[Senate Hearing 118-666]
[From the U.S. Government Publishing Office]
S. Hrg. 118-666
THE NEW INVISIBLE HAND?
THE IMPACT OF ALGORITHMS ON
COMPETITION AND CONSUMER RIGHTS
=======================================================================
HEARING
before the
SUBCOMMITTEE ON COMPETITION POLICY,
ANTITRUST, AND CONSUMER RIGHTS
of the
COMMITTEE ON THE JUDICIARY
UNITED STATES SENATE
ONE HUNDRED EIGHTEENTH CONGRESS
FIRST SESSION
__________
DECEMBER 13, 2023
__________
Serial No. J-118-47
__________
Printed for the use of the Committee on the Judiciary
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
www.judiciary.senate.gov
www.govinfo.gov
______
U.S. GOVERNMENT PUBLISHING OFFICE
60-435 WASHINGTON : 2026
COMMITTEE ON THE JUDICIARY
RICHARD J. DURBIN, Illinois, Chair
SHELDON WHITEHOUSE, Rhode Island LINDSEY O. GRAHAM, South Carolina,
AMY KLOBUCHAR, Minnesota Ranking Member
CHRISTOPHER A. COONS, Delaware CHARLES E. GRASSLEY, Iowa
RICHARD BLUMENTHAL, Connecticut JOHN CORNYN, Texas
MAZIE K. HIRONO, Hawaii MICHAEL S. LEE, Utah
CORY A. BOOKER, New Jersey TED CRUZ, Texas
ALEX PADILLA, California JOSH HAWLEY, Missouri
JON OSSOFF, Georgia TOM COTTON, Arkansas
PETER WELCH, Vermont JOHN KENNEDY, Louisiana
LAPHONZA BUTLER, California THOM TILLIS, North Carolina
MARSHA BLACKBURN, Tennessee
Joseph Zogby, Chief Counsel and Staff Director
Katherine Nikas, Republican Chief Counsel and Staff Director
Competition Policy, Antitrust, and Consumer Rights
AMY KLOBUCHAR, Minnesota, Chair
SHELDON WHITEHOUSE, Rhode Island MICHAEL S. LEE, Utah, Ranking
CHRISTOPHER A. COONS, Delaware Member
RICHARD BLUMENTHAL, Connecticut CHARLES E. GRASSLEY, Iowa
MAZIE K. HIRONO, Hawaii JOSH HAWLEY, Missouri
CORY A. BOOKER, New Jersey TOM COTTON, Arkansas
PETER WELCH, Vermont THOM TILLIS, North Carolina
MARSHA BLACKBURN, Tennessee
Keagan Buchanan, Democratic Chief Counsel
Wendy Baig, Republican Chief Counsel
C O N T E N T S
----------
OPENING STATEMENTS
Page
Klobuchar, Hon. Amy.............................................. 1
Lee, Hon. Michael S.............................................. 4
WITNESSES
Alford, Roger P.................................................. 14
Prepared statement........................................... 33
Responses to written questions............................... 39
Baer, Hon. Bill.................................................. 7
Prepared statement........................................... 46
Responses to written questions............................... 52
Epstein, Robert, Ph.D............................................ 10
Prepared statement, pp. 1-8.................................. 54
Responses to written questions............................... 62
Hewitt, Damon T.................................................. 9
Prepared statement, pp. 1-14................................. 71
Responses to written questions............................... 85
West, Sarah Myers, Ph.D.......................................... 12
Prepared statement........................................... 98
Responses to written questions............................... 107
APPENDIX
Items submitted for the record................................... 109
THE NEW INVISIBLE HAND?
THE IMPACT OF ALGORITHMS ON
COMPETITION AND CONSUMER RIGHTS
----------
WEDNESDAY, DECEMBER 13, 2023
United States Senate,
Subcommittee on Competition Policy,
Antitrust, and Consumer Rights,
Committee on the Judiciary,
Washington, DC.
The Subcommittee met, pursuant to notice, at 3:02 p.m., in
Room 226, Dirksen Senate Office Building, Hon. Amy Klobuchar,
Chair of the Subcommittee, presiding.
Present: Senators Klobuchar [presiding], Hirono, Welch,
Lee, and Hawley.
OPENING STATEMENT OF HON. AMY KLOBUCHAR,
A U.S. SENATOR FROM THE STATE OF MINNESOTA
Chair Klobuchar. All right. Thank you everyone for being
here today. A few things on our agenda in Washington, but as we
await word, like, you know, the white smoke coming from the
Vatican to see if we have any kind of progress on an agreement,
it does give us this pause in time to focus on a really
important issue for our country.
