[Senate Hearing 118-225]
[From the U.S. Government Publishing Office]
S. Hrg. 118-225
HARNESSING AI TO IMPROVE GOVERNMENT
SERVICES AND CUSTOMER EXPERIENCE
=======================================================================
HEARING
before the
COMMITTEE ON
HOMELAND SECURITY AND
GOVERNMENTAL AFFAIRS
UNITED STATES SENATE
ONE HUNDRED EIGHTEENTH CONGRESS
SECOND SESSION
__________
JANUARY 10, 2024
__________
Available via the World Wide Web: http://www.govinfo.gov
Printed for the use of the
Committee on Homeland Security and Governmental Affairs
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
U.S. GOVERNMENT PUBLISHING OFFICE
54-718 PDF WASHINGTON : 2025
COMMITTEE ON HOMELAND SECURITY AND GOVERNMENTAL AFFAIRS
GARY C. PETERS, Michigan, Chairman
THOMAS R. CARPER, Delaware RAND PAUL, Kentucky
MAGGIE HASSAN, New Hampshire RON JOHNSON, Wisconsin
KYRSTEN SINEMA, Arizona JAMES LANKFORD, Oklahoma
JACKY ROSEN, Nevada MITT ROMNEY, Utah
JON OSSOFF, Georgia RICK SCOTT, Florida
RICHARD BLUMENTHAL, Connecticut JOSH HAWLEY, Missouri
LAPHONZA BUTLER, California ROGER MARSHALL, Kansas
David M. Weinberg, Staff Director
Lena C. Chang, Director of Governmental Affairs
Evan A. Freeman, Counsel
Carter A. Hirschhorn, Research Assistant
William E. Henderson III, Minority Staff Director
Christina N. Salazar, Minority Chief Counsel
Kendal B. Tigner, Minority Professional Staff Member
Laura W. Kilbride, Chief Clerk
Ashley A. Gonzalez, Hearing Clerk
C O N T E N T S
------
Opening statements:
Page
Senator Peters............................................... 1
Senator Carper............................................... 11
Senator Hassan............................................... 13
Senator Butler............................................... 20
Prepared statements:
Senator Peters............................................... 25
WITNESSES
WEDNESDAY, JANUARY 10, 2024
Jennifer Pahlka, Author, Recoding America: Why Government Is
Failing in the Digital Age And How We Can Do Better, Former
U.S. Deputy Chief Technology Officer........................... 3
Beth Blauer, Associate Vice Provost for Public Sector Innovation,
Johns Hopkins University....................................... 5
Beth Simone Noveck, Ph.D., Chief Innovation Officer, State of New
Jersey, Professor of Experiential AI, Northeastern University.. 7
Alphabetical List of Witnesses
Blauer, Beth:
Testimony.................................................... 5
Prepared statement........................................... 40
Noveck, Beth Simone:
Testimony.................................................... 7
Prepared statement........................................... 46
Pahlka, Jennifer:
Testimony.................................................... 3
Prepared statement........................................... 27
APPENDIX
Partnership for Public Service statement for the Record.......... 69
Responses to post-hearing questions for the Record:
Ms. Pahlka................................................... 76
Ms. Blauer................................................... 77
Ms. Noveck................................................... 78
HARNESSING AI TO IMPROVE GOVERNMENT
SERVICES AND CUSTOMER EXPERIENCE
----------
WEDNESDAY, JANUARY 10, 2024
U.S. Senate,
Committee on Homeland Security
and Governmental Affairs,
Washington, DC.
The Committee met, pursuant to notice, at 10 a.m., in room
562, Dirksen Senate Office Building, Hon. Gary Peters, Chair of
the Committee, presiding.
Present: Senators Peters [presiding], Carper, Hassan,
Blumenthal, Ossoff, Butler, and Hawley.
OPENING STATEMENT OF SENATOR PETERS\1\
Chairman Peters. The Committee will come to order. We all
interact with the government throughout our lives, whether it
is applying for a small business loan, receiving financial aid
to pursue an education, accessing essential health care, or
applying for Social Security.
---------------------------------------------------------------------------
\1\ The prepared statement of Senator Peters appears in the
Appendix on page 25.
---------------------------------------------------------------------------
Our citizens turn to the government for all sorts of
critical services. As elected officials, we must ensure that
agencies provide those services as effectively as possible, and
artificial intelligence (AI) can help succeed in doing that
work.
This technology has the potential to make government
services more efficient, effective, and accessible for all
Americans. For instance, AI can easily translate crucial
documents into multiple languages.
It can operate 24/7 chatbots that provide our citizens with
interactive assistance. It can allow employees to tackle more
requests in less time with greater accuracy. AI can make
complex processes easier to navigate. The government's use of
AI to deliver services is certainly not new.
Three decades ago, the U.S. Postal Service (USPS) used it
to create faster mail delivery services. Today, in my home
State of Michigan, AI tools are providing translation services
for citizens applying for permits and for licenses.
This technology is already making a difference in the lives
of countless American civilians. But this is a watershed moment
for AI. These technologies grow more advanced nearly every day,
reaching more and more aspects of American life.
That is why, in the last year, I have convened five
hearings on artificial intelligence, and we have passed
legislation out of this Committee to address the challenges as
well as the opportunities posed by AI.
This builds on my previous legislation to provide
educational opportunities on AI, establish adequate training
resources, and provide agencies with guidance on how to
implement AI tools most effectively.
This Congress, our Committee has passed my bipartisan
Improving Government Services Act to encourage Federal agencies
to adopt proven customer service tools to help ensure Americans
get the assistance that they need. This bill builds upon the
hearing we convened last Congress to explore ways agencies can
build trust by improving the customer experience (CX).
This is a key moment to understand the capabilities of AI,
and how it can benefit both government employees and the
citizens that they serve, and under what circumstances. This
hearing will help us do just that.
We will examine the ways that AI can help deliver critical
services and improve the public's experience in receiving them.
We will outline the guardrails that this technology requires,
the training resources, privacy standards, and performance
metrics we need to properly implement AI tools, and our expert
panel of witnesses will help us understand how this can happen
at all levels of government.
We have a responsibility to do everything we can to ensure
government provides the most effective and efficient government
services. Succeeding in that work will help our constituents.
It will improve trust in government and strengthen our
democratic institutions. This hearing is an opportunity to
examine how artificial intelligence can achieve that mission.
It is the practice of the Homeland Security and Governmental
Affairs Committee (HSGAC) to swear in witnesses.
If each of you would please stand and raise your right
hand. Do you swear that the testimony that you will give before
this Committee will be the truth, the whole truth, and nothing
but the truth, so help you, God?
Ms. Pahlka. I do.
Ms. Blauer. I do.
Ms. Noveck. I do.
Chairman Peters. Thank you. You may be seated. Our first
witness is Jennifer Pahlka, and she is a renowned author and
expert on public digital services. She has served in a range of
roles across all levels of Government, including the U.S.
Deputy Chief Technology Officer.
And during her time in Government, she helped found the
United States Digital Service, fix healthcare.gov, and improve
the functionality of California's unemployment insurance
program.
Prior to her public service, Ms. Pahlka founded Code for
America, a nonprofit working to enhance government services
through digital tools. She is also author of Recording America,
which was named as one of National Public Radio (NPRs) best
books of 2023 and made President Obama's AI reading list.
Congratulations on that. Ms. Pahlka, you are recognized for
your opening comments.
