-traded funds 
(ETFs) have proliferated, relying on algorithmic models to 
ensure the fund's holdings of shares are properly weighted to 
whatever index or benchmark it is tracking. Quantitative hedge 
funds, or quant funds, use algorithms that scour all sorts of 
market data to find the stocks that have the most price 
momentum or the highest dividends or look for correlations in 
the market and in the external data feeds to provide the most 
value for investors.
    And I think it is very notable that a lot of the shakeup 
that we are seeing in those markets is really a reflection of, 
sort of, the winner-take-all nature of digital economies--that 
any digital business, purely digital business is a natural 
monopoly, and as more of finance becomes digitized, you are 
going to see more and more of the rewards go to a smaller and 
smaller number of dominant players. And I would like to 
emphasize, that doesn't mean they are evil; it is just simply a 
natural reflection of the nature of the digital marketplace.
    Other asset managers may use algorithms to perform complex 
research and analysis in real-time on big data sets. This could 
include scouring of social media sites, satellite information, 
internet traffic, online transactions, and just about anything 
else you can think of. This is, I guess, good in terms of 
having the market reflect all known data, but there are abusive 
corners. For example, imagine what it would be worth if you had 
a 10-second early look at Trump's Twitter feed, how much money 
you could make trading off that, for example.
    The three types of computer-managed funds--index funds, 
ETFs, and quant funds--make up about 35 percent of the 
approximately $31 trillion American public equities market. 
Human managers, such as traditional hedge funds and other 
mutual funds, manage just 24 percent of the market.
    The rise of the so-called computerization of our stock 
markets has a number of benefits. The costs of executing trades 
has gone way down, sometimes to zero dollars, and there is more 
liquidity in the market. Passive funds charge less than 1 
percent of assets under management each year, while active 
managers often charge 20 times that much.
    It certainly creates additional questions as well, however, 
as in the 2010 flash crash. And the more recent mini flash 
crashes have shown algorithmic trading can sometimes cause 
unpredictable consequences that create market volatility. It 
can also exacerbate information asymmetry between different 
types of investors, as firms with more and faster access to 
enormous data sets are able to obtain a competitive advantage.
    Another broader question is how these developments are 
impacting the nature of jobs in the financial services 
industry. A recent Wells Fargo research report estimated that 
technological efficiencies would result in about 200,000 job 
cuts over the next decade in the U.S. banking industry. While 
these cuts will certainly affect back-office, call-center, and 
customer-service positions, the pain will be widespread. Many 
front-office workers, such as bankers, traders, and financial 
analysts, could also see their head count drop by almost a 
third, according to a McKinsey & Company report released 
earlier this year.
    The report also found that 40 percent of existing jobs at 
financial firms could be automated with current technology. So, 
if you spend your whole day staring at a big screen, and 
particularly if you are receiving a large paycheck, your job 
will be at risk.
    Understanding the skills that will be needed to excel in 
the financial services industry of tomorrow and how we can 
encourage these skills is one of the issues that we must tackle 
head-on and tackle early. In a world where many functions can 
be done by automated AI models, what role does that leave for 
humans?
    So, I very much look forward to hearing from our witnesses 
on these issues.
    With that, I would like to recognize the ranking member of 
the task force, my friend from Georgia, Mr. Loudermilk, for 5 
minutes.
    Mr. Loudermilk. Thank you, Mr. Chairman.
    And I want to thank each of our witnesses who are here 
today. Thank you for taking the time to be here to discuss this 
issue. While the rest of America is fixated on other things 
going on here, this is something that may not resonate on the 
major networks, but it is something that is very important, and 
has an impact on our lives, positively but also potentially 
negatively, and it is important that we look into this.
    And, as you know, today, the task force will examine the 
intersection between technology and the capital markets. In 
recent years, there have been many technological developments, 
including the adoption of artificial intelligence and 
automation, that have redefined and reshaped trading and 
investing.
    The first trades on the New York Stock Exchange were made 
in the late 1700s using a manual, paper-intensive process. For 
many years, buyers and sellers communicated about orders over 
the phone. Today, trading and investing are done on digital 
platforms, and investors can trade securities from virtually 
anywhere in the world using modern technology.
    Electronic trading has benefited the markets in many ways. 
It has been positive for investors by leading to lower overhead 
and transaction costs, which has contributed to record 
investment returns over the last decade.
    Several major asset management firms now offer zero-percent 
commissions, which means investors can buy and sell stocks 
essentially for free and can capture more of the growth of 
their investments. This would not be possible without 
electronic trading.
    Digital trading platforms also provide investors with 
access to low-cost financial research and advice 24 hours a day 
using robo-advisors.
    Electronic trading also makes markets more efficient by 
allowing faster searches for prices, better processing of large 
sets of data, and more transparent price information. The 
proliferation of technology can also lower firms' barriers to 
entry, foster more competition, improve risk management, and 
increase market access for investors.
    In addition to these core benefits, there are many other 
cases of companies using AI to improve efficiencies in the 
capital markets in unique ways. For example, some clearing 
companies are using AI to optimize the settlement of trades and 
enhance cybersecurity and fraud detection. Some self-regulatory 
organizations are also using AI in regtech and market 
surveillance.
    While there are many benefits to electronic trading, it can 
also present new challenges.
    One challenge, which is at the forefront of our discussion 
today, is the disruption of the job market. While the rise of 
automated trading has displaced many floor traders, job 
opportunities in fields like code writing, cloud management, 
telecommunications, fiber optics, and data analysis are 
growing.
    There is some concern that high-frequency trading can 
contribute to volatility, but new evidence suggests that high-
frequency trading does not increase volatility and can actually 
improve liquidity. There is also some concern that firms that 
don't have the latest technology could be competed out of the 
markets.
    It is important to keep in mind that not all types of 
electronic trading are the same, and I look forward to learning 
more from the witnesses about the differences between automated 
trading, algorithmic trading, high-frequency trading, and 
computer trading.
    Finally, I look forward to exploring the legislative and 
regulatory issues in this space. One issue that I think needs 
to be addressed is the protection of source code, because 
algorithms are traders' core intellectual property. They must 
be protected.
    We passed a bill out of this committee and the House on a 
bipartisan basis last Congress to ensure that the Securities 
and Exchange Commission issues a subpoena before obtaining 
these algorithms, rather than getting them through routine 
exams. Mr. Chairman, I hope that we will be able to work 
together on a bill this Congress.
    I thank you, and I yield back.
    Chairman Foster. Thank you.
    And, today, we are welcoming the testimony of Dr. Charlton 
McIlwain, vice provost for faculty engagement and development, 
and professor of media, culture, and communication at NYU; Dr. 
Marcos Lopez de Prado, professor of practice at the Engineering 
School, Cornell University, and chief investment officer of 
True Positive Technologies; Ms. Rebecca Fender, CFA, senior 
director, Future of Finance Initiative at the Chartered 
Financial Analyst Institute; Ms. Kirsten Wegner, chief 
executive officer of the Modern Markets Initiative; and Ms. 
Martina Rejsjo, head of Nasdaq market surveillance, Nasdaq 
Stock Market.
    Witnesses are reminded that your oral testimony will be 
limited to 5 minutes, and without objection, your full written 
statements will be made a part of the record.
    Dr. McIlwain, you are now recognized for 5 minutes to give 
an oral presentation of your testimony.

     STATEMENT OF CHARLTON MCILWAIN, VICE PROVOST, FACULTY 
 ENGAGEMENT AND DEVELOPMENT, AND PROFESSOR OF MEDIA, CULTURE, 
          AND COMMUNICATION, NEW YORK UNIVERSITY (NYU)

    Mr. McIlwain. Chairman Foster and Ranking Member 
Loudermilk, thank you for inviting me to testify.
    While my written remarks cover four key areas, my oral 
remarks focus on two: the implications of automation on the 
workforce; and mitigating algorithmic discrimination and bias.
    We have ample reason to be concerned about automation's 
future in the financial services sector. First, the financial 
services sector is ripe for automation and algorithm-driven 
innovation. Second, the fintech sector is on the rise. Third, a 
large number of workers will likely be displaced in the 
financial services sector even if automation and AI development 
is projected to create new types of jobs.
    If all of this is true, then the cause for concern is 
clear. It lies with the fact that African Americans and Latinx 
workers, in particular, are already vastly underrepresented in 
the financial services sector workforce.
    African Americans, Hispanics, and Asians make up only 22 
percent of the financial services industry workforce. African-
American representation in the financial services sector, at 
both entry-level and senior-level jobs, declined from 2007 to 
2015. Less than 3.5 percent of all financial planners in the 
U.S. are Black or Latinx. African Americans make up just 4.4 
percent and Hispanics just 2.9 percent of the securities 
subsector. Asians make up just 2.8 percent of the central 
banking and insurance subsectors.
    My point is simple: Racial groups that are already 
extremely underrepresented in the financial services industry 
will be most at risk in terms of automation and the escalation 
of fintech development. This is especially true given the vast 
underrepresentation of African Americans and Latinx in the 
adjacent technology sector workforce.
    If we are to mitigate the likelihood that automation will 
disproportionately and negatively affect those already 
underrepresented in the financial services industry, we must 
plan ahead long into the future rather than allowing the market 
to run its course towards predictable outcomes.
    Now, to the subject of deterring algorithmic bias. 
Certainly, one way to mitigate against algorithmic bias is to 
develop best practices for constructing and deploying 
algorithmic systems and providing more oversight from industry, 
government, and nongovernmental bodies who are able to assess 
how such systems are used and the outcomes they produce.