And so, I call to order the hearing of the Subcommittee on
Competition Policy, Antitrust, and Consumer Rights. This
hearing is titled, ``The New Invisible Hand? The Impact of
Algorithms on Competition and Consumer Rights.''
I'd like to welcome our witnesses. I got to say hi to all
of you, and also, of course, Ranking Member Lee. Mike and I
work together really well, and I appreciate his work, and his
staff's work in working with our staff to plan this hearing.
In today's digital economy, we are constantly interacting
with algorithms, whether we're shopping online, paying rent,
booking a flight, hailing a ride, buying insurance, or looking
for a new app to download. Often, the products we see, the
prices we pay, or the services we receive are determined by an
algorithm.
And I think even for some of our more astute visitors today
to this hearing, and people who may be joining us, including
Senator Hirono who just joined us, I don't think when we look
at our phones and we see things pop up, that even when we know
these algorithms are playing a role, do we really think that we
are being manipulated by them?
I think we think we're smarter than that, but actually, it
is happening all the time. And so, the focus of this hearing is
really what's going on, and how it is affecting competition and
prices that consumers are facing.
In many ways, the services that are built with algorithms
can improve our lives. We know that algorithms help make sense
of vast amounts of information to surface and order search
results, identify patterns, and improve products.
But algorithms also have the potential to create or
exacerbate competition problems. Even when used with the best
intentions, algorithms that are built with data that reflect
past patterns of bias or discrimination can perpetuate those
patterns into the future. And algorithms also are responsible
for pushing harmful content to kids. Today's hearing will
explore these issues to help Congress understand how to
safeguard competition and protect consumers.
In October, this Subcommittee heard testimony about the
potential for competitors to use a third-party algorithm to fix
prices in housing markets, driving up the cost of rent and
leading to higher vacancy rates.
Corporate landlords have figured out that it's better for
them and their bottom lines to delegate their pricing decisions
to algorithms to push rent so high that some rental properties
sit empty rather than offering housing that people can actually
afford. And they're doing this not the old-fashioned way by
talking to each other, which would be clearly--some of this
would be illegal depending on the course of conduct, but
they're doing it by delegating it to an algorithm.
As I said in October, landlords shouldn't be sharing prices
in an effort to boost prices. Instead, they should be competing
on price and on quality.
Price fixing and other forms of collusion are illegal under
the antitrust laws. When competitors that should be competing
decide instead to delegate their independent pricing decisions
to an algorithm, the result is little more than a sophisticated
cartel hiding in code. Whether the conspiracy takes place in a
server room or a boardroom shouldn't matter under the antitrust
laws, but it isn't clear whether our current antitrust laws are
sufficient to stop that practice.
Most courts require proof of an explicit agreement to fix
prices before condemning the conduct. But that agreement may
not exist when a computer is coordinating prices instead of
humans. It may even be possible for independent price-setting
algorithms programmed to maximize profits to learn to collude
on price without any human intervention.
Price fixing isn't, of course, the only concern. Dominant
platforms can and do use algorithms to preference their own
products and services, and bury those of their competitors.
The United States needs laws that are up to date, and as
sophisticated as the monopolists we are trying to reign in.
That's why Senator Grassley and I have introduced the American
Innovation and Choice Online Act--you may have heard about it
on TV. There was a lot of ads run against it. Hundreds of
millions--which is co-sponsored by several Members of this
Committee, and is specifically targeted to address this type of
anticompetitive conduct by Big Tech platforms.
Dominant firms may also use algorithms to protect
themselves from competition. For example, the FTC recently sued
Amazon for monopolizing e-commerce markets, in part, by
surveilling prices across the internet and punishing
independent businesses that sell on Amazon's marketplace that
offer lower prices anywhere else. This conduct deters price
competitions, hurts consumers, and gives Amazon free reign to
charge exorbitant fees to small businesses that uses the
platform.
The public understands that Big Tech tactics hurt consumers
and deter innovation. Just this week, a jury in California
found Google's App Store policies--allowing it to take up to 30
percent of a developer's revenue--violate antitrust laws. The
Department of Justice has also sued Google for monopolizing the
search engine market. The recent trials showed that Google's
exclusive contracts, which guarantee its search engine default
status, starved competitors of the data necessary to develop
competing algorithmic search tools.