TESTIMONY OF JENNIFER PAHLKA,\1\ AUTHOR, RECODING
AMERICA: WHY GOVERNMENT IS FAILING IN THE DIGITAL
AGE AND HOW WE CAN DO BETTER, FORMER U.S. DEPUTY
CHIEF TECHNOLOGY OFFICER
Ms. Pahlka. Thank you so much, Chair Peters and Members of
the Committee. Thank you for inviting me here today.
---------------------------------------------------------------------------
\1\ The prepared statement of Ms. Pahlka appears in the Appendix on
page 27.
---------------------------------------------------------------------------
My message today is simple, a powerful body like the U.S.
Senate has many tools at its disposal to shape the future of
the country. Those tools can be used to create mandates and
controls on the institutions that deliver for the American
people, or they can be used to enable these institutions to use
their judgment in the service of agreed upon goals.
A mandate and control framework will add rules and
processes. An enablement framework will focus on building
capacity, both by strengthening the workforce and reducing the
burdens on everyone in it. This critical AI moment calls for
enablement. The recent Executive Order (EO) on AI already
provides some new controls and safeguards.
The order strikes a reasonable balance between
encouragement and caution, but I worry that some of its
guidance will be applied inappropriately. For example, some
government agencies have long been using AI for day to day
functions like handwriting recognition on envelopes, and
agencies may now subject these benign, low risk uses to red
tape.
Caution is merited in some places and dangerous in others,
where we risk moving backward, not forward. Moreover, in many
areas of government service delivery, the status quo is frankly
not worth protecting.
We understandably want to make sure, for instance, that
applicants for government benefits are not unfairly denied by--
because of bias in algorithms. The reality is that to take just
one benefit, 1 in 6 determinations of eligibility for
Supplemental Nutrition Assistance Program (SNAP) is
substantially incorrect today. If you count procedural errors,
the error rate is 44 percent.
Worse are the applications and adjudications that have not
been decided at all, the ones sitting in backlogs, causing
enormous distress to the public and wasting taxpayer dollars.
Poor application of AI in these context could indeed make a bad
situation worse.
But for people who are fed up and just want someone to get
back to them about their tax return, their unemployment
insurance check, or even their company's permit to build
infrastructure, something has to change. We cannot double down
on the remedies that failed in the Internet age, and hope that
somehow it would work out in the age of AI.
We must finally commit to the hard work of building digital
capacity. the Office of Personnel Management (OPM) and the
White House have stated their intentions to build that capacity
by hiring tech talent. This is a case where strengthening the
workforce is also a matter of reducing burdens.
OPM's recent memo, for instance, will grant direct hire
authority for several AI related job classifications, and that
will remove a bit of the red tape agencies need to bring on
experts. That direct hire authority does not allow for the use
of pooled hiring across agencies, despite the fact that pooled
hiring has gotten us many excellent data scientists and other
tech roles much more quickly.
Agencies will have to run separate hiring processes for
each open position, which will take enormous amounts of time
and paperwork even with the direct hire authority. Congress
should ask OPM what authorities they need in order to change
this, and what resources they need to scale programs like this
highly successful Subject Matter Expertise Qualifying
Assessments (SME-QA) program. Then ask, what is the next
obstacle that they need removed.
I do not presume to know everything that is needed, only
that they operate in a highly constrained environment no longer
fit to the purpose it must serve. We must reduce burdens in
other areas as well. What the public wants from customer
service is answers, where is my check? Why did I get this
Internal Revenue Service (IRS) notice?
If we use AI to make it easier to talk to government, but
not to get those answers, we fail. The key reason those answers
are hard to come by is the enormous complexity of government
programs.
I worked on the pandemic unemployment insurance crisis and
encountered what is close to 10,000 pages of regulation
governing what should be a relatively simple program. A claims
processor working with my colleague kept calling himself the
new guy because he was still learning the ropes. He had been
with the agency for 17 years.
Recall that unemployment insurance dates back to the 1935
Social Security Act. We have been adding rule upon rule and
mandate upon mandate for close to 90 years now. We almost never
remove them. It is no wonder the program is still in peril. It
is collapsing under its own weight that Federal and State
agencies cannot shed on their own. It is tempting to say that
AI will help us by understanding those 90 years of accumulated
policy cruft for us.
This is appealing in the short term, but dangerous in the
long run. We cannot have a government so complex that one
algorithm talks to the other at such a level of complexity that
humans are out of the loop. But we can and should use AI to
suggest dramatic simplifications to those overwrought
frameworks, and make those new, leaner frameworks the law of
the land.
The greatest gift this body could give to the agencies and
the American public they serve is a massive, thorough
decluttering. AI makes it possible, but only the people in this
chamber can make it happen. Thank you very much.
Chairman Peters. Thank you, Ms. Pahlka, for your testimony.
Our next witness is Dr. Beth Blauer. Currently serves as
Associate Vice Provost for the Public Sector Innovation at John
Hopkins University.
In 2015, she founded the Bloomberg Center for Government
Excellence (GovEx) at Johns Hopkins to help Government use data
effectively to improve the lives of people across the United
States and around the world.
Since then, the GovEx has worked alongside thousands of
public officials, and Ms. Blauer has become a leading expert on
the use of innovative technology in Government services.
Ms. Blauer, you are recognized for your opening statement.
TESTIMONY OF BETH BLAUER,\1\ ASSOCIATE VICE PROVOST
FOR PUBLIC SECTOR INNOVATION, JOHNS HOPKINS UNI-
VERSITY
Ms. Blauer. Chair Peters, Committee Members, thank you for
inviting me to participate in today's hearing. It is an
absolute honor to share my views on AI, which reflect my
personal and professional expertise and are not necessarily a
position of Johns Hopkins University or medicine.
---------------------------------------------------------------------------
\1\ The prepared statement of Ms. Blauer appears in the Appendix on
page 40.
---------------------------------------------------------------------------
I come to this Committee with more than 20 years of public
sector experience. My first job in government was as a juvenile
probation officer, and I ended service more than a decade later
as the de facto Chief Operating Officer (COO) of the State of
Maryland, where I had the opportunity to oversee about 85
percent of the State's budget and a comparable amount of the
workforce through an innovative, data driven program called
StateStat.
Together with leaders across our State, we used data to
create better outcomes for all Marylanders, leading to stronger
schools, safer communities, a healthier Chesapeake Bay, and a
strong economic outlook, all accomplished by maintaining
fidelity to a robust data practice and evidence based
decisionmaking implemented across the whole of government.
In 2015, I founded the Bloomberg Center for Government
Excellence at Johns Hopkins University with a simple yet
ambitious goal, to help government, primarily cities, use data
effectively to improve the lives of residents across the United
States and around the globe.
Over the past eight years, GovEx has established itself as
a global leader, working alongside thousands of public
officials to achieve our goal. We are the go to resource for
governments as they seek to improve data proficiency and
develop data driven policy solutions, drawing on the most
cutting edge technology, including AI and advanced analytics.
Technology progress can present unique challenges in the
public sector, but where some see barriers, we see opportunity.
An effective AI approach relies on investment in core data
practices.
When public institutions attempt to dive into the deep end
of analytics, automation, or machine learning without first
building a strong data practice, they run the risk of
propagating disparate impacts, poorly informing decisions, and
creating bad outcomes.
When done right, a strong data practice will lead to safer
AI implementation that will anticipate resident need and
promote safer, more efficient, healthier, thriving communities.
We have learned from our work with cities that there is already
a strong body of evidence that AI and local government can
improve lives.