    This includes technical solutions that make algorithms more 
transparent and auditable and mitigate against potential biases 
before such systems gain widespread use rather than trying to 
simply correct their effects once their damage is done.
    But I want to emphasize that, especially when it comes to 
mitigating the potential disparate outcomes that biased 
algorithms might have on individuals and communities of color, 
simple reliance on technical fixes by technologists is not a 
complete solution.
    I want to end by drawing on the wisdom of Bayard Rustin, a 
former civil rights leader who had a sophisticated 
understanding of computerized automation and algorithmic 
systems as they existed in his time. He said, ``Today, the 
unskilled and semi-skilled worker is the victim, but 
cybernation invades the strongholds of the American middle 
class as once-proud white-collar workers begin sinking into the 
alienated world of the American underclass. And as the new poor 
meets the old poor, we find out that automation is a curse. But 
it is not the only curse. The chief problem is not automation 
but social injustice itself.''
    Take as a final example the findings from a recent National 
Bureau of Economic Research study, titled, ``Consumer-Lending 
Discrimination in the FinTech Era.'' Their researchers sought 
to determine whether an algorithmic system could reduce 
discrimination in mortgage lending as compared to traditional 
face-to-face lending processes.
    Their findings were mixed. Yes, the algorithmic system 
discriminated 40 percent less than the traditional process, but 
that also meant that the process still discriminated against a 
large number of Black and Latinx loan applicants. Further, even 
though the algorithmic system did not, on balance, discriminate 
in terms of loan approval, it did discriminate against Black 
and Latinx users in terms of price.
    One of the key conclusions of the study states that both 
fintechs and face-to-face lenders may discriminate in mortgage 
issuance through pricing strategies. We are just scratching the 
surface of the role of pricing strategy discrimination in the 
algorithmic area of data use.
    In short, algorithmic lending may reduce discrimination 
relative to face-to-face lenders, but algorithmic lending is 
not, alone, sufficient to eliminate discrimination in loan 
pricing. Even with the aid of a fair, accurate, and transparent 
algorithmic system, racial discrimination persists.
    Thank you again for allowing me the opportunity to 
contribute to these proceedings.
    [The prepared statement of Mr. McIlwain can be found on 
page 66 of the appendix.]
    Chairman Foster. Thank you.
    Dr. Lopez de Prado, you are now recognized for 5 minutes to 
give an oral presentation of your testimony.

  STATEMENT OF MARCOS LOPEZ DE PRADO, PROFESSOR OF PRACTICE, 
 ENGINEERING SCHOOL, CORNELL UNIVERSITY, AND CHIEF INVESTMENT 
              OFFICER, TRUE POSITIVE TECHNOLOGIES

    Mr. Lopez de Prado. Thank you, Chairman Foster, Ranking 
Member Loudermilk, and distinguished members of this task 
force. It is an honor to be asked to contribute to this 
committee today.
    As a result of recent advances in pattern recognition, 
supercomputing, and big data, today, machine-learning 
algorithms can perform tasks that until recently only expert 
humans could accomplish.
    An area of particular interest is the management of 
investments, for two reasons. First, some of the most 
successful hedge funds in history happen to be algorithmic. The 
key advantage of algorithmic funds is that their decisions are 
objective, reproducible, and can be improved over time. The 
second advantage is that the automation enables substantial 
economies of scale and cost reductions. Automated tasks include 
ordered execution, portfolio construction, forecasting, credit 
rating, and fraud detection.
    Financial AI creates a number of challenges for the over 6 
million people employed in the finance and insurance industry, 
many of whom will lose their jobs, not because they will be 
replaced by machines but because they have not been trained to 
work alongside algorithms. The retraining of these workers is 
an urgent and difficult task.
    But not everything is bad news. As technical skills become 
more important in finance than personal connections or 
privileged upbringing, the wage gap between genders, 
ethnicities, and other classifications should narrow. In 
finance, too, math could be a great equalizer.
    Retraining our existing workforce is of critical 
importance; however, it is not enough. We must make sure that 
the talent that American universities help contribute and 
develop remains in our country. The founders of the next 
Google, Amazon, or Apple are this very morning attending a math 
or engineering class at one of our universities. Unlike in the 
past, odds are these future entrepreneurs are in our country on 
a student visa and that they will have a very hard time 
remaining in the United States unless we help them. Unless we 
help them, they will return to their countries of origin with 
their fellow students to compete against us.
    On a different note, I would like to draw your attention to 
two practical examples of regtech--that is, the application of 
machine-learning algorithms to regulatory oversight.
    A first embodiment of regtech is the crowdsourcing of 
investigations. One of the most challenging tasks faced by 
regulators is to identify market manipulators among oceans of 
data. This is literally a very challenging task, like searching 
for a needle in a haystack.
    A practical approach is for regulators to enroll the help 
of the data science community, following the example of talent 
competitions or the Netflix Prize. Accordingly, regulators 
could anonymize transaction data and offer it to the worldwide 
community of data scientists, who would be rewarded with a 
portion of the fines levied by regulators against wrongdoers. 
The next time that financial markets experience something like 
the flash crash, this tournament approach could lead to a 
faster identification of potential market manipulators.
    A second embodiment of regtech is the detection of false 
investment products. Academic financial journals are filled 
with false investment studies as a consequence of backtest 
overfitting. Financial firms offer online tools to overfit 
backtests, and even large hedge funds fall constantly for this 
trap, leading to investor losses.
    One solution is to require financial firms to record all 
the backtests involved in the development of a product. With 
this information, auditors and regulators could compute the 
probability that the investment strategy is overfit, and this 
probably could be reported in the funds' promotional material.
    Finally, I would like to conclude my remarks with a 
discussion of bias. Yes, machine-learning algorithms can 
incorporate human biases. The good news is we have a better 
chance at detecting the presence of biases in algorithms and 
measure that bias with greater accuracy than on humans. The 
reason is that we can subject algorithms to a batch of 
randomized, controlled experiments, and we will calibrate those 
algorithms to perform as intended. Algorithms can assist human 
decision-makers by providing a baseline recommendation that 
humans can override, thus exposing biases in humans.
    As algorithmic investing becomes more prevalent, Congress 
and regulators can play a fundamental role in helping reap the 
benefits of this technology while mitigating its risks.
    Thank you for the opportunity to contribute to this 
hearing, and I look forward to your questions.
    [The prepared statement of Dr. Lopez de Prado can be found 
on page 53 of the appendix.]
    Chairman Foster. Thank you.
    Ms. Fender, you are now recognized for 5 minutes to give an 
oral presentation of your testimony.

 STATEMENT OF REBECCA FENDER, CFA, SENIOR DIRECTOR, FUTURE OF 
FINANCE INITIATIVE, CHARTERED FINANCIAL ANALYST (CFA) INSTITUTE

    Ms. Fender. Chairman Foster, Ranking Member Loudermilk, and 
members of the task force, thank you for inviting me to testify 
here today. My name is Rebecca Fender, and I am the senior 
director of the Future of Finance Initiative at the CFA 
Institute, which is our thought leadership platform.
    CFA Institute is the largest nonprofit association of 
investment professionals in the world, with 170,000 CFA 
charterholders in 76 countries. CFA Institute is best known for 
its Chartered Financial Analyst designation, the CFA Charter, 
which is a rigorous, three-part, graduate-level exam. To earn 
the designation, charterholders must also have at least 4 years 
of industry experience.
    CFA Institute is a nonpartisan organization and seeks to be 
a leading voice on global issues of transparency, market 
efficiency, and investor protection.
    Earlier this year, CFA Institute published a paper on the 
investment professional of the future, examining the changing 
roles and changing skills of the industry in the next 5 to 10 
years. Among the CFA Institute members and candidates we 
surveyed, 43 percent think the role they perform today will be 
substantially different in 5 to 10 years' time. And it was 
greater than 50 percent among financial advisors, traders, and 
risk analysts. Another 5 percent do not think their role will 
exist by then.
    One of the catalysts is technology. CFA Institute sees the 
impact of technology on jobs in the investment industry as a 
pyramid. At the foundation, we have basic applications. 
Everyone will need to learn to do things differently, and they 
must be more comfortable using and understanding technology. 
Some people will face tech substitution, but many more will 
have their roles adapted. In the middle, there are specialist 
applications, where technology will enhance work. And at the 
top, there are hyperspecialist roles that will be less common 
but very valuable. This includes roles at quant firms and AI 
labs.
    CFA Institute believes the key to this evolution is ongoing 
learning. Our exam curriculum now includes material about 
machine learning. And among the members and candidates we 
surveyed in our recent report, 58 percent have interest in 
data-analysis coding languages, like Python and R. Similarly, 
data visualization and data interpretation are areas that more 
than half have expressed interest in.
    In terms of the role of artificial intelligence in the 
investment industry, the organizing principle we see is: 
artificial intelligence plus human intelligence, or AI+HI.
    In these middle and top levels of that technology 
hierarchy, investment management and technology teams work 
together. AI techniques can augment human intelligence to free 
investment professionals from routine tasks and enable smarter 
decision-making. Investment professionals will spend less time 
finding and entering data and more time ensuring models are 
consistent with how markets work. AI unlocks the potential of 
unstructured data and can identify patterns in information more 
efficiently than humans. AI can amplify an investment team's 
performance but cannot replicate its creativity.