And because algorithms run on data, the more the companies
rely on algorithms, the more they will be incentivized to track
consumers to collect data raising privacy concerns. This is why
I've long supported comprehensive Federal privacy legislation
so that consumers can control how their personal data is being
used, including the right to opt out.
Moreover, the output of an algorithm is only as good as the
data fed into it. ``Garbage in, garbage out,'' as they say.
This means that if an algorithm is trained on data that
reflects racial, economic, or other biases, it can lead to bias
or even discriminatory results. For example, after being
trained on a company's past hiring data, an algorithmic tool
used to screen job applicants weighed two factors most heavily
when making recommendations about whom to interview: the name
Jared, and the interest in lacrosse. True fact.
Finally, the proliferation of artificial intelligence and
machine learning has a potential to supercharge these issues
and to create new ones. That's why earlier this week, Ranking
Member Lee and I sent a letter to the Department of Justice and
the FTC asking for more information about their enforcement
efforts related to algorithms and additional tools they need to
do that work.
While many of these issues raise tough questions, I think
one thing is clear. We need vigilant enforcement of our
antitrust laws. We, of course, passed our Merger Filing Fee
Modernization Act on a, I think, 88 vote in the U.S. Senate, to
move up when that money was going to be collected to allow our
antitrust enforcers to do their jobs, as well as passing
Senator Lee and my Venue bill at the same time.
So that is our part, but we know there's so much more we
have to do, and that's why we're having this hearing today. I
consider this the beginning of this discussion. We know there
will be AI legislation coming down the pipe. We also know that
many other Members in the Senate are interested in this, as
well. So with that, I turn it over to Senator Lee.
OPENING STATEMENT OF HON. MICHAEL S. LEE,
A U.S. SENATOR FROM THE STATE OF UTAH
Senator Lee. Thanks so much, Madam Chair. And that was a
good day getting that passed. Wasn't it? We like that bill.
As our economy continues to modernize, algorithms have
drawn increased attention as a tool for calculating prices, and
calculating costs, and doing all sorts of things that we would
never have dreamed just a few short years ago could be made by
reference to or with the assistance of an algorithm.
Although algorithms are efficient, they can be used by
businesses to collude in setting prices, and to reduce the
quality of products in a way that can perpetuate bias.
Algorithms are neither good nor bad, but rather, like a
company's size, it depends on the context. It depends on how
they are used.
There's nothing inherently sinister about an algorithm.
They can do only what they've been designed to do, and for that
reason, we can fairly say that the devil is in the details. It
depends on the quality of the data that's being fed into them,
and what businesses do with the algorithms that may be
problematic, especially if the business in question has
monopoly power or engages in collusive price setting.
Today's hearing concerns algorithms and their potential
impact on competition. Now, it can be used in a number of ways
that may impact competition, in ways that are either pro-
competitive, pro-competition, or anti-competition.
First, algorithms are able to help determine prices. Using
an algorithm is a much more efficient, and in some cases fast,
perhaps accurate way to set prices rather than engaging in a
more traditional, price-setting method.
For example, consumers experience this when using, just to
cite one example, popular rideshare apps. Algorithmic pricing
could be very problematic if used improperly. When executives
use algorithms to collude and to set prices, this can impede
competition and undermine the natural balancing of the market.
Algorithms present a new and easier way to engage in this
behavior, and to engage in it in a way that makes its detection
a little more difficult.
Second, algorithms impact competition by affecting the
quality of products. Facebook, Google, TikTok, Twitter, and
others have been a sale for the content that their algorithms
sometimes produce, or alternatively restrict, or throttle, or
advance, or silence, or partially mute.
These algorithms are likely written with the intent to fuel
engagement and keep users on the platform to consume more and
more content in a period of time, or to keep the user on the
app for a longer period of time. It's the clickbait model,
basically, but on a trillion-dollar societywide basis.
Why fuel engagement? Why use the clickbait model, or
something else, algorithmic, or otherwise, to fuel engagement?
Well, more time spent on the platform means more digital ads,
and that, in turn, as those eyeballs are scanning more and more
ads, means a lot more money for the platform.
Now, Google and Facebook are a digital advertising duopoly
rife with all sorts of conflicts of interest. Those conflicts
enable what basically amounts to insider trading, exacerbating
the monopoly rents that Google and Facebook already collect,
and further incentivizing those companies to keep people on
their platforms for as long as possible.