For example, cities are using AI to reduce traffic
congestion and manage fleet, ensuring prompt service delivery
and saving money. AI also have the potential to predict the
needs of residents before it is too late by identifying
households that are experiencing the early onset of economic
distress.Early detection allows public agencies to provide
immediate interventions that reduce barriers to aid and
expedite connecting the right services to those in need. For
example, cities have used AI to predict which residents are at
risk of becoming homeless by analyzing utility and property
violation data. This analysis has led to public interventions
coming sooner, keeping families in their homes, and saving
public resources. When local leaders ask us how to get started,
this is what we suggest.
No. 1, designate a senior leader. This should be a trusted
partner who knows when to deploy AI and when not to. Someone
who can influence service design, and someone with the power
and authority to lead governance and technical strategy while
asking the right questions.
No. 2, share early guidance. Be transparent about when AI
is deployed. Assess the risks. Take precautions to avoid bias
in delivery.Create, communicate an acceptable use policy, and
bring people to the table to discuss.
No. 3, you already are investing in technology that
incorporates AI, explore what you already have. Many technology
platforms are already doing this important integration, and big
and expensive investments should not be your first move.
No. 4, create space for experimentation. Integrate a review
of your AI implementation into ongoing performance routines. Be
clear before implementation about what success looks like.
Track progress and give your team the right balance of space
and support. We are learning as we go.
An iterative stance with reflection and learning periods
baked into the process creates fertile ground for cautious
experimentation. At the same time, some organizations may not
be ready to implement AI. For example, some municipalities may
learn that they need to get their data houses in order before
they can use data for predictive modeling. Some may realize
they do not have sufficient expertise on staff and need to
hire.
Others may find that their policies prevent AI
implementation, and they need to address that first. In order
to maximize the use of AI in public organizations, government
must invest in the capacity for technology adoption across all
levels and program areas. This means that we need to attract
new skills to the sector.
We need to elevate the capacity of the current workforce,
and we need to start investing in the creation of public space
to experiment with the new AI powered technologies. But there
are positive signs on the horizon. I would like to thank Chair
Peters and the Senators of both parties on this Committee for
your leadership on this important issue, as demonstrated by the
introduction of the bipartisan Improvement Government Service
Act.
This important piece of legislation would require agencies
to develop, implement, and submit to Congress a written,
comprehensive customer experience strategy that includes the
adoption of some of these best practices from the private
sector, and long term planning for customer service
modernization.
This is exactly the type of jumpstart and focus on the
basics that will greatly benefit the American people. Thank you
for your time today, and I am very happy to take any questions.
Chairman Peters. Thank you, Ms. Blauer. Our final witness
is Dr. Beth Noveck. Yesterday, Governor Murphy appointed Dr.
Noveck as the first Chief AI Strategist of New Jersey, where
she also currently serves as the State's first Chief Innovation
Officer (CINO).
Prior to that post, she served as a founding U.S. Deputy
Chief Technology Officer and head of the White House Open
Government Initiative under President Obama.
Outside of her dedicated public service, Dr. Noveck works
as a Professor of Experimental AI at Northeastern University,
where she directs the Burn Center for Social Change, as well as
its partner project, the Governance Lab.
For the last 20 years, Dr. Noveck has designed and built
public tools to improve governance and strengthen our
democracy. Dr. Noveck, you are now recognized for your opening
comments.
TESTIMONY OF BETH SIMONE NOVECK, PH.D.,\1\ CHIEF INNO-
VATION OFFICER, STATE OF NEW JERSEY, PROFESSOR OF
EXPERIENTIAL AI, NORTHEASTERN UNIVERSITY
Ms. Noveck. Thank you, Chair Peter, and thank you so much
for this opportunity to appear before the Committee, especially
in the company of such a distinguished panel. Despite having
funded the creation of the Internet, the public sector, we can
all agree, I think has really lagged in its adoption.
---------------------------------------------------------------------------
\1\ The prepared statement of Dr. Noveck appears in the Appendix on
page 46.
---------------------------------------------------------------------------
But now I think government can lead in the use of
artificial intelligence for public good. Because especially of
the power of AI to sort, to organize, and to summarize vast
amounts of data and text, AI has the potential to make
government more effective, more efficient, and more responsive
to residents.
AI is already helping us to create what I would call a more
conversational government, where we can supply information and
benefits and services, as you alluded to, 24/7. In New Jersey,
we are using AI enabled chat bots to answer resident questions
via the web.
We also support our call centers with AI, which helps staff
quickly write menu options and answers to questions in response
to public demand. That means the public is getting accurate
answers faster, and that allows our call center operators to
increase the resolution of calls now by 50 percent by human
operators.
We are able to create now a more accessible government as
well, with AI driven translation and summarization. My
wonderful students at Northeastern University and the AI for
Impact co-op program, where students are doing a six-month full
time paid internship working for governments on AI projects.
Students, for example, have been working with the
California nonprofit Innovate Public Schools to help families
of schoolchildren with disabilities translate and summarize the
child's individualized education program (IEP). For most
families, that is a 50 to 100 page portable document format
(PDF) of inscrutable legalese and government speak.
Students in this, as we call it, Apex International
Education Partners (AIEP) project team worked with a network of
parents now to design a tool that enables families to upload
their IEP and then securely and privately ask questions of the
documents, such as what are the accommodations to which my
child is entitled.
Because generative AI makes it so much faster to write
code, students built a fully functional product in just one
semester. AI is also helping us to support public workers in
their effort to do a better job serving residents. In New
Jersey, our innovation team engineers have integrated AI
copilot tools.
That is to say, tools that suggest to them which line of
code to write next, and that is enabling us to write code 55
percent faster, creating a wide array of digital benefits much
faster than we were able to do before. Governments are also
using AI to process data, as we have heard, more efficiently.
In New Jersey, where we design government operations with
residents rather than for them, we actively invite citizen
comment. That is to say, we often receive thousands of comment,
for example, from unemployment claimants telling us what their
questions are, what they would like to see next, what
information they need, and then an AI can help us parse those
comments in a way that we could not do with the current staff
that we have.
We can then, based on that understanding, put front and
center the information that people want, and as a result of
which, over the last year, we have brought down the time it
takes to apply for unemployment benefits by 48 minutes per
application. Governments are also turning to AI to improve how
we do public engagement more broadly.
I am working now with the Museum of Science to run a
national consultation to understand why it is, shockingly, that
only 33 percent of fourth graders were proficient readers last
year.
The free and open source tool, Policy Synth, that we
designed together with the nonprofit Citizens Foundation
automated the creation of thousands of research questions,
allowing us to synthesize then 150 root causes of the problem
of illiteracy.
Now at unlockingliteracy.ai, we are asking parents and
students and educators around the country, what do they think
is the real cause of the problem, and to tell us how we can do
things better.
To realize this vision of AI enabled public administration,
where residents do not have to wait on hold, do not have to go
to a government office, and do not have to, frankly, go from
website to websites searching for the information they need,
let me just end with a couple of recommendations.
I have a much longer list in my written submission. You can
support Federal efforts to implement AI systems to provide this
kind of round the clock information services, about policies,
about benefits, about services, again, 24/7 and in multiple
languages. We need to employ AI for more robust public
engagement--first to improve services. We can turn to AI to
help us synthesize comments in a human centered design process.
I would add, we can also engage the public using AI to talk
about how we are using AI and how we are governing the use of
AI in agencies. We need to intensify efforts to open government
data, ensuring that it is in machine readable formats, to
facilitate the training of better AI models, and, frankly, to
build and grow the American AI industry.