    In our recent paper, ``AI Pioneers in Investment 
Management,'' authored by my colleague, Larry Cao, we have 
identified three types of AI in big-data applications that are 
emerging in investment managemen: first, the use of natural 
language processing, computer vision, and voice recognition to 
efficiently process text, image, and audio data; second, the 
use of machine-learning techniques to improve the effectiveness 
of algorithms used in investment processes; and third, the use 
of AI techniques to process big data, including alternative and 
unstructured data, for investment insights.
    We find that relatively few investment professionals, about 
10 percent, are currently using AI and machine-learning 
techniques in their investment processes. However, here are a 
few examples from our case studies of what the AI pioneers are 
doing.
    First, Goldman Sachs' sell-side research team is better 
able to analyze national concrete companies supplying the 
construction industry by using geospatial data of 9,000 U.S. 
quarries that each act as local businesses.
    Second, the data science team at American Century 
Investments studied psychology textbooks to determine patterns 
of deception in children and criminals. They then applied 
machine learning to these patterns in their earnings calls to 
determine where spin, omission, obfuscation, and blame are 
being used.
    Finally, Bloomberg has had a sentiment analysis product 
available since 2009 which analyzes the potential effect of 
news stories on valuations. They process 2 million documents a 
day through their machine-learning platform. This was 
alternative data used only by hedge funds at first, but now 
many of their clients use it.
    Just as the investment industry is beginning to employ 
greater technology, regulators can look at new data in the 
world of regtech. This speed and volume of data presents a new 
surveillance challenge. Regulators will need to have the tools 
and resources to keep pace with changes.
    Thank you again for the opportunity to testify today, and I 
look forward to your questions.
    [The prepared statement of Ms. Fender can be found on page 
34 of the appendix.]
    Chairman Foster. Thank you.
    Ms. Wegner, you are now recognized for 5 minutes to give an 
oral presentation of your testimony.

 STATEMENT OF KIRSTEN WEGNER, CHIEF EXECUTIVE OFFICER, MODERN 
                    MARKETS INITIATIVE (MMI)

    Ms. Wegner. Thank you, Chairman Foster, Ranking Member 
Loudermilk, and members of the AI Task Force. It is an honor to 
discuss the role of automation of the markets and our 
deployment of artificial intelligence in the financial services 
industry and our future workforce.
    I am Kirsten Wegner, chief executive officer of Modern 
Markets Initiative. We are an education and advocacy 
organization comprised of automated trading firms. We operate 
in over 50 markets globally and, together, employ over 1,600 
people. Our advisory board, which is half women, promotes 
responsible innovation, including advancing a diverse workforce 
in our industry.
    Over the past decades, we have seen automated trading 
leading to much of the replacement of the exchange-floor-based 
intermediaries you see in 1980s Wall Street movies. Technology, 
as you have noted, has reduced the cost of trading for the 
average investor by more than half over the past decade, both 
in direct trading costs and in savings through tighter bid-ask 
spreads.
    So if you are an investor in a 529 college savings plan, a 
pension fund, or a 401(k), then you have benefited from today's 
low-cost trading and all of the dependable liquidity that we 
see in the markets. And studies have shown that over a lifetime 
of savings, investors have 30 percent more in their bank 
accounts as the result of the automation.
    Now, as we look ahead, there are four points that I want to 
discuss here in the oral testimony.
    First, global competition to adopt the latest AI 
technologies will make human decision-making more efficient in 
terms of speed, processing time, depth of data, and it is going 
to confirm more efficiencies and cost savings for U.S. 
investors across-the-board. Already, competition in the markets 
has resulted in near-zero-commission online trading from 
Fidelity, Charles Schwab, and Robinhood, and we have seen a 
rise in the ETF industry from those efficiencies. Similarly, 
automated trading has brought down overall trading costs to a 
fraction of the price from decades ago.
    Second, we can expect to see a proliferation of regtech as 
AI becomes increasingly valuable for individual firms and 
regulators to police the markets more efficiently. AI 
functionality in regtech includes monitoring, reporting and 
compliance, and processing of regulatory filings; loan 
origination processing; detection and reporting of illegal and 
irregular trading; and detection of cyber risk.
    And, notably, I want to point out that through public-
private partnerships, firms can play a role in working with a 
regulator to share those limited resources in AI and to share 
cutting-edge technology. Since 2017, several Modern Markets 
Initiative members have welcomed the opportunity to work 
together with FINRA in public-private partnerships. We are 
contributing our know-how while welcoming deploying artificial 
intelligence together to surveil the markets.
    So automated trading firms are incentivized to detect bad 
actors, because we, too, can be the victims of fraud. And as 
bad actors become more sophisticated globally, it is absolutely 
vital that financial regulators have the funding resources so 
they, too, have the technological capacity and access to AI and 
automated technologies to be a strong and effective cop on the 
beat.
    Third, as AI technology matures, we can expect increased 
demand for high-quality, robust data, including alternative 
data, to provide what I call the crude oil for the engines of 
AI. This entails large quantities of complex data that humans 
alone cannot digest. So I think we are going to see policy 
questions arise around this proliferation of data; I think it 
was already noted, questions of competition and antitrust in 
the digital marketplace. We are going to see increasing 
discussion of intellectual property rights and ownership rights 
of that data and questions of access to that data and the cost 
of data.
    I think alternative data has been successful in helping 
establish a credit history for the underbanked. That is one 
positive. But I think we need to continue discussions 
surrounding algorithm bias. And, in my prepared testimony, I 
have noted next steps, including industry-led initiatives, to 
share best practices, utilize ethics officers, and regtech 
approaches.
    And, last, I want to talk about the future of the 
workforce. AI and automation can and should be a tool rather 
than a replacement for humans. Some jobs will disappear, and 
others will grow. Areas of growth we can expect to see are in 
the computer occupations, jobs related to the transmission, 
storage, security, privacy, and integrity of data, the fiber-
optics industry. They are all going to be fueling the AI 
economy.
    There is massive existing demand for qualified 
technological talent across virtually all sectors of our 
economy, particularly in the financial sector. The current 
baseline participation for women, and particularly women of 
color, is something that leaves room for substantial 
improvement, and that is something we are focused on. And a 
skilled workforce for tomorrow's Wall Street is only as good as 
the companies that are there to invest in technology.
    I thank you for your time.
    [The prepared statement of Ms. Wegner can be found on page 
89 of the appendix.]
    Chairman Foster. Thank you.
    And, Ms. Rejsjo, you are now recognized for 5 minutes to 
give an oral presentation of your testimony.

STATEMENT OF MARTINA REJSJO, VICE PRESIDENT, NASDAQ MARKETWATCH

    Ms. Rejsjo. Thank you, Chairman Foster and Ranking Member 
Loudermilk, for the opportunity to testify on the impact of AI 
on our capital markets.
    Many people associate AI with high-tech and movies such as 
``The Matrix,'' and ``Terminator,'' but we at Nasdaq strongly 
believe that we can use this technology to target a wholly 
different prey: the fraudster.
    As you know, Nasdaq has extensive experience leveraging 
technology to operate our markets and markets around the world 
to protect participants and investors. We operate 25 exchanges 
and 6 clearinghouses around the globe. And we sell marketplace 
technology--trading, clearing, and surveillance systems--to 
hundreds of the world's markets, regulators, exchanges, 
clearinghouses, and broker-dealers.
    Our internal surveillance department is monitoring the 
markets for insider trading, fraud, and manipulation, as well 
as handling real-time events in the market. The accessibility 
of the markets and the increase in players with the ability to 
deploy manipulative strategies using their own technology and 
the exponential increase in data quantities can act as the 
perfect ecosystem for market manipulators to hide amongst the 
noise. This increased complexity in monitoring presents new 
challenges for the surveillance team relying on preconceived 
parameters and known factors to detect manipulative patterns.
    Our surveillance program is using algorithmic coding to 
detect unusual market behavior, running over 40 different 
algorithms in real-time, utilizing over 35,000 parameters. In 
addition to real-time surveillance, there are over 150 patterns 
covering post-trade surveillance to identify a wider range of 
potential misconduct. The team proactively develops tools and 
procedures to increase the quality of surveillance and to meet 
changing demands in the markets.
    But with the manner in which patterns are currently 
recognized, relying on known factors to describe behavior, it 
can be difficult to capture new behavior and to remain 
proactive rather than reactive to threats in the market.
    In addition, predefined expectations of what patterns look 
like can often limit alert results, depending on how alert 
parameters are calibrated. Calibration also presents a 
continued challenge when determining the best balance between 
false positives and true alerts.
    These challenges led to a calibration between the Nasdaq 
Machine Intelligence Lab, Nasdaq's market technology business, 
and the Nasdaq U.S. Surveillance Team, to enhance surveillance 
capacities with the help of artificial intelligence.
    Using AI to detect abnormal behavior patterns is based on 
the notion that manipulative behavior can be identified by 
signals in the markets, that a scheme to defraud market 
participants often has a specific pattern to it. There is a 
price rise or decline, an action is taken, and the trading is 
then back to normal. So, this signaling concept leads to new 
ways to look at pattern detection.
    By leveraging AI, detection models are not tied to static 
logic or parameters. We are able to train the AI machine based 
on visual patterns of manipulation, and we started to look at 
this spoofing pattern.
    The machine must then further train with human input, and 
then transfer learning was used to expand the scope of this 
project beyond spoofing. Transfer learning leveraged AI to 
apply a model developed for a specific task at the starting 
point for a model on the second task.
    By using deep-learning and human-in-the-loop techniques, 
the new models for detecting market abuse with our initial 
spoofing examples indicated usable results with 95 percent 
fewer examples than typically required.