Breaking up the ad tech stack is an important first step
toward reigning in the power of Big Tech platforms, and making
sure that they don't use algorithms in an anticompetitive,
predatory, harmful manner. Additionally, Big Tech companies
have exploited algorithms to exert their market position in
order to push content, perpetuate bias, and control expression,
speech, and even elections.
This even applies to what our children are viewing online.
We can all agree, I think, that this is bad when it
specifically happens to children, especially when Big Tech
pushes inappropriate content to underage viewers, whether
teenagers, or in many cases, much smaller children. They don't
yet have the maturity, the judgment, the self-control to make
wise decisions about the content that they consume. And in some
cases, they can be really affirmatively harmed by it.
All of us have the instinct to protect children. I believe
that the effective enforcement of antitrust laws in this area
could result in increased competition and new alternatives. And
with those new alternatives, you could have some companies
doing a better job than others at making sure that the content
provided to children wouldn't be harmful to them.
Now, this is also true of course, when it comes to adults.
Neither the Government nor Big Tech should be telling people
what to believe or how to vote, or at least should proceed with
the knowledge that if they tell people what to believe or how
to vote, they may have backlash in the marketplace, at least if
there is a marketplace with robust competition to be had.
Because if there isn't, they're going to continue to do
whatever they want, relying on their monopoly or duopoly
status, relying on the absence of meaningful competition,
knowing full well that they can simultaneously advance their
economic, you know, their financial, and their political, or
other interests.
It's nothing short of alarming that Google, precisely
because of its monopoly and its lack of algorithmic
transparency that impairs in the nature of the use of
algorithms, can control the information that a user sees to the
extent that it does with no repercussions in the market.
According to Dr. Epstein's research, Google is feeding
people biased information quite deliberately that it furthers
the progressive political agenda and reminds liberals to vote
on election day while intentionally failing to give
conservatives those same reminders.
Now, Google significantly ended the partisan vote reminder
and began sending the vote reminder to all users, but only
after they were caught doing so in the last election. That is
significant. That is itself, perhaps, an indication that
there's a breakdown in the market. That, perhaps, is an
indication that the antitrust laws can have something
meaningful to say here--and that they should.
Now, I want to be clear, every American, including any
business, has whatever right to pedal whatever sort of
political message that that company, or that business, or that
person may want, whether it's for a political party represented
by an elephant, or one represented by a donkey, or for that
matter, a giraffe. They can do that if they want. It's part of
the First Amendment. They enjoy that First Amendment
protection.
But if they're doing that, especially doing that in an
audience that doesn't consist overwhelmingly of Republicans, or
of Democrats, or whoever they're trying to pitch these messages
to, they would be doing so at the risk of alienating, in our
country, about half the population anytime they side with one
political party over the other. The fact that they did so, so
boldly, so effectively, so unapologetically in this instance,
to me, suggests there's something terribly, terribly wrong. We
know that there is, in that marketplace. It's one of the
reasons why we need to address the stack.
We have an obligation to our constituents to protect them
from this significant threat and to defend competition itself--
not for the purpose of any political party, or cause, or
message, but because consumers themselves deserve better than
this. They deserve what happens naturally when there is
competition in the marketplace. I look forward to our hearing
today and to hearing from each of you.
Chair Klobuchar. Very good. Thank you, Senator Lee.
I'm now going to introduce the witnesses.
Bill Baer is a visiting fellow in governance studies at The
Brookings Institute. He previously served as Assistant Attorney
General for the DOJ Antitrust Division. Previously, he was a
partner at Arnold & Porter, and headed up the firm's antitrust
practice group. He was also the director of the FTC's Bureau of
Competition.
Damon Hewitt is the president and executive director of the
Lawyers' Committee on Civil Rights Under Law. Previously, he
was an attorney at the NAACP Legal and Educational Defense
Fund, and served as executive director of the Executives of
Alliance for Boys and Men of Color.
Dr. Robert Epstein is a senior research psychologist and
co-founder of the American Institute for Behavioral Research
and Technology. He is the former editor-in-chief of Psychology
Today, and is the founder and director emeritus of the
Cambridge Center for Behavioral Studies. He has taught at
several universities.
Sarah Meyers West is the managing director of the AI Now
Institute, where she researches topics at the intersection of
artificial intelligence, society, and policy. She is also a
visiting scientist at the Network Science Institute at
Northeastern University. She was formerly a senior advisor on
AI at the Federal Trade Commission.
Last, but not least, Roger Alford is a professor of law at
the University of Notre Dame Law School, where he focuses on
topics at the intersection of international business and the
law. Previously, he served as Deputy Assistant AG for
International Affairs at the DOJ Antitrust Division.