Finally, we need to build on the success of the AI
Leadership Training Act that came out of this Committee by
implementing free, accessible training not just for leadership,
but for all Federal employees, going beyond senior officials,
focusing not just on knowledge of what AI is, but on practical
applications in government services, ensuring that we collect
data and evaluate impact about how such upskilling translates
into better public service efficiency and better innovation. I
look forward to your questions. Thank you very much.
Chairman Peters. Thank you, Dr. Noveck. Ms. Pahlka,
numerous surveys suggest that satisfaction with government
services is a lot lower than it is in the private sector. My
question for you is, what is your assessment of the State of
Government service delivery today? What should we be striving
for?
Ms. Pahlka. It is lower, but it is getting better. I think
what we need to recognize is that now it must get better,
faster. It has to get better faster in part because the gap
between public and private sector is about to widen as private
sector actors adopt AI.
Also, because AI is going to complicate the landscape
within Federal agencies, and so there is a need to step up. I
think the way to get it better is both to support a lot of
current efforts that are already underway. My colleagues have
mentioned several of them, and there is a great foundation to
build on. But also, to look at new tactics, like policy
simplification, which I mentioned in my opening statement.
The goal is a good question. I think many people will say
that we need to be on par with the private sector. There are,
of course, differences between the public sector and the
private sector. I think the way I would put it is that the
level of service needs to recognize what is at stake for the
citizen.
Very often when you are engaging with the private sector,
you are getting your cable bill or getting a car, it may be
critical to you in that moment it but it is nowhere near as
critical if you are doing something like expecting a tax return
check or getting your unemployment insurance, when that may be
really life changing for you.
The level of service should be appropriate to how critical
it is to the human on the other end.
Chairman Peters. Thank you. This next question is going to
be actually all three witnesses. Ms. Blauer, I will start with
you and then we will work down the dais.
This Congress, I introduced the Improving Government
Services Act to encourage agencies to improve customer service
and also reduce backlogs, which some have tremendous backlogs,
unfortunately.
My question, though for each of you is that how can AI help
agencies provide quality outcomes for these key services but
also protect individual privacy, something that we are
concerned about and hear from constituents all the time,
related to their concerns.
Ms. Blauer, what are your suggestions?
Ms. Blauer. Thank you so much for the question. I am going
to take the question in two parts. The first one is how do we
make sure that AI actually aligns the outcomes that we want to
see, and that requires leadership.
We need leaders who can articulate goals for customer
service, that can create those expectations for what customers
should expect in interactions with the government, and we
should make sure that we build the habits within our government
to measure whether or not we are hitting those goals.
We need to make sure that if we have set a standard, that
we are holding to that standard. Then the way that we think
about AI and the way that we protect individual information is
that when we think about this culture of experimentation, AI
right now, we are in a research and design phase, and we need
to create a ring fence around these opportunities to experiment
with a technology that does not include utilization of
personally identifying information (PII).
We can get very far by using data that is de-identified,
that is safe, and that will not expose individual level data.
We can do that by thinking really strategically about those
frameworks that we set out in the bill, in the way that we
think about building out these opportunities for
experimentation.
Chairman Peters. Thank you. Dr. Noveck.
Ms. Noveck. Thank you. Let me also take this in two parts
then. The discussion that we have already begun to have and
that you have alluded to around using chat bots and the ability
to use AI to help answer people's questions 24/7 will help us
begin to reduce the backlog.
The ability to use AI to help us write better, clearer
instructions and directions, again, will help people get
customer service faster. The ability for us to use AI to bring
information together across agencies, across departments,
across those 10,000 regulations that you talked about, that
ability to now synthesize information, I think can help us to
create this more conversational ethos where we are responding
to people's questions and then reducing the need for them to
call in, to wait, etcetera.
But we also then need to speed up how we do that work, and
that is where some of these copilot techniques, this ability to
use AI to help us write code faster, really comes into play.
But I do want to respond to the second half, and as Beth
alluded to already, this ensuring that we are not using, and we
do not need to in most cases, use PII or sensitive information,
and how we do this, I think, is really crucial.
It is one of the guidelines that we have put out in our
policy in New Jersey as we have embraced and encouraged people
to responsibly use AI in order to improve customer service, but
we said, be very careful.
Do not type personally identifiable information into these
tools, not that of residents, not that of yourselves, in order
to ensure that we are protecting privacy, even at the same time
as we are improving customer service.
Chairman Peters. Thank you. Ms. Pahlka.
Ms. Pahlka. Wise words from my co-witnesses here. I support
everything that they just said and simply add that I guess from
the perspective of the pair mandates and directives with
enablement.
In other words, agencies want to do these things. They want
to improve government service delivery, and there are many
things that often stand in their way that make them take a long
time to do things, jump through a lot of hoops.
As you say, as you direct them to take these actions, also
ask them what can we remove to make this easier.
Chairman Peters. Thank you. Senator Carper, you are
recognized for your questions.
OPENING STATEMENT OF SENATOR CARPER
Senator Carper. Thank you, Mr. Chair. To our witnesses,
welcome. Is it Pahlka?
Ms. Pahlka. Yes.
Senator Carper. Pahlka, right, Noveck, and Blauer. Those
were my only questions. [Laughter.]
Ms. Pahlka. That was easy.
Senator Carper. Not really. I think everybody on this
Committee has been fortunate to hold a whole lot of different
jobs in our lives. I think of myself as a servant, as Governor
of Delaware. Maggie was Governor of her State as well. We got
several Governors on this panel.
But we focused a whole lot on customer service. I remember
I was right out of the Navy, moved to Delaware from California
and went to go to the Department of Motor Vehicles (DMV) in
Wilmington, Delaware to get my Volkswagen Karmann Ghia, with a
rebuilt engine, through inspection and it took forever to find
a place.
When we got there, I was convinced that the people who work
there hated me and anybody else who showed up for an
inspection. I said, boy, would it be great to have a chance to
fix that someday and change that.
Today, when people go into a DMV in Delaware, the first
thing they see is customer service, how can I help you. It used
to be that people hated the Division of Revenue, which is like
our IRS in Delaware.
We did have a lousy reputation for service. My last year as
Governor, the folks at the Division of Revenue won the customer
service award for best service of any nonprofit or business
entity in the State of Delaware. We did not have AI then.
Imagine how good we could be if we had AI then.
AI could be a good thing, a real good thing, or it could be
a real bad thing. Why I am excited about this hearing today to
hear what the upside might be. When Chuck Schumer, our leader,
told us a couple of months ago, we are going to have a series
of briefings on AI, I said to my colleagues, I said I could
barely spell AI.
Several of them said to me later, they could barely spell
AI as well. But I think I have almost got it down, at least the
spelling. Now to find out, how can we harness this to provide
better service. We have only three counties in Delaware. We
have a million people.
We have county directors in each county. We have folks who
do constituent service in each of our three county offices. We
measure every month, we send out a customer satisfaction survey
to the people that we have served through our county offices,
and people can evaluate our service excellent, good, fair, poor
or excellent. We do this every month.
For 22 years we are running--96 percent excellent or good.
About three percent fair, one percent poor. If folks evaluate
us, they come back and say that we were fair or poor, the
service was fair or poor, we call them.
We call them and see, how could we do it better. None of
them have mentioned AI to us yet in terms of doing better. But
I just lay that down as a predicate. This is an important part
of our jobs.
Folks think about what we do about here is we pass budgets
and we deal with--foreign international issues. We also focus
on a good customer service. This may be a way that we can
provide even better service.
Question, Ms. Pahlka, Ms. Blauer, now that I have your
names straight, what can Federal agencies do with the current
tools at their disposal to deploy AI technology more
efficiently in order to meet the evolving needs of the people
we serve?