    The inclusion of AI into the detection function will allow 
us to focus the effort on in-depth investigations of potential 
manipulative behavior instead of triaging a high number of 
false positives.
    But, to be clear, the human input is still of critical 
importance, both in analyzing the output from the surveillance 
system but also in continuously training the machine to produce 
more and more accurate outputs.
    The massive growth in market data is a significant 
challenge for surveillance professionals. Billions of messages 
pass through a larger market on an active day. In addition, 
market abuse attempts have become more sophisticated, putting 
more pressure on surveillance teams to find the needle in the 
data haystack.
    By incorporating AI, we are sharpening our detection 
capabilities and broadening our view of market activity to 
safeguard the integrity of our financial markets.
    Surveillance is a critical use case for AI, but Nasdaq is 
also looking to apply it in other businesses. For example, we 
are using a version of AI, natural language processing, in the 
listings business to facilitate the compliance review of public 
company filings.
    In closing, we are convinced that this use case for AI will 
benefit investors and the resiliency of the U.S. market and the 
other markets that we serve.
    Thank you for the opportunity to testify, and I am happy to 
answer your questions.
    [The prepared statement of Ms. Rejsjo can be found on page 
80 of the appendix.]
    Chairman Foster. Thank you.
    And I will now recognize myself for 5 minutes for 
questions.
    I should also mention to the Members present, it looks like 
the latest estimate for votes is now 11:30, so we may, in fact, 
have time for a second round of questions for Members who are 
interested. We will have to play it by ear.
    Dr. Lopez de Prado, you note in your testimony that today, 
data vendors offer a wide range of data sets--and I think other 
witnesses mentioned that--things that were not available a 
couple of years ago. And not only the data itself but the 
processing power to analyze it and the real-time delivery of 
that data is becoming more and more important to successfully 
trade on it.
    Could you just illuminate for us what are some of the more 
interesting data sets that you now see being used?
    Mr. Lopez de Prado. Certainly. It is a combination of data 
sets. On one hand, we have access now to credit card 
transactions, geolocation data, satellite images, 
transcriptions from earning calls, engineering data, and data 
from engineering processes like exploration and production 
companies that allow us to better estimate where the wells are 
for extraction of oil or fracking--all sorts of data.
    Keep in mind, please, that 80 percent of all data recorded 
today was generated over the past 3 or 4 years. Going back to 
history, going back to Mesopotamia, there is a lot of data 
around, data that we aren't even aware of but is just being 
scraped from websites and such.
    So all of this data can be used to understand what is the 
psychology of people, what is the state of mind of people, 
understanding people are more inclined today to take risks or 
to, for instance, relocate their assets to fixed income instead 
of stocks; trying to understand from news articles, as one of 
my colleagues mentioned, what are the narratives associated 
with particular companies.
    The amount of data today is staggering, and this is only 
going to increase because the storage of data is becoming 
cheaper every day and the processing power is increasing. So, 
this is definitely a trend that is not going to stop.
    Chairman Foster. Yes. And as I think I mentioned in my 
opening remarks, that has a danger of driving monopoly, the 
returns to scale--because you get more correlations to look at 
with your AI if you have the full range of data.
    And so, this will naturally cause those smaller players in 
the market to be less effective, and less profitable. And I 
think, that is probably what you are seeing in high-frequency 
trading, the consolidation that you are seeing there.
    Now, is there any way around this? And how hard should we 
lean against the natural tendency to monopoly here in financial 
trading?
    Mr. Lopez de Prado. There are two schools of thought in 
this regard.
    Number one, there are a number of academics who believe 
that this consolidation is not necessarily negative, in the 
sense that the few survivors that are able to consolidate, for 
instance, high-frequency trading, today are operating like 
utilities. They are not making the kind of returns that they 
were able to obtain 9 years or 10 years ago. Essentially, what 
happens is that they break even. These technologies are 
becoming so expensive that they have to spend this time and 
money in order to achieve a profit that is dwindling.
    There are a number of academics who believe that, actually, 
consolidation is not necessarily negative. There is, on the 
other hand, of course, the problem that a small number of 
operators could have a grip on the market, and it also could 
cause a domino effect if one of them fails to provide 
liquidity.
    So, there is a need to strike a balance between, on one 
hand, preventing too much consolidation, and on the other hand, 
also favoring competition between these operators.
    Chairman Foster. Yes. Ms. Wegner, you mentioned that this 
actually netted out--or, at least, electronic trading generally 
netted out very positively for someone's retirement account, 
that it actually, because of the lower bid-offer spreads and 
transaction costs, that, actually, it was--I think you quoted 
30 percent--
    Ms. Wegner. Correct.
    Chairman Foster. --more in your retirement account as a 
result of this.
    So, similarly, when AI is widely deployed, if it is very 
effectively deployed, in principle we get a more efficient 
capital allocation across our country. And so is, actually, the 
best strategy to let a small number of very dominant players 
have access to all the data sets to get a more efficient 
economy?
    Ms. Wegner. Yes. I think it is absolutely--
    Chairman Foster. Or are we better off just letting a 
thousand flowers bloom?
    Ms. Wegner. Sure. I think it is absolutely vital that we 
encourage policies that promote strong competition in this 
space. And with high-frequency trading and automated trading, 
we have seen such fierce competition over the past decade or 
two that we are approaching near-zero latency speed, we are 
approaching the speed of limits of--
    Chairman Foster. But also more monopolization. I think my 
time is up here, but this is something I intend to return to--
    Ms. Wegner. Sure, absolutely.
    Chairman Foster. --if we get a chance here.
    Ms. Wegner. I am happy to respond.
    Chairman Foster. Thank you all.
    I now yield 5 minutes to the ranking member, Mr. 
Loudermilk.
    Mr. Loudermilk. Thank you, Mr. Chairman.
    Ms. Wegner, as you know, the SEC has experienced some 
cybersecurity difficulties, especially in the 2016 EDGAR data 
breach. I think it is important for the SEC to only obtain 
proprietary trading algorithms, if absolutely necessary, with a 
subpoena. So I was wondering if you could discuss why it is 
important for source code to be protected?
    Ms. Wegner. Sure. That is a very good question.
    The real lifeblood of automated trading and the, kind of, 
secret sauce is the source code--that is the valuable 
intellectual property that the different firms are competing 
against each other with, not just domestically but globally. 
And just like a self-driving car company needs to keep its 
algorithms and source-code intellectual property protected from 
misappropriation, so do algorithmic traders rely on government 
protection for their intellectual property.
    There was a proposal a number of years ago to perhaps 
collect IP source code and put that in a government repository 
just in case it was needed. That never came to light, but it is 
still something we are absolutely educating policymakers on. 
This should be, I think, a bipartisan area of interest, to 
ensure that we have a globally competitive marketplace that 
protects intellectual property rights.
    Mr. Loudermilk. I appreciate that from my time in the 
military working in intelligence. We had a principle we lived 
by because of the sensitivity of the data that we collected and 
maintained, which was, ``If you don't need something, don't 
keep it,'' which means you don't have to protect what you don't 
have.
    And my concern is how vulnerable the industry becomes, 
because, quite frankly, the government tends to be the weakest 
link when it comes to data security in some aspects. So, I 
think obtaining that source code is not only just a violation 
of the privacy right of the business, the coder, but it could 
also be a national security risk.
    Ms. Wegner. I think that is right. If bad actors were able 
to breach the source code, it would be presenting an 
opportunity for manipulating the markets or cyber risks. So it 
is absolutely vital that we protect the intellectual property 
rights of source code.
    Mr. Loudermilk. Thank you.
    Ms. Fender, the adoption of artificial intelligence in 
electronic trading can disrupt the job market and displace 
floor traders, but technologies also create a need for more 
workers in other fields.
    Today, we have about a million people working in the 
airline industry, but in the early 1900s, The Washington Post 
led with a headline that said, ``Man Will Never Fly and 
Shouldn't,'' and part of their argument was the displacement of 
people in the job market.
    Could you touch on the job fields that are growing because 
of the use of AI in the capital markets space?
    Ms. Fender. Yes. Thank you.
    As you noted, there are many ways that jobs are changing, 
and adaptation is really the key.
    We surveyed industry leaders, the people who are doing the 
hiring, and we asked, ``What are the most important skills 
going forward? Maybe it is not necessarily in the job 
description. What are the skills underlying who will succeed in 
the future?''
    And they talked about something called T-shaped skills. 
This is an idea that, if you think about the letter ``T,'' you 
have the vertical bar where there is deep subject-matter 
expertise and a horizontal bar where you can cut across 
different disciplines.
    And if you think about fintech, we have big risk if there 
is ``fin'' over here and ``tech'' over here, and they aren't 
talking. So, the ability to connect the two is where there is a 
lot of opportunity.
    These are the innovators. This is an area where you will 
see more research needing to be done so that we understand what 
the trends are.
    And the key thing is that people have to ask the right 
questions. Firms are realizing you have to think about the 
return on investment (ROI) of gathering this data. And many of 
the machine-learning people will say a large percentage of the 
data isn't that useful. So you have to be smart about how to do 
that and start the process with investment professionals.
    Mr. Loudermilk. Okay. So, what you are getting at is not 
all the jobs are going to be just as deep intellectual, being 
able to code and understand algorithms and all that, but there 
are ancillary jobs that come about because of the development, 
is that a fair statement?
    Ms. Fender. Yes, definitely. We don't think, for example, 
that all CFA charterholders need to become programmers, but we 
think they are going to have data scientists on their teams, 
and they are going to need to speak the language and work 
together.