All right. So now we're going to swear you guys in. You
never know, people and algorithms--you've got to swear.
I will now swear in the witnesses. If the witnesses could
stand--you are. Thank you.
[Witnesses are sworn in.]
Chair Klobuchar. Thank you, and you may be seated. So, I'll
start by recognizing you, Mr. Baer. Thanks for being here.
STATEMENT OF HON. BILL BAER, VISITING
FELLOW IN GOVERNANCE STUDIES, THE
BROOKINGS INSTITUTION, WASHINGTON, DC
Mr. Baer. Chair Klobuchar, Ranking Member Lee, Senator
Hirono, it's good to be back. And thank you for the opportunity
to address, as you both have noted, one of the many challenges
we face in harnessing the power and maximizing the potential of
AI.
The growing use of pricing algorithm presents one such
challenge. I'm no expert in AI, believe me. But from the
vantage point of a long-term antitrust enforcer, I am concerned
that misuse of this tool is growing and puts consumers at risk
of paying supracompetitive prices for all sorts of goods and
services.
Sometimes competitors agree directly on a pricing
algorithm. We brought a case when I was at DOJ that addressed
that kind of competitor-to-competitor agreement. But often
companies outsource to a third-party AI price fixing by using
the same vendor to collect the data on supply and on demand,
and recommend pricing or output behaviors that facilitate price
coordination.
Your hearing last October on rental housing markets showed
how property management companies allegedly were using third-
party vendors to collect competitively sensitive pricing
information from competitors, feed that data through
sophisticated algorithms, and recommend unit-by-unit prices so
landlords could charge higher rent.
Similar antitrust challenges are pending involving hotel
operators in Las Vegas and in Atlantic City, using that same
kind of third-party pricing algorithm to suggest profit
maximizing strategies to the hotels that increase margins and
limit a consumer's ability to bargain hunt online.
These AI-facilitated hub-and-spoke conspiracies extend
further up the food chain, too, literally. This fall, DOJ and
six attorney generals, including those from Utah and Minnesota,
charged Agri Stats with operating an information exchange that
obtained sensitive price and output information from turkey,
chicken, and pork producers, analyzed the data, and provided
feedback that allowed these competitors to reduce output and
increase price with confidence that their competitors would do
the same.
Again, the basic behavior's not new, but the use of pricing
algorithm seems to make coordination both easier and quicker.
The good news is that these hub-and-spoke conspiracies have
traditionally been held to violate the antitrust laws. The bad
news is that algorithmic collusion using third parties seems to
be on the increase. Detection is not easy, and AI makes success
more likely,
But my big worry is whether our current antitrust
jurisprudence can handle fact patterns where the machines learn
how to collude with little or no human involvement. An unlawful
agreement under Section 1 of the Sherman Act requires a showing
of a meeting of minds between rivals, a conscious commitment to
a common scheme.
At the same time, the courts have held that consciously
parallel behavior, sometimes called ``tacit collusion,'' is not
enough to violate Section 1. Courts require plus factors, proof
that something more than parallel unilateral action was afoot.
But what if competitors individually develop pricing algorithms
that set profit maximization as the goal, and machine learning
leads to pricing outcomes that result in widespread
oligopolistic pricing in markets where discounting previously
was the norm?
I think about the 1983 ``WarGames'' movie. There, a young
computer nerd, played by Matthew Broderick, hacks into the
super computer controlling the U.S. military's nuclear arsenal,
and activates a game called Global Thermonuclear War. He thinks
it's just a game, but the computer, the Whopper, treats it as
the real thing and takes actions that trigger escalating
responses from the then-Soviet Union.
We were on the brink of thermonuclear war until Broderick
directs the computer to play tic-tac-toe. In seconds, the
whopper--this is the 1980s--runs every series of possible
moves, learns that the game is unwinnable, and stops the
escalation.
I worry about an algorithmic pricing scenario where
companies individually write code that simply instructs the
machine to profit maximize. The machine gathers public pricing
info from competitors, learns in nanoseconds that price
competition is eluding losing strategy, and it stops
discounting, and stabilizes prices.
Does the industry-wide implementation of pricing algorithms
that predictably lead to such a result, even without direct
communication between competitors, constitute and illegal
agreement under Section 1? That's a little unclear under
current law.
So, what can be done? My written statement makes a couple
suggestions.