How will this improve the public's perception and
satisfaction of government services? Ms. Pahlka.
Ms. Pahlka. I just want to honor your focus on AI--focus on
customer service in State of Delaware. I think that does so
much to contribute to people's trust and faith in government
and in democracy.
Thank you for sharing that. I think what you said about the
successes in customer service without AI actually speak to
something that each of us are trying to say, which is it is
really about the foundation and the basics. Ms. Blauer
mentioned leadership.
Dr. Noveck mentioned the infrastructure. We are not
necessarily going to succeed or fail because of AI. We are
going to succeed or fail because we have the right foundation
to build on. We do not quite yet have that entire foundation.
We need to keep building it. We need to build it faster.
There are all sorts of things that we should have done in
the past decade or two, to bring customer service from the
Federal Government into the Internet age and meet the
expectations of the American public as they have been able to
use services that become much more convenient for them.
You do not have to go stand in line anymore to deposit a
check, you use your phone to do it. Expectations are risen, and
we have to meet those rising expectations. Those things that we
should have done to move us effectively into the Internet era
they are unfinished business.
We still have to do them. We now have to do them quite
quickly so that we can be set up to be a great customer service
organization in the age of AI. Those things that are basic have
to do with our human capital strategy.
They have to do with procurement. They have to do with the
kind of leadership that Ms. Blauer talked about. I think we
agree, and I very much appreciate your comments and your
perspective.
Senator Carper. Thank you. Ms. Blauer, same question. Just
follow up. I think I saw you nodding your head as Ms. Pahlka
was speaking.
Ms. Blauer. Yes. Thank you for your very thoughtful
question. As someone who would always measure whether or not
what we were doing was effective in the State of Maryland by
looking at what was being said about the Motor Vehicle
Administration (MVA) on Twitter, I can say with great authority
that it was also a very big measure for us as to whether or not
the residents of our State were having their needs met.
I think the most effective implementation of AI is actually
not felt at all by the public. It is working in the background
to optimize an experience that is led with the good people that
are helping usher in people through a process, that are
connecting people to services that they are trying to access,
that are actually freed from a lot of the sort of bureaucratic
processes that have been labored by regulations, by frameworks,
by some artificial inflation of bureaucratic processes.
When we can skim those processes down and free up the
public sector workforce to actually focus on creating those
meaningful service delivery opportunities, that is when we are
going to see the kind of combination of the power of AI and the
ability for us to provide very responsible and very satisfying
customer experiences.
I think the most frustrating implementations that I have
seen to date have been where we are leading with a new chat bot
or a new interface that is hard, that is difficult for people
to navigate or to connect with, and where they feel even
further distance between themselves and the work that the
government is doing.
We have to think really strategically about how we achieve
both of those ends by focusing on what is that experience, what
do we want that experience to feel like in the moment, and then
how can we relieve those frontline workers of some of those
bureaucratic processes so that they can focus on that
experience and let the AI work in the background to optimize
it.
Senator Carper. Great. Thank you both for the responses.
Mr. Chair, thanks for holding this hearing. This is important
stuff.
Chairman Peters. Thank you, Senator Carper. Senator Hassan,
you are recognized for your questions.
OPENING STATEMENT OF SENATOR HASSAN
Senator Hassan. Thank you, Mr. Chair. thank you to our
witnesses for being here, not only for your testimony, but for
the work that you do. My questions are really going to be
follow-ons or expansions of, I think, the questions you have
heard, because there is a common focus among all of us about
how we can maximize the potential of AI, but also protect the
things that people care about in their interactions with
government in the process.
Ms. Pahlka, I will start with you. In your testimony, you
discussed how the Federal Government needs to build capacity
and competency to deploy AI in a way that meets the diverse
needs of each agency and improves service delivery. I am
interested in a little bit more of the specifics.
What specific ways can agency leadership foster greater
technological capacity and competency to deploy AI, while
balancing each agency's ultimate mission? In particular, what
can we do to develop AI-competent workforces?
Ms. Pahlka. That is such a great question and thank you for
asking it. I think that there are things that agencies need to
do, and I think there is also things that the center of
government needs to do to support them in doing it.
If you go into any agency today and you ask who is actually
doing the customer service, who is doing the delivery, who is,
say, creating the website that people are using, and I am sort
of speaking in an Internet era answer right now rather than the
AI answer, it is pretty much all contracted out and the number
of people in an agency who have a real pulse on the actual
mechanisms of delivery is very small.
You can go in and say, OK, there is so many teams I know
right now who are doing really incredible work where they are
building interfaces to government online for instance that
really work for people, and they are struggling. You go in and
say, great, this is amazing work, what do you need?
They say, well, we are a team of three, how many other
people are working on this? 300, 3,000. What are they doing?
They are writing reports. They are doing various kinds of
compliance. But we really need to go look at who are the people
doing this work. Do we have enough of them who have the right
competencies and capacities? Part of what I mean by a human
capital strategy.
There is a huge relook that we have to do on who we hire
and in what capacity in government and really change things to
be more oriented toward the capacities and competencies of
delivery. Then we also need to be able to hire those people. I
mentioned in my written testimony that it just takes too long
to hire people today. And you know, the how to make that a
shorter process is a very complicated question.
You see good movement around that right now with things
like direct hire authority for agencies for AI enabled
positions. Of course, it is much more than just experts in AI,
right. It is all of the people who do the data, for instance or
a wide range of AI enabling positions.
But we need to further ask, how can we speed the hiring of
those positions? I now know literally hundreds of people from
the private sector with exactly the competencies, capacities,
and values that we want who want to come into government.
These are the people we need. They are really interested in
a way that they were not as interested 10 years ago, but they
are languishing in nine month processes and then we are losing
them. Let us just look at who we need to hire and then figure
out how to hire them quickly.
Senator Hassan. Thank you. I think we could talk about this
a great deal more, but I have a question for each of the other
witnesses that I want to get to. To Dr. Noveck, we know the
public expects better digital customer service from their
government.
However, many Americans, including retirees, older
individuals on Medicare or Social Security, veterans, or those
who have just complex cases actually need to speak to a
customer service representative to address these issues.
The whole panel has touched on this a little, but what are
some of the ways that AI could supplement or complement the
work of Federal customer service representatives so that they
can focus on the people who actually need the human one-on-one
case management or solving?
Ms. Noveck. I think it is a great question. That is why in
New Jersey, we are really using AI to support our call center
workforce. We are not looking to offload people to phone trees
and to put you in an endless loop of pushing buttons.
Quite the opposite. What we want to do is by listening to
residents and understanding what it is that they need to know
and want to know. Looking at what they are searching for on the
web, what are the questions that they are asking. They call in
and then we can put that information front and center, first of
all.
Second of all, we can provide good responses to people on
the web, and that is going to offload some of the easy
questions such that they are not passing through to humans. If
we can get rid of your simple question about when is this open
or closed, or which form do I use, etcetera, and we are putting
that front and center, that means that fewer people are needing
to actually talk to a person, and as a result--so for example,
on just one service, by reducing by 15 percent the number of
queries coming through, we have increased by 50 percent the
number of people that were then able to serve with that phone
service.
But also for more vulnerable residents, whether they are
non-native English speakers, whether they are elderly,
etcetera, the more that we design things in a human centered
way that is talking to and listening to residents about what do
they need and what do they want, the more we can then take
advantage of AI to do things like provide multilingual
services, or for example, translate language, complex
government speaking legalese into something at a ninth grade
level or in simple terms.