    Mr. Loudermilk. Okay.
    Ms. Rejsjo, I want to talk about the use of artificial 
intelligence in fraud detection. I view cybersecurity as the 
biggest challenge that we face in this nation, from a business, 
government, and personal perspective.
    Can you touch on quickly--I'm running out of time--how 
algorithms are used to detect unusual market behavior?
    Ms. Rejsjo. Yes. As I said, we really rely on the algorithm 
coding to pick up on the unusual patterns that we see. 
Everything needs to be compared to something that is usual, 
right? So we program things to pick up on the unusual things 
based on historical comparison on specific stocks, how they 
have been trading in the past. So that is what we do already 
and we have done for a long time.
    The new thing here--
    Mr. Loudermilk. Thank you.
    Chairman Foster. At this point, I think we will leave that 
hopefully to your next round of questioning.
    The gentlewoman from North Carolina, Ms. Adams, is now 
recognized for 5 minutes.
    Ms. Adams. I thank the Chair very much for putting this 
hearing together. We appreciate it.
    And, also, those of you who have come to testify, thank you 
very much for your comments and for your work.
    Automation technologies, which enable the transfer of tasks 
from human labor to machines, affect approximately 6.4 million 
workers employed in the financial services industry. Specific 
industries like credit lending and capital markets are being 
affected by AI, as human tasks involving data analysis, 
decision-making, and compliance are replaced by machine-
learning robots. This shift in job automation could predict 
which jobs in financial services will be replaced and what new 
jobs could be created.
    Ms. Wegner, specifically examining loan underwriting 
compared to the traditional methods of meeting a loan 
application in person, to what extent does AI replace or 
augment the work done by loan officers, credit counselors, or 
other credit underwriters?
    Ms. Wegner. That is a very good question.
    In the consumer lending context, I think it is very 
important that AI is the tool for humans when they are 
extending credit and extending loans, that there are systems in 
place to ensure that there isn't any sort of algorithmic bias. 
And, in my prepared testimony, I noted some suggestions. Our 
members are not engaged in the consumer lending context, but we 
have our own insight.
    I think that loan companies, individually or collectively, 
could employ ethics officers to ensure that there isn't 
algorithmic bias in the lending context. I think it is 
important that industry members share lessons learned as they 
explore how they are democratizing access to credit and finding 
the most efficient ways to extend that credit.
    I think it is really vital that we act now to make sure, as 
we are building out this system, that we minimize the risk for 
algorithmic bias in consumer lending. I think it is very vital.
    Ms. Adams. Thank you, ma'am.
    Is the U.S. properly equipped to remain competitive in the 
financial services workforce?
    This question is to Dr. Lopez de Prado and to Ms. Fender.
    Mr. Lopez de Prado. The U.S. is the leader in the financial 
services industry today. My concern is that this leadership is 
being challenged by the fact that: first, we are not investing 
as much in AI as other countries; and second, the fact that we 
are educating our competitors.
    In my remarks, I mentioned that I am very concerned that 
the innovators of the future are attending today a class in our 
universities but they will not be allowed to stay. And, as a 
result, yes, we are very competitive, and this ability to train 
these skills is going to turn against us if we are not able to 
retain this talent.
    Ms. Adams. Okay.
    Ms. Fender?
    Ms. Fender. We have seen that--again, it is early days for 
how this changes our industry, with only about 10 percent 
actually using these techniques. But what we are seeing is that 
firms are doing AI labs, they are doing innovation hubs. They 
realize that this is something they need to be proactive about.
    And so we are seeing--out of our case studies, we had a 
criteria that things in our case studies had to actually be in 
practice. There is a lot of talk out there, but things that are 
actually in practice, five of the nine are here in the U.S.
    Ms. Adams. Great. Thank you.
    Dr. McIlwain, are we adequately teaching the skills needed 
for the jobs of the future?
    Mr. McIlwain. Thank you for the question.
    I think we are adequately teaching those skills; I think 
the question is, who has access to that teaching?
    And so, when we think about underrepresentation of certain 
individuals and members of the workforce who are not getting 
the types of education that are needed for the jobs that may be 
coming online as a result of automation and AI development--and 
so I think, if we are to have a full pipeline of folks who are 
able to receive what it is that we teach in our colleges, 
universities, even high schools and younger, then we have to be 
more proactive about making sure that all people have access to 
that teaching and that information.
    Ms. Adams. No one left behind.
    Mr. McIlwain. Absolutely.
    Ms. Adams. Okay. I appreciate that.
    I am going to yield back, Mr. Chairman. Thank you very 
much.
    Chairman Foster. Thank you.
    The gentleman from Indiana, Mr. Hollingsworth, is now 
recognized for 5 minutes.
    Mr. Hollingsworth. I appreciate each of you being here 
today, and I appreciate the chairman for holding this hearing. 
This is an important topic, something I have been really 
passionate about since arriving here in Congress.
    And, Dr. Lopez de Prado, I appreciate your comments, 
because what you have touched on is something that I have been 
an ardent believer in for a long time, which is that the big 
arm of the Federal Government isn't going to stop the growth of 
this technology, isn't going to cease the investment in AI 
either here or around the world. And while we can shape the 
context by which that technology flows, we are not going to dam 
up and stop that technology.
    And so, when people say job losses may result on account of 
this, there is a lot of fear and a lot of desire to put an end 
to that and to stop that, but I like how you referenced a lot 
of training and retraining that may need to happen--training 
individuals who are graduating from school to ensure they have 
the skills that are necessary in a 21st-Century workplace, but 
also ensuring that those who are already in the workplace have 
the opportunity to get the retraining to continue their 
competitiveness. And as we see further growth and development 
in AI, it will require more and more frequent retraining to 
stay ahead of that, to stay relevant in that field. That is a 
very competitive field, right?
    But the second thing you touched on is something I am even 
more ardent about. We educate a lot of kids in this country. We 
do higher education in this country better than anywhere else 
in the world. We bring a lot of talent into this country. We 
invest a lot in those kids, and then we politely ask them to 
leave at the end of their tenure here, right? That is 
embarrassing, that is idiotic, that is stupid, and I hate that.
    I want to find a way to attract talent into this country 
and retain talent into this country, not because I believe it 
is a zero-sum game but because I believe that this country can 
provide a crucible for technological development that you can't 
find elsewhere in the world. And I think that technology will 
benefit humankind over all the world in the long run, and I 
want to make sure we do that.
    So, I really appreciate you touching on those topics, and I 
really appreciate that investment of time.
    Ms. Wegner, I know that you have a source-code event coming 
up. Today? Tomorrow?
    Ms. Wegner. This afternoon.
    Mr. Hollingsworth. This afternoon, to talk about source 
code again. And I really appreciate you continuing to educate a 
lot of people about how important that is. Where I go, all the 
way across the district in Indiana, I hear more and more about 
how much technology, how much investment, how much IP is in 
things that aren't readily seen, either in business processes, 
in the source code, in the technology underpinning automation 
itself. And so I know how important that is, and I really 
appreciate you bringing that to light.
    All that being said, I wanted to ask Ms. Rejsjo a question 
that is maybe a little bit far afield from what we are talking 
about today.
    I had some people in my office earlier this week who were 
very complimentary, frankly, of Nasdaq surveillance services. 
They were very complimentary--they were public companies--and 
how, when something seems amiss in the markets, Nasdaq was very 
quick to pick up the phone and say, ``Something seems amiss. 
Let's figure out what is going on here.''
    One of the things that is very important back home is 
biotech. A lot of biotech firms are based in Indiana. People 
don't know that. We are trying to get the word out about it. 
They are concerned about market manipulation, specifically with 
regard to short-selling. And they are promoting this idea that 
there should be more disclosure around short-selling, similar 
to many long positions.
    Now, they came in and said that disclosure around short-
selling would really help us, as a firm, better understand 
those that might have interests adverse to us, because we can't 
really track that right now. But the counter-argument that they 
made was, gosh, Nasdaq seems to be doing a really good job of 
figuring out when there is potential manipulation.
    I wondered if you might touch on that. Is disclosure in 
short-selling something that would benefit the market, 
something that would benefit these firms? Or do you feel like 
you have enough of the ability to track potential market 
manipulation on the back end?
    And, again, I am not pejorative against short-sellers. I 
just want to make sure that it is legitimate action, not market 
manipulation.
    I wonder if you might comment on that in the last minute.
    Ms. Rejsjo. I think disclosure is a big part of 
surveillance.
    Mr. Hollingsworth. Yes.
    Ms. Rejsjo. Information is always needed to understand what 
is happening.
    Mr. Hollingsworth. Okay.
    Ms. Rejsjo. I do think that what we have today is 
sufficient. As you say, we have a lot of patterns that are 
detecting manipulation such as short-selling, or, I might say, 
the troublesome part of short-selling.
    Mr. Hollingsworth. Right.
    Ms. Rejsjo. I mean, short-selling is legal, right?
    Mr. Hollingsworth. Right. Of course.
    Ms. Rejsjo. So it is really to detect what is then being--
how it is used in an abnormal way or in a sort of manipulative 
kind of way.
    Mr. Hollingsworth. Yes. So you feel like you can detect the 
activity that would be illegal or abnormal or different 
adequately. The question is, what do we do with it after that 
point, is maybe where we should focus public policy attention? 
Is that fair?
    Ms. Rejsjo. Yes. But to be fair, also, there are other 
parts within Nasdaq that handle more of the policy questions.