Listening to residents using the data we have from what
they tell us to then improve what we are putting forward, then
means we are putting fewer things through to the call center
and having more responsive, more human, and more conversational
interactions.
Senator Hassan. Terrific. Thank you so much for that. Ms.
Blauer, one of the main benefits of AI is that it can make
sense of vast amounts of data.
In terms of customer service, and you have touched on this
a little bit, but this could help connect disparate pieces of
information from all across the government to streamline
benefit applications or answer regulatory compliance questions.
However, we cannot use AI to make these connections without
adequate data-sharing policies or data-sharing infrastructure.
We have talked a little bit about the privacy issues, which are
real, but what are the main barriers that prevent data sharing
across the government?
Ms. Blauer. Thank you so much for this question. It is one
that I spend a lot of time focused on. We have replicated the
silos of government in the systems where data lives in
government.
One of the actual promises and opportunities that AI
presents is that it is making it much easier to actually link
data sets together. Data sets that do not have the same types
of metadata that connect them.
We may identify addresses differently in one platform, we
may identify the name format differently in another, and when
you work at that level, you can start to link those data sets
together.
AI is actually very powerful in helping us do something
that used to take sometimes years to do, which is to link data
and to allow you to do that analysis. That is one of the things
that I am very excited about.
But interagency data sharing agreements are critical. We
need to clear the way for data to be able to be shared not just
across agencies within the Federal Government, but also between
the Federal Government, the State Governments, and city
governments.
We need inter-agency--we need inter-governmental data
sharing agreements that will free up our abilities to have data
move through these complex webs of systems so that we can
actually design the most impactful services. Right now, that
does not exist.
Senator Hassan. I will just note, and I am over time, and I
note that, but as we talk about this capacity of data sharing,
we also, to your previous answers, know that there are ways we
can do that while protecting people's privacy, because
obviously my constituents, I am sure all of our constituents,
care a great deal about their sensitive personal information
and their privacy.
But there are ways in these interagency agreements that
look at the metadata to protect privacy too, right?
Ms. Blauer. Absolutely. It is one of the best ways for us
to protect privacy is to have these structures built around it.
Senator Hassan. Thank you very much. Thank you, Mr. Chair.
Chairman Peters. Thank you, Senator Hassan. Ms. Pahlka,
your testimony mentions that our current Federal procurement
framework is simply too rigid and that it prevents us from
timely obtaining some technological capabilities like AI, as
well as cloud computing.
But this issue is actually a major priority for this
Committee, and we are in the process of looking at some reforms
to the procurement process going forward.
My question for you is, is how can we change the framework
to better obtain and use these essential capabilities like AI
and cloud computing, and any other types of reforms that you
think are essential for us to consider?
Ms. Pahlka. Thank you so much for bringing up procurement.
I know there are a bunch of procurement nerds listening who
will be very excited that this topic is being spoken of in this
chamber.
There are a bunch of procurement nerds because it is such a
complex issue. The Federal Acquisition Regulation (FAR) is, as
you well know, 2,400 pages, and the number of things that could
be done to it to make it easier for contracting officers and
people in charge of delivery in government are numerous.
I think in the time that we have today, what I would say
is, it has to be simpler. Any time something has that many
different provisions, you are going to have people fighting
over interpretations of things. You are going to have people
saying, we cannot use this, we have to apply this particular
provision of it.
Ultimately, if it is a little bit simpler, I think people
will be able to get the outcomes that they desire from a
particular procurement more quickly. I think speed and
flexibility are the two things that you should be designing for
if you think about any kind of procurement reform.
There is a lot else, I think, to talk about there, but
ultimately, what you want to end up with something that is not
2,400 pages. The great thing about this particular moment in
time is that it is very hard to understand what those 2,400
pages actually mean. Now, there are many people are in
government who really do know it backwards and forwards, but AI
can actually help us suggest some simplifications to that.
It is part of the opportunity of this particular moment.
Chairman Peters. Thank you, Dr. Noveck and Ms. Pahlka. We
are going to ask both of you to respond to this question.
Competition is absolutely critical to ensuring that the Federal
Government uses the latest technology and gets good value for
its purchasers as well.
My question to you is, what are some of the measures that
we can take in Federal procurement to promote competition with
these advanced technologies? Dr. Noveck, I will start with you,
and then Ms. Pahlka.
Ms. Noveck. I think Ms. Pahlka has already begun to touch
on this question, the need for greater simplicity and greater
streamlining to enable smaller vendors to be able to compete in
this space.
One needs to not simply have mastered the General Service
Administration (GSA) schedules which should not be the sole
threshold to being able to do business with the Federal
Government. Streamlining and simplicity is going to be
incredibly important in this space if we are going to enable
smaller people to compete.
I think, and GSA has been making strides on this with their
wonderful guide on agile procurement that they have put out,
also really trying to push forward what is known as agile
procurement.
That is to say, not trying to anticipate every single
requirement that we are going to have five years ahead of time
in advance. These technologies are so new, our uses and
experiences with them are so new, and designing tools correctly
means actually doing it in little increments, testing what is
working, and improving.
The more that we are streamlining by going smaller in the
scope, being more agile, and ensuring that we are actually then
refining as we go, I think the more that we are going to both
open things up to procurement, but also ensure that the tools
that we are procuring translate into better services and better
customer experience.
Where we make the mistake is that, again, I am going to
anticipate everything that we need five, 10 years down the
road, again, on technologies that are changing now really by
the day, if not by the week, and there are new capabilities
emerging all the time.
We can use again AI to help us in conceptualizing and
drafting some of those requirements, but we need to be able to
make more agile procurement the norm rather than the exception.
Chairman Peters. Thank you. Ms. Pahlka.
Ms. Pahlka. I think Dr. Noveck said 90 percent of it. If I
can just plus 1,000 that. All I would add to it is, there is
some controversy over the notion of open source AI. I think in
the kinds of contexts we are talking about, the dangers are
much less, and the benefits can be pretty big.
AI could have the effect of real consolidation of power
because it is so hard to create these models. The investment
that OpenAI put into Chat Generative Pre-trained Transformer
(ChatGPT) is staggering.
They have a lot of power because of that. The more that we
can actually look at models that are more distributed, more
democratized, and that is going to mean oftentimes open source,
and make sure we can consider those when we are trying to find
a solution in government, the more we will be I think promoting
competition broadly in technology.
Chairman Peters. Good. Ms. Blauer, this question is for
you. We know that a lot of innovation occurs at the local level
and State Governments. My question for you is how are Mayors
and local officials approaching the procurement of AI systems
in their cities, in their communities, in their States? Maybe
give us some specific examples for the record.
How can we learn from their work specifically? What are
some lessons that we should take to heart here?
Ms. Blauer. Thanks so much for the question. For this one,
I think it is important for us to look at local examples
because there is some really promising practices, particularly
around outcomes based procurement.
AI has the potential to deliver value across a whole host
of outcome areas that we really care about, and local
governments are leading with outcomes based procurement.
Instead of scoping what the technology need is, I think it is
prudent for us to think about how can we partner with AI
developers and with AI platforms to actually achieve some of
the most pressing outcomes that we have been trying to crack
for generations?
In places like Seattle, where they are using outcomes based
budgeting model in order to advance how they fund homeless
services or cities have been using outcomes based procurements
in order for them to instead of having the provision of how you
deploy light-emitting diode (LED) light systems across your
city system, have the light bulb providers buy into your energy
efficiency goals.