    Mr. Hollingsworth. Okay.
    Ms. Rejsjo. But for me as a surveillance practitioner, I do 
think that the disclosure we have and the tools we have to 
monitor the markets are--
    Mr. Hollingsworth. Are adequate.
    Ms. Rejsjo. Yes.
    Mr. Hollingsworth. Great. I think that is an important 
question. Because when they were in my office, I think that is 
the question: Where do we need to focus public policy 
attention? And perhaps it is beyond surveillance, and focus 
more on some of the penalties or some of the actions that 
happen with the enforcement agencies.
    With that, I will yield back.
    Chairman Foster. Thank you.
    And I am very encouraged that one of the areas of 
bipartisan agreement here is the insanity of this business of 
awarding people their Ph.D.s and pushing them back on an 
airplane.
    And so that is one of the reasons I was proud to introduce, 
this session of Congress, H.R. 4623, the Keep STEM Talent Act 
of 2019, designed to--it is a rifle shot to just exactly solve 
this problem. And I really look forward to my colleagues' 
support on this.
    And now, I recognize the gentlewoman from Texas, Ms. 
Garcia, for 5 minutes.
    Ms. Garcia of Texas. Thank you, Mr. Chairman. And thank you 
again for holding this hearing.
    And thank you to all the witnesses. Good morning, and 
welcome.
    I wanted to focus on a couple of issues that some of you 
have already talked about. Like Ms. Adams, I am particularly 
concerned about jobs. My district is in Houston, and is 77 
percent Latino. It is also working-class, so we are always 
concerned about jobs. I am encouraged that you all seem to have 
the consensus that, although there will be some job 
displacement, there will be new jobs created.
    My main concern, of course, is whether or not we do have 
the skill sets, Mr. McIlwain, to transfer those skills or to 
make sure that we can fill those jobs. Because, in the end, 
that is what really matters to families in my district.
    But I am also concerned with automation and the difference 
between AI and automation, and how it can work together, 
specifically in the area of regulatory compliance.
    Ms. Fender, in your experience, has AI and automation 
affected institutions' regulatory compliance? Is it improving? 
Is it still a work in progress? Or how are we doing?
    Ms. Fender. Thank you. That is a very good question. And, 
again, I think it is about--it is still kind of early to know. 
We hear so much about what is coming, and yet--so compliance 
areas are growing in firms clearly. And now we have more and 
more data, and regulators are going to be able to have the same 
sort of data.
    The question is, is there a greater risk, maybe, of insider 
information now? You collect more data, and people can see lots 
of different patterns out there. And if they see that and can 
trade on it before the market, then you have challenges for the 
SEC, I think, in terms of Reg FT and so forth.
    Ms. Garcia of Texas. Okay.
    Ms. Wegner, can that be used to simplify and ensure 
regulatory compliance with the Federal agencies in charge of 
supervising the capital markets.
    Ms. Wegner. Sure. I think, as the data sets become more 
complex, as you just alluded to, I think it is going to be 
vital that the regulators have the resources to have their own 
AI, either independently of the companies, or together with the 
companies through public-private partnerships as the bad actors 
become more sophisticated, and we are talking about global bad 
actors. We need a strong cop on the beat here in the U.S. And I 
think it is very important that the private sector work 
together with regulators to ensure that they have those 
resources, and that Congress really ensures that the SEC and 
the CFTC have the resources they need, because the systems are 
becoming much more complex and regtech is evolving, but needs 
to keep up with the pace of that technology.
    Ms. Garcia of Texas. I think that is a big concern of this 
committee, those bad actors, as you have described them.
    So how can AI assist us with anti-money-laundering 
compliance's suspicious activity reporting? Are we prepared for 
that? I know we did a codel to several countries. And things 
are getting more and more sophisticated, and it seems like the 
bad actors have more money and better things, to find ways to 
hide the money. Do we have what we need to detect it and to 
ensure that we can catch it?
    Ms. Wegner. It is vital that we focus on this. And I would 
say Haimera Workie, who is the new head of innovation at FINRA, 
has an excellent group. They just established themselves this 
year. They are a fantastic resource. They are working together 
with other regulators, with private sector participants to 
gather information about best practices, and to really make 
sure we have the best technology. This is 100 percent something 
we need to be focused on.
    Ms. Garcia of Texas. In your opinion, do you think that our 
regulators and our oversight entities are well-prepared in this 
arena, or what else should we be doing?
    Ms. Wegner. I think we need to be investing in technology. 
There is always room for more technology with the regulatory 
agencies. I think MIDAS at the SEC has been a very positive 
example of the SEC using very sophisticated technology to 
surveil the markets, but I think this is a constantly evolving 
space, as everyone here has noted. We have to just keep very 
much on our tiptoes on this, and keep on investing in this 
area.
    Ms. Garcia of Texas. Okay.
    Ms. Fender, did you want to add something?
    Ms. Fender. I think the more data we have, the more complex 
it gets, right? And one of the other things that we are really 
concerned about is the investor protection side too.
    If bad data goes into these models, they can be marketed in 
many different ways. And so, disclosures are really important. 
Understanding your clients, understanding where the money comes 
from, and understanding what clients are really getting all 
kind of goes together.
    Ms. Garcia of Texas. Thank you. Thank you, both of you. And 
I yield back.
    Chairman Foster. Thank you. The gentleman from Virginia, 
Mr. Riggleman, is recognized for 5 minutes.
    Mr. Riggleman. Thank you, Mr. Chairman. I want to thank all 
of the witnesses for being here today. I am so happy that all 
of you are here, so I am not showing any favoritism. I would 
particularly like to welcome Ms. Fender from the CFA Institute, 
which is located in my district in Charlottesville, Virginia. 
The CFA Institute provides a host of resources for 
professionals who work in the financial services industry, who 
are among the most qualified and adhere to the highest codes of 
standards in the financial industry. I am honored to have such 
a distinguished group reside in the Fifth District. Although 
Ms. Fender is not a constituent herself, her organization 
employs many of them, so I am thrilled to see you here today. 
Welcome, and welcome to all of you.
    I will start with you, Ms. Fender. You probably knew that 
was going to happen. Can you talk about how CFA is adapting the 
charter to these AI and machine-learning innovations in the 
investment industry?
    Ms. Fender. Thank you very much. And I'm pleased to be here 
representing Virginia.
    The CFA Institute is really the global standard for 
investment practitioners. The people who have our credential 
are the portfolio managers for your 401(k). They are the chief 
investment officers at the public pension funds. They are the 
people who are really safeguarding the financial futures of so 
many people. And so, it is imperative for us to keep up-to-date 
on what we teach. I mentioned earlier in my testimony that we 
just added machine learning into our curriculum. And this is a 
significant indication that we are seeing the market change. 
And we need to prepare people.
    We have a group called our practice analysis team. And they 
are out there all the time going to these conferences, figuring 
out what is the next thing that people need to know, because 
global demand for investment management is growing, and 
especially for those who really combine both competence and 
ethics.
    Mr. Riggleman. There really is a reason I asked that 
question. My prior job, and we talked about monopolization of 
the data and things of that nature. I wanted to monopolize as 
much data as I could for data interactions when looking at sort 
of critical infrastructure analysis when I worked for the 
Office of the Secretary of Defense. We had about 40 people 
looking at this, so we had to look at all data, multi domain 
across stovepipes, and see how that actually includes that data 
or to analyze or aggregate that data, consolidate it, aggregate 
it, analyze it, and then execute using that data based on how 
we templated human behavior.
    So looking at AI and ML rules, I guess that I will start 
with Ms. Rejsjo, I am going to ask a few of these, because this 
is the exciting part for me, is the technology part. When we 
did this, we had multiple data sets that people had never seen 
before. We talked about the challenges of data. We had multiple 
data sets. We had data we had never sort of aggregated, and 
combined with other data sets.
    So we thought we had the right answer, and we found out we 
didn't, in trying to template human behavior analysis. Do you 
think that is something you are going to see more of in the 
future, is that there won't be a human in the loop, and there 
will be more sort of human templating, or machine-learning 
rules to sort of mimic what human behavior does with certain 
rule sets? Do you think we are going to see more and more of 
that, taking humans out of the loop, looking at actually any 
type of analysis, or fraud, or anything of that nature?
    Ms. Rejsjo. I think that we are a long way from that. I 
think that for now, the way that we do it is really to have the 
data that we have. For us, it is really much more the order and 
trade data that we already have and that we analyze. Now, we 
are just applying a new technique to give us a better overview 
that is not that parameter-driven. But for us, still, I really 
think that the human in the loop is the way to go, because 
there is much more analysis that needs to be applied after the 
output has come. And I think that is going to be there for a 
while.
    Mr. Riggleman. It is interesting that you said that. We 
actually thought we could take the human out of the loop in 
some of our processes, and found out it was not a good idea, 
with some of the things that we did. I see some heads nodding 
back there. We tried to do that.
    Dr. Lopez de Prado, you were talking about, there could be 
some advantages to sort of aggregating as much data in one 
place as we can, right? And then looking in the gaps of that 
data. That is the thing I have been trying to wrap my arms 
around. My whole job was not competition. It was to monopolize 
all the data. And then to use competition to give us the best 
algorithmic solutions that we could first for first, second, 
and third order effects of what happened to a specific part of 
the network.