Outcomes based procurements have the ability for you to
actually scope AI, actually scope technology interventions in a
way where you are partnering with the organizations for the
actual improved customer experience, for the actual reduction
in greenhouse gas emissions, for the actual system safety and
preservation outcomes you want to see across your
transportation structures.
If we can pivot out of the silo of procurement, being
totally focused on the technology and get the technologists to
partner with you through the procurement vehicles on the
outcomes, it will be transformational. We are seeing it in
cities. I think it is absolutely possible also at the Federal
level.
Chairman Peters. Wonderful. The next question is going to
be for all three of you. I will start with Dr. Noveck and then
Ms. Pahlka, and then Ms. Blauer, you can wrap it up for this
question. Studies have indicated that roughly 45 percent of
chief data officers believe that they do not have the
responsibilities for AI at their individual agencies.
Yet, quality data is absolutely key to quality outcomes, as
we all three have already discussed and we are well aware of.
My question for you is, how can the Federal Government ensure
that Federal Chief Data Officers (CDOs) are involved in the
efforts to actually improve the delivery of services with AI?
Dr. Noveck.
Ms. Noveck. I think the question suggests the answer, which
is ensuring that they do. As we have all talked about, data is
of course integral to training AI models. Without data to
analyze, the AI becomes fairly useless.
What we are trying to do is actually use data to ensure
that we are proactively delivering services in the way that Ms.
Blauer already talked about. I think having Chief Data Officers
at the table is really essential and mandating that their role
includes, again, the proactive publication and availability of
data, as is already required by law, and ensuring that we do
more of that.
That is going to require then more training, I think, for
chief data officers and for data professionals in government
around AI to talk about where those collaborations can occur.
Ensuring that agencies do create those cross-functional
teams and that cross-agency collaboration, as well as within
agencies and then across agencies to ensure that these new AI
officers, data officers, policy people, everywhere that we can
bring disparate disciplines together to talk about how we can
better serve residents, I think the better job that we are
going to do.
I think the question is really vital because it suggests
the answer, which is we need the data people to be at the
table, and we need to invest in the proactive publication and
availability of data is really integral.
Chairman Peters. Thank you. Ms. Pahlka.
Ms. Pahlka. My colleagues at the table here are national
experts on this, and I would like to cede my time to Ms. Blauer
there. Both of these folks have so much experience in this
topic.
Chairman Peters. Great. Ms. Blauer, you have the final say
on this question.
Ms. Pahlka. All right. Thank you to both of you.
I was not a classically trained data person when I became
essentially the de facto Chief Data Officer of the State of
Maryland. What I was someone who understood what it meant to
work in government.
I connected to the people of the front lines of government,
and I was able to usher in a transformation of data use in
government because I could relate to the challenges that the
people in the organizations were trying to solve. I knew what
it was like to be a probation officer, to sort of sit under the
crushing weight of that bureaucracy.
When we think about how we attract CDOs into our
government, if we are only hiring technologists into those
roles and not people that have the softer skills, that can be
deferential to subject matter expertise that exists in your
government, that can actually connect and translate from the
technology to the policymaking, to the legislative agenda
setting, to the budget making, to all of the things that we
know are the levers of government that are going to create
change, then we are doing a disservice both to the position of
CDO and to government itself. CDOs need to be able to have
those rounded expertise so that they can, like I said, connect
to the people who are delivering for your government and also
for those that are leading change in your government.
Where we have seen failures, both at the local and State
level, is when we place a CDO in an information technology (IT)
agency and their role is sort of ordained just to be the person
that hounds people to move data or to get data express for
other use.
Where we have seen great success is where they are
empowered from a leadership level, and that they are materially
part of the big substantive change that government is embarking
on.
Chairman Peters. Right. Thank you. Senator Butler, you are
recognized for your questions.
OPENING STATEMENT OF SENATOR BUTLER
Senator Butler. Thank you, Chair Peters. Thank you to all
of you for participating. So great to see a Californian on the
panel.
For that reason, little hometown love, I would love to sort
of start with you, Ms. Pahlka. We are one of the most diverse
States in the country, representing 40 million Californians and
a vast array of racial diversity, ethnic diversity, and
language diversity.
We, given our size and scope, our diversity that makes us
such a great State, the challenges of making government work
for California residents is a challenge that you have taken on
behalf of Governor Newsom and our Employment Development
Department (EDD) team.
I am curious about the ways in which these technologies can
be deployed to further accessibility, as well as just getting
people what they need on an everyday basis. The pain of getting
a driver's license.
The pain of getting a passport, or getting your call
returned. Talk with me a little bit, share with the Committee a
little bit about how we can embrace all that makes us great,
our diversity, and all of the things that really make our
country the melting pot that it is, and deploy technology in
order to make our government, again, more accessible for those
whose first language may not be English, and make sure that it
works in a way that people can get exactly what they need when
they need it.
Ms. Pahlka. That is that is an inspiring question, thank
you. I am glad that it came from our Senator from California.
We have mentioned earlier that translation is one of the things
AI is fantastic at.
When I worked on the task force for unemployment insurance
during the pandemic, translation of the whole process, not just
the application form people would use online--of course, we had
a problem with the call center, but if you were able to call,
were you able to get through and talk to somebody in the
language that you knew so that you could actually get the help
you need to get your unemployment insurance check?
Of course, we are at a moment in time when there are tools
available that could make that a lot easier. I have spoken
before about the need to streamline procurement so that tools
like that could be procured more easily. I also think it is
really important to recognize that when people call, they do
not just want to be able to talk to somebody in their language,
they actually need an answer.
It is wonderful if someone can pick up the phone, but much
better if they can say, your check is coming. In order to do
that, I do think we need to simplify the policies and processes
and regulations that a program like unemployment insurance is
staggeringly complex for something that really is not that
complicated.
We are going to take your pay, your previous pay and some
portion of it, and you are going to get that when you are out
of work. We have made it into 10,000 pages of regulation. I
mentioned how people working at the EDD in California were
still learning the ropes after 17 years.
Of course, it is hard to get an answer under that
incredible burden of complexity, and we happened to be at a
moment in time when AI could help us reduce that complexity. It
cannot do the work for us, but it can help us understand what
is in those 10,000 pages, suggest simplifications, and then
lawmakers could work to make that a program that is a little
bit easier to administer.
When I worked on unemployment insurance, and I hope that
speaks to your question, not just the particular example, but
we had a Legislature who passed a law right when we were trying
to get through that backlog that was designed to help people
who did not speak English.
As much as I am a fan of the need to do that and feel very
connected to the need to serve those folks, it was an example,
I felt, of adding a mandate when what was needed was
enablement.
One of the things we could have done at that time to
subtract rather than add, would have been to remove the
requirement that call center agents had to get permission
before they could use the translation service that was
currently available.
Instead, we said, here are new mandates for translation of
the website, which distracted, frankly from the ability of the
EDD at that moment in time to work through the backlog. The
thing they needed to do the most at that time was get checks to
what turned out to be 1.3 million people.
Sometimes we need to recognize that we have to have
priorities in order to serve people. There were many people in
that 1.3 million claims backlog who were underserved,
represented the diversity of California, and they actually did
not need at that moment a different way to read through this
web form, which was already so complicated.
The language in it was so complicated in English, that
translating it into another language was not going to help.
What they needed was their check. Sometimes we need to have
better communication between agencies and legislators to
understand what is the thing at this moment in time that will
enable the right outcome and set some priorities and have the
translation stuff come later.
I hope the translation stuff does come later, but I also
hope we use the tools of AI to do it, instead of doing it in
the mode that we have been doing it over the past 10 years.
Senator Butler. Thank you.
Ms. Pahlka. Thank you.
Senator Butler. Very responsive to the question, and a
continued incredible example for California. My last question,
I want to quickly move from maybe customer service to thinking
about helping you help us to think about how to better create
the guardrails, enable the tools to also supplement direct
services, and challenges that we have as a Nation.
Across the country, there is both a mental health crisis as
well as a mental health worker crisis. I was reading, Los
Angeles being one of the largest mental health providers in the
country, that they also have more than 1,800 vacancies in
mental health provision.
But have been reading a lot, I have, about chat bots that
are being able to be deployed to offer initial support, where
we may not have a worker at the ready to support the kind of
one on one interaction that is required during mental health
counseling and services.
Nationally we have is, at the Federal level, a national
health care worker task force, understanding that we have a
shortage around the country and a challenge that we have to
address.
Can you talk more, either of you, about the deployment of
artificial intelligence, the usage of chat bots in a way to
support the existing workers and their work, even as we have
these sort of challenges of attracting more and creating more
opportunities for workers in the space.
Ms. Noveck. It is such a wonderful question and I want to
combine it with some of our focus also on government customer
service here. I think one of the exciting things about these
technologies is that they are what is sometimes referred to now
as multimodal. That is to say, they talk as well as they type.
That means we have conversational agents, a tool I am
thinking of like Pie, which was designed by the person who
actually created the United Kingdom (UKs) largest helpline for
Muslim youth, precisely in order to offer that responsiveness,
that sense of talking to a person, which can really provide a
lot of psychological support for people in the first instance.
We are seeing this ability not only to give people chat
bots in the sense of typed responses to questions, whether it
is for residents or whether it is for employees, but also
spoken responses to questions to provide that added level of
customer service.
I think that is hugely important, whether it is in the
mental health space, which is outside my area of expertise, or
whether it is in the government customer support area. This
ability now to provide people with that kind of responsiveness,
I think is a hugely important and we are going to see the use
of these tools only grow.
Of course, we need those guardrails to ensure that we are
confident about what the information is that is being provided
to people, that we are not taking advantage of people in a
vulnerable state.
That is why doing things in a human centered way, whether
that is with residents or whether that is also with government
workers and government employees, talking to them about what do
they need and what will actually help them is really crucial.
Last point here is, what is exciting about a lot of these tools
is you can also train them to respond on a specific corpus of
information.
It is not the world of ChatGPT and the open web that we
know where it might pull from anywhere on the Internet. But
rather the idea now, for example, that my students are working
with the Massachusetts Department of Transportation (MassDOT)
to say, let us upload their manuals, their procedures, their
policies, and then provide a tool so that new engineers coming
into the agency can ask a question like, what is the rule on
this or what is the policy about that, and to pull the answers
only from those documents.
That helps provide an added level of safety and security,
to be sure that we know what the answers that people are going
to get. Provides that conversational element. Again, it
provides that opportunity, though, to make it a much easier
than reading 1,000 documents and having to pore through them,
enabling those new employees that we are bringing in not to
burn out quite so fast, but instead to be excited about the
opportunity to do public service.
Chairman Peters. Thank you, Senator Butler. Dr. Noveck, New
Jersey is one of the first States to develop guidance on how
employees should use AI. If you would, for the Committee's
benefit, kind of describe that guidance. More importantly,
perhaps, how is it actually working in practice?
Ms. Noveck. Thank you. I am very excited, and Governor
Murphy gave his State of the State speech yesterday, in which
the topic of use of AI, responsible and ethical use of AI
featured very front and center.
I am very proud that we are one of the first States to
actually have clear guidance, which says we want people to go
out and try these things. It is not to say that we will not
revise that guidance, and we will not upgrade it, and we will
not change it as we learn better what these tools do and as
they evolve, but the fact that we are going out and saying we
want people to actually try this to the end of improving how we
deliver services to residents.
We are coupling, however, that encouragement to try things
with broad scale availability of training. I lead a partnership
of multiple States, including California, called Innovate US,
run by public servants for public servants, where we are
providing free training in responsible AI and other technology
use to government workers.
I might say my co-panelists here have been or will be soon,
speakers who have been helping us to train government workers
at no cost to public servants. I think that is really
important. Right now, OPM has a mandate whereby they have to
use fee recovery to charge for training that they provide to
government workers.
We want to ensure that the training we are providing is
actually free, so that there is no disincentive to using it.
But coming back to the policy question, we are telling people,
go out and try these tools, but ensure that when you are trying
them, please do not enter any personally identifiable
information.
Please factcheck anything that comes out of these tools. We
do not want to see those instances like the lawyer who filed
the brief that still said in it, written by ChatGPT. We want to
be sure that you are actually fact checking what is coming out
of these things in the way that you would if you asked an
intern to write something.
We want to also be sure that people are taking advantage of
these tools in ways that are actually supplemental to and
aiding, as Senator Butler referred to, actually aiding
government workers in how they do their work, not supplanting
them.
Ensuring that we are using AI as an auxiliary to workers is
very important. But again, undergirding all of that is
training, training, training, and training.
Chairman Peters. OK. Ms. Pahlka, your book, Recoding
America, outlines the disconnect that can occur between
policymakers, policy implementers, and the general public. My
question for you is, how can government actually prioritize the
customer when it comes to service delivery in the life cycle?
Ms. Pahlka. I think this is so critical that we have a
fundamental shift in what we value in order to put the customer
at the center.
I know this is a very high level answer, but we have to
value delivery as highly as we value policy, both here on the
Hill and in the agencies, and to the points earlier, in States
and in cities.
We are very focused here on the words that we write, and we
forget that the people on the other end do not look at those
words. They go on a website and try to get unemployment
insurance. They are trying to get their driver's license or God
forbid, they are in our criminal legal system and they are
experiencing incredible burdens.
They see delivery. As I said, our human capital system in
government broadly, not just in Federal Government, is not set
up to value that. We are set up to write words and hire
contractors.
I believe that putting the customer at the center of
government is going to involve a pretty high level shift to
thinking about what is it that government actually does, what
is our function, and valuing what it looks like to the person
at the end this incredible waterfall hierarchy, as much as we
do the business of actually writing policy.
Chairman Peters. Thank you for that. I would certainly like
to thank all of our witnesses for your time and for being here
today, and I am certainly grateful to your contributions to
this important discussion.
Delivering services to the American people is a core
function for all levels of government. Long lines, slow
customer service, complicated applications for Federal services
hurt the public satisfaction, as well as the trust that they
have in their government.
As I think we have heard very clearly today, artificial
intelligence can help make service delivery more efficient,
more effective, and accessible. Our witnesses emphasized that
strong data infrastructure, a trained workforce, and proper
privacy safeguards can ensure that the government is ready for
these new AI powered tools.
We also know that developing the right procurement
frameworks and recruiting the best talent is going to be needed
for wide and successful AI adoption.
As Chair of this Committee, I will continue to find ways to
build a more efficient and more effective government for all
Americans, and your testimony today will help inform the
Committee's future legislative activities, which are
forthcoming.
We look forward to your thoughts on that legislation as it
is being developed, as well as our oversight actions that we
take regularly throughout the Federal Government.
The record for this hearing will remain open for 15 days
until 5.00 p.m. on January 25, 2024, for the submission of
statements, as well as questions for the record. This hearing
is now adjourned.
[Whereupon, at 11:16 a.m., the hearing was adjourned.]
A P P E N D I X
----------
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
[all]