    This is a tough question, because to be this objective in 
40 seconds is going to be probably ridiculous. But when you are 
looking at this, do you think--and I know this is a tough 
question--with all the proprietary technologies out there, do 
you think there will be a voluntary sharing of that data if we 
find something that is very good, across multiple sets? So, for 
example, another company, do you think we will have that type 
of sharing for proprietary solutions based on algorithmic types 
of analysis? Do you think that will ever happen? Or do we think 
we have to sort of force that to happen when we monopolize that 
kind of data, if that is makes sense.
    Mr. Lopez de Prado. Are you referring to sharing these 
technologies--
    Mr. Riggleman. Yes.
    Mr. Lopez de Prado. In private and public companies?
    Mr. Riggleman. Yes.
    Mr. Lopez de Prado. When you look at the NASA model, 
actually, there has been a lot of transfer of IP between the 
agency and various contractors. So that could be a model that 
could work for the CFTC and the SEC. In particular, in my 
remarks I mentioned the crowdsourcing of investigations, how 
the companies or private participants could establish 
tournaments to help agencies identify market manipulators.
    Mr. Riggleman. Thank you very much. And I yield back the 
balance of my time.
    Chairman Foster. Thank you.
    The gentleman from Illinois, Mr. Casten, is recognized it 
for 5 minutes.
    Mr. Casten. Thank you, Mr. Chairman. And thank you all so 
much for being here today.
    Back in my prior life, I had a head of engineering who had 
a theory that I have yet to prove wrong. He said, every advance 
in technology gives us more precision and less knowledge. This 
was a guy who started with slide rules, where he had to know 
the order of magnitude of his answer. And now, he got 16 
significant digits, and can never remember whether it was 
millions or billions. And, of course, in my lifetime, we have 
gone from foldable maps to GPS that can give me the exact 
latitude and longitude. And I can't tell you whether I am 
north, south, east, or west of where I started.
    AI has always struck me as sort of putting that 
acceleration on steroids. At one point, I built a genetic 
algorithm to predict the revenues of our utility business, and 
it was amazing. I cut our revenue forecast variance by 90 
percent, and I have no idea how it worked.
    And that is the power and the frustration. I mention that 
because I think most of you have talked about the consumer 
benefit that comes when we get all these AI algorithms out in 
the markets, and we get lower trading costs, lower bid ask 
spreads. And that is all terrific.
    A lot of you have also talked about bad actors and we can 
put up monitoring for that and that is also great. The concern 
I have is this tension between the transparency of the model, 
and whether the model can actually effectively replicate a bad 
actor that we don't understand, because it is fairly easy for 
me to imagine the trading algorithm that is tracking a bunch of 
data and has figured out how to bet on one country invading 
another and making money. I can imagine a trading algorithm 
that is looking at changes in currency flows for illegal 
activity that is not in itself illegal, but is arbitraging some 
spread that results from that.
    So Ms. Wegner, I wonder if you would comment on that 
tension between transparency and algorithm robustness? And to 
what degree we have or need regulatory tools to stipulate where 
we sit on that continuum?
    Ms. Wegner. Sure. I think transparency is absolutely vital. 
I think it is also very vital that regulators and the exchanges 
have the resources that, if they note any sort of irregularity 
in the markets, they can immediately identify that. And to the 
question of whether or not one needs to get source code, if 
there is a detection of some sort of illegal or irregular 
activity, then the regulator requests--
    Mr. Casten. But if I could just clarify, first, would you 
agree that the more transparent the algorithm, potentially the 
less powerful the algorithm?
    Ms. Wegner. I think to the extent that the algorithm is not 
subject to intellectual property rights, that transparency is 
absolutely vital. If we are talking about intellectual property 
rights in this source code of algorithm, that is proprietary 
information that if it was--
    Mr. Casten. By transparency, I am not referring to whether 
or not the public has access to the algorithm. I am referring 
to whether or not our human brains can understand how the 
algorithm works. I could give you the genetic algorithm I 
wrote. You couldn't understand what it is doing.
    Ms. Wegner. Sure. That question becomes more complicated in 
the machine-learning context, especially. You point to an 
interesting question, as the commands become self-acting in a 
way, they are basing their analysis on the existing data sets. 
I don't think we are totally there yet, but I think that is 
something we definitely need to explore, what does our policy 
answer to, because that is an interesting balance.
    Mr. Casten. This question is for you, but really for all of 
the panelists. I think thinking about that problem before it 
gets there, because it strikes me that there will be pressure 
for every trading firm to develop the most powerful algorithms, 
which, by definition, at some level, are going to be the ones 
that we have the least ability to unpack and understand.
    Ms. Wegner. I think this is an important question that the 
industry should get together on and share their best practices, 
how do you balance that push for trade?
    Mr. Casten. For anybody who thinks they have a great answer 
in this, number one, how should we do that? And number two, to 
what degree do we need to coordinate internationally? Because 
even if we do everything in our country, since all of these 
markets are so interlinked, is this a U.S. problem, or is this 
an international problem? Does anybody have thoughts on that?
    Mr. Lopez de Prado. If I may, this is a very important 
distinction. Black boxes in finance tend to be less reliable 
than transparent solutions, particularly in finance, because we 
are dealing with problems where the signal-to-noise ratio is 
very low. Unlike, for instance, in astrophysics research, why 
is the signal-to-noise ration low in the finance world because 
of competition, because of arbitrage? Otherwise, everybody 
would be able to extract profits from the market. So because of 
that, when we deploy black box solutions in finance, the 
solutions can identify patterns that are not real, they are 
just balancing the noise and confound these patterns with these 
noise patterns with actual signal, leaving to investment 
studies that will fail. So one solution would be for in 
investor to understand very carefully when a product is based 
on a black box solution as opposed a transparent machine-
learning solution.
    Mr. Casten. Thank you. I yield back. I would welcome any of 
your comments. If you have any follow-up in writing, please 
share.
    Chairman Foster. As I mentioned, we are likely to have 
another round for Members who are interested here.
    The gentleman from Missouri, Mr. Cleaver, who is also the 
Chair of our Subcommittee on National Security, International 
Development and Monetary Policy, is recognized for 5 minutes.
    Mr. Cleaver. Thank you, Mr. Chairman. And I really 
appreciate you calling this hearing and we appreciate all of 
you giving us your time.
    I don't know how we are going to deal with AI and human 
beings. Long before we had flip phones, Captain Kirk had one, 
and long before we had the smartwatches, Mr. Spock had one. And 
a lot of attention is always paid to Hollywood, particularly in 
science fiction, and the military, our own military.
    So a lot of people have their eyes on a fearful future, as 
it relates to AI. And to be straight, I am one of those, I am 
conflicted. I know we can't hold back the wind. It is 
inevitable that we are going to see more and more of this in 
the future. And I am not sure that we ought to try to hold it 
back. But to the degree that we can control it, that is what I 
think we ought to do. And that is where I am concentrating most 
of my answers.
    Dr. McIlwain, first of all, thank you for being here. But I 
am wondering how inclusive this new technology is right now, 
and what can we do to make sure that in the future, every 
component of our great mosaic in the United States is a part of 
it?
    Mr. McIlwain. Thank you for that question, and I share a 
little bit of your fear, because what we know persists as 
technology changes, as technological advances are made is that 
some people, and typically, the same groups of people, are left 
out, left behind, disadvantaged. And so, even as technology is 
unpredictable, some of those exclusions are very much 
predictable.
    I think those exclusions are present in our current market, 
as most of the folks in this panel have at least alluded and 
nodded to; that is, when we look at our technology sector, 
those who are prepared to be part of that sector, those who are 
currently working, building the technologies of today and 
tomorrow are tremendously unrepresentative of our full 
democracy of all the citizens of our country. And I think 
representation makes a tremendous difference. I think the place 
we are in today with respect to some of the inequalities and 
devastations that technologies, AI and automation included, 
have wreaked, because not everyone has been included in the 
decision-making about what technologies will be built, why, for 
what purposes, who they will advantage and disadvantage.
    So I think moving forward, we have to change that. That is, 
we have to invest strategically in building a more inclusive 
workforce in these sectors that are growing. That is the 
technology sector and the financial sector as well.
    Mr. Cleaver. What do you think we should do, or any of you 
do right now, if we--we have young people interested in and 
committed to the future, and AI is an inevitable part of it. 
What should they do next week? What should young people be 
doing? How should we direct young people right now, who are 
scientifically gifted? What should we do?
    Ms. Wegner. I think we need to promote responsible 
innovation. I know our members support trying to get out there 
to the middle school students geographically across the 
country, a diverse population of people, and get them 
interested in STEM fields. I think there is a lot of 
opportunity for companies to partner with some of the public 
schools in a geographically diverse part of the country and 
help fund that, and just recruit now. Kids get interested in 
these fields from a young age. And we just have to get in there 
early and make sure that people see role models at the firms 
that we promote those public-private partnerships.
    Mr. Cleaver. My time is up, so thank you very much. Thank 
you, Mr. Chairman.
    Chairman Foster. Thank you.
    And, I guess, we have time for a brief second round of 
questions here.
    We have had sort of two different narratives that have been 
going on here. One is the, sort of, optimistic narrative of 
the--well, I guess the T-shaped skills or machine intelligence, 
human intelligence paring, albeit second, augmented human 
intelligence. And then we also have the sort of intermediate 
way of transfer learning, where you would actually use one 
field of expertise and transfer what was learned there to 
another field thereby replacing multiple machine parings. An 
example of that was the example from the geniuses at Goldman 
who were analyzing satellite imagery of quarry activities to 
predict cement pricing, and so on in the future; and then, 
potentially, using transfer learning so that knowledge could be 
transferred to copper mining or whatever else it was.
    On the other hand, there is an alternative narrative that 
you just aggregate all the data you can, and just say, I want a 
general purpose, learning, trading algorithm to look at all 
satellite data and look for all market correlations. And that 
would detect not only the cement market, it would look at the 
parking lots of Toys R Us to predict that Toys R Us was going 
bankrupt because they didn't have many cars on Black Friday.
    And so this sort of thing could be written once and 
deployed to replace tens of thousands of machine-human pairings 
on here, and, obviously, with much, much smaller labor input 
and need for humans. So which of the two narratives is going to 
end up winning, and how is it going to net out for human 
participation in this? Anyone who wants to tackle that tar 
baby?
    Ms. Fender. I can start, and just say that one of the 
foundational concepts in investing is that correlation is not 
necessarily causation. And so, we have a lot of data and we can 
see these patterns, but you need a human to ask, what is the 
right question? I mention also the example of going through the 
news stories with Bloomberg. And they said, the key question 
there was to go through the news article and not say, what do 
we think the author of this article wanted to get across, but 
what do we think people are hearing?
    So there are a lot of nuances really about how this is 
going to play out. And that is why, again, having sort of the 
collective intelligence and diverse perspectives is going to be 
important.
    Chairman Foster. Dr. Lopez de Prado?
    Mr. Lopez de Prado. Yes. I think that the two narratives 
have some part of truth. I think in the short term, we have 
reasons to be worried in terms of the transfer of knowledge, 
and the potential displacement that will occur as these 
technologies are more broadly deployed. But I think in the long 
term, we have reasons to be optimistic, because the next 
generation would be better prepared than our generation, or 
previous generations. It is very important that we give equal 
access to education. It is very important that we encourage 
kids to learn how to program, participate in math and 
engineering classes, and that we form the flexible workforce, a 
workforce that in the future, we don't know what these 
technologies will do in 20 years, that they are able to engage 
proactively.
    Chairman Foster. Is there a danger that this is going to 
squeeze all profitability out of financial services? That if 
you had complete knowledge of everything, and very efficient 
algorithms immediately trading on that knowledge, the 30 
percent improvement in your retirement savings, all of that 
money used to end up in the pockets of people with nice homes 
in Oyster Bay, and that is sort of the nature of things. And it 
may be that when we get this much more efficient economy with 
extensive deployment of AI, just the total amount of money left 
to be extracted will continue to go down the same way high-
frequency trading is sort of suffering that decline in margins.
    Mr. Lopez de Prado. One view, if I may, is that, in fact, 
having such a perfect market is not necessarily bad for 
society, meaning that the day that we go to our financial 
adviser and we receive the same treatment that we receive when 
we go to the doctor essentially, there is a product goal of, 
this is what you need to invest to achieve your retirement 
goals. I think that is a good outcome.
    Ms. Wegner. And as we see greater efficiencies, global 
advisors and other more efficient, I would say, asset managers, 
we will be able to deploy that efficiency to the masses. But I 
think it also raises a global competition question, because we 
are not just talking about competition domestically, we are 
talking about internationally, and we are not going to stop 
time all across the world, right? Other countries are 
innovating in AI. So it is inevitable we are going to be 
competing in that space and we want to keep the U.S. markets 
the envy of the world, I think.
    Chairman Foster. So if the future of financial advising is 
conversations with Alexa, I guess it comes down to, is the 
objective that a function that the AI running Alexa is 
maximizing, is that Amazon's profit? Or is it some linear 
combination of Amazon's profit and diversity inclusion, a 
secure retirement rather than steering people into products 
that are profitable for Amazon?
    Ms. Wegner. Right. I think the vital part here is, as you 
mentioned, we have competition, that there is not too much 
aggregation of power in one entity. We need to have policies 
that promote robust competition amongst, let's say, the robo 
advisors that make sure that data is accessible at competitive 
prices, so there is not a barrier to entry. This is going to 
be, I think, an exciting space where a Financial Services 
Committee meets a Judiciary Committee on antitrust issues, and 
meets a Commerce Committee. Finance is becoming more 
technology, and technology is becoming more finance. So, those 
are the right questions.
    Chairman Foster. Thank you. And I will yield 5 minutes to 
the ranking member.
    Mr. Loudermilk. Thank you, Mr. Chairman.
    Ms. Rejsjo, I would like to go back and kind of continue 
our conversation that we were talking about, cybersecurity, and 
using AI, and fraud. And I wasn't managing that time very well 
before, so could you explain further how Nasdaq is using AI in 
fraud detection?
    Ms. Rejsjo. Yes, I think it is important, just to start, 
that--I mean, the future is here, right? We have billions of 
data points. It is a massive amount of data that needs to be 
analyzed to capture anything that is fraudulent or manipulative 
in the market. So, we have that environment already. And what 
we have been doing so far is deploying algorithmic coding to 
sort of be able to process all of this data very fast. Our 
real-time surveillance is picking up on unusual behavior within 
seconds after it has happened in the market.
    So there is, really, a fast and efficient way to go through 
the data, but as it is growing and exponentially growing, there 
is the need, of course, to continue to invest in other ways of 
looking at it, where AI then comes in. It is a broader 
approach, and it doesn't have to be those parameters specific 
that we are today, so we can capture more things that are more 
sophisticated. Because as we have been talking about, it is not 
only us using this technique, the participants in the market 
are using it as well. So I think it is important for us to 
match their technology with our technology, when we are look at 
the types in the market with the behavior.
    Mr. Loudermilk. Thank you.
    Dr. Lopez de Prado, can you touch on the differences 
between automated trading, algorithmic trading, high frequency 
trading, and computer trading, how they are not all the same 
and what differentiates each of those?
    Mr. Lopez de Prado. Yes. Algorithmic trading consists of 
following some rules. A computer follows some rules in order to 
achieve a particular outcome. It does not require machine 
learning. Machine learning is the learning of patterns from a 
set of data with us directing that learning. Essentially, what 
happens is, you give to an algorithm a data set, and the data 
set identifies the pattern we were not aware of. So, that is 
machine learning.
    What was the third one?
    Mr. Loudermilk. The automated trading and high frequency.
    Mr. Lopez de Prado. Yes. Well, high frequency trading is 
algorithmic trading at a fast speed. It can happen with or 
without machine learning, so in the earliest stages, 2005, the 
high frequency trading or core without intelligence. Today, 
what we see is liquidity providers, market majors, hedge funds, 
deploy high frequency solutions with machine learning embedded.
    Mr. Loudermilk. Thank you.
    Ms. Wegner, we have had some discussion on the cost savings 
that have resulted from AI and automation in the capital 
markets. Do you see that these efficiencies are a significant 
reason behind the record returns investors have enjoyed in the 
last decade?
    Ms. Wegner. This has definitely contributed to the returns, 
every reduced incremental cost of trading adds up with 
compounding interest over time. So as the markets become more 
efficient, investors are going to have more in their 
pocketbooks, whether half of America invested in a 529 plan or 
otherwise for the net positive.
    Mr. Loudermilk. Okay. Thank you.
    I have no further questions, Mr. Chairman. I yield back.
    Chairman Foster. Thank you. The gentleman from Missouri is 
recognized for 5 minutes.
    Mr. Cleaver. Thank you, Mr. Chairman.
    I am interested in, how do we do planning now for the 
future? For example, we just updated our anti-money-laundering 
deal, or the Bank Secrecy Act. And I am sitting here now, and I 
introduced the bill, so I have been feeling pretty good about 
myself until you guys came up today. And I am thinking, why did 
we go through all of that? Because the bad guys are out there 
trying to figure out how they can exploit whatever we pass 
legislatively. How do you see an AI involved in anti-money-
laundering efforts, like the legislation that we hope the 
Senate will take up during our lifetime? Is there any way you 
think that can play a role, that AI can play a role in our 
money laundering bills or how we are trying to reduce it? We 
know we are probably never going to eliminate it.
    Mr. Lopez de Prado. This is a gargantuan problem. We have 
to tackle tremendous amounts of data, terabytes of data, and 
identify this needle in the haystack. I think a practical 
solution is for regulators to work together with data 
scientists, with the entire community, and crowdsource these 
problems. We need to anonymize this data, and give this data to 
the community so that the community help us enforce the law. Of 
course, they could be rewarded with part of the fines levied 
against wrongdoers, but I think that is a very doable approach, 
given: number one, how difficult it would be for the agencies 
to develop the kind of techniques that the wrongdoers are 
developing for bad purposes; and number two, the amounts of 
data that we need to parse through.
    Mr. Cleaver. We had the Treasury Secretary before our 
committee yesterday. I, of course, didn't even raise this 
issue. We have an agency, FinCEN, which is an investigatory 
part of Department of the Treasury. So, I am here wondering 
what they are doing to try to keep up with the technology, and 
what challenges they are going to face in the future. And you 
all have destroyed almost everything I was proud of, but we 
appreciate you coming here anyway. Thank you very much.
    I yield back, Mr. Chairman.
    Chairman Foster. Thank you. I would like to thank our 
witnesses for their testimony today.
    The Chair notes that some Members may have additional 
questions for this panel, which they may wish to submit in 
writing. Without objection, the hearing record will remain open 
for 5 legislative days for Members to submit written questions 
to these witnesses and to place their responses in the record. 
Also, without objection, Members will have 5 legislative days 
to submit extraneous materials to the Chair for inclusion in 
the record.
This hearing is now adjourned.
[Whereupon, at 11:05 a.m., the hearing was adjourned.]

[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]