First, enforcers need to determine whether my doomsday
scenario is actually a real-world concern. One way to do that
is in merger investigations to make sure they use second-
request to see how the pricing algorithms employed by merging
parties react to each other's pricing decisions.
In addition, the FTC could use its authority in under
Section 6(b) of the FTC Act to do a deep dive into selected
industries and learn, better understand the prevalence and
real-world impact of pricing by algorithm. The FTC potentially
could employ its unfair methods of competition authority under
Section 5 to challenge the use of AI that results in
anticompetitive outcomes, even if the evidence is not
sufficient to show an agreement in violation of Section 1. In
addition, companies need to monitor the pricing behavior of
their machines just as they're responsible for the actions of
employees that lead to anticompetitive results.
Finally, Congress, I think, should consider legislation
that addresses the growing risks that competition posed by
algorithmic pricing, either as part of this broader effort to
set guardrails for the use of AI, or antitrust-specific
legislation that holds competitor responsible for the knowing
use of pricing algorithms that that they know or should have
known results in tacit collusion and reduced competition. Thank
you.
[The prepared statement of Mr. Baer appears as a submission
for the record.]
Chair Klobuchar. Very good. Thank you. Mr. Hewitt--where we
challenge you to cite a 1983 movie.
[Laughter.]
Chair Klobuchar. Go ahead.
Mr. Hewitt. I've seen that movie, and it's a very good
film.
STATEMENT OF DAMON T. HEWITT, PRESIDENT AND
CHIEF EXECUTIVE OFFICER, LAWYERS' COMMITTEE
FOR CIVIL RIGHTS UNDER LAW, WASHINGTON, DC
Mr. Hewitt. Good afternoon, Chair Klobuchar, Ranking Member
Lee, Senators Hirono and Welch, and others in absentia. My name
is Damon Hewitt, president, and executive director of the
Lawyers' Committee for Civil Rights Under Law, where our
mission is to ensure that Black people and other communities of
color have the voice opportunity and power to ensure that all
the promises of our democracy are made real, and not just words
on paper.
As technology grows in this country and plays a growing
role, our work has expanded to include what we call ``digital
justice.'' We're looking at the intersection of racial justice,
privacy technology, focusing on the ways that predatory data
practices, discriminatory algorithms, and other online harms
disproportionately impact consumers of color.
Across our economy, algorithms are used to make decisions
about all aspects of our lives, determining who can run a
house, who can get a loan, who can get a deal, and
consequentially who cannot.
One of the greatest civil rights challenges of our
generation is to ensure that our new data-driven economy, does
not replicate or amplify existing discrimination. As Ranking
Member Lee said earlier in his remarks, AI is not the problem,
but it's like putting discrimination and the problems that do
exist on steroids if we don't watch out.
We have to ensure that the technology well serves all of
us. Because algorithmic technologies are built using societal
data that reflect generations of discriminatory practices, such
as redlining and segregation, they often reinforce past
patterns of discrimination.
For example, scoring algorithms used by auto insurers
result in higher rates and fewer options for residents of
majority Black neighborhoods who don't have other options.
Retail websites have been found to charge different prices for
the same products based upon consumer demographics. Mortgage
approval algorithms are more likely to reject Black applicants
even when they have less debt than white applicants. Things
that absolutely make no sense.
The harms of algorithmic discrimination are already denying
millions of Americans equal opportunity in our economy. Instead
of aiding consumers, AI tools often create gross distortions in
the marketplace, reflecting exclusion rather than fairness for
consumers and creating closed doors in the virtual world that
have discriminatory effects in real life.
In my written testimony, I catalog dozens of examples
showing how algorithmic systems harm Black communities and
other communities of color. Left unchecked, these harmful
impacts will continue to grow as AI becomes ingrained in every
aspect of our daily lives.
Senators, my message today is simple: The tools of the
future are locking us into the mistakes of the past. But it
does not have to be that way. Congress must act quickly to
enact legislation that can ensure algorithmic systems are safe,
effective, and fair for all consumers.
Now, we've faced similar challenges in this country before
to unlock opportunity. Congress passed the Civil Rights Act of
1964 to prohibit segregation and interstate commerce alongside
other legislation to address discrimination in employment,
housing, and other critical aspects of the lived experiences of
daily lives of Americans in the marketplace.
Today, the mass accumulation of personal data and the use
of algorithmic technologies call for return to our civil rights
legacy to ensure that everyone has an equal opportunity in the
new digital marketplace and fair access to the information
goods and services it enables. We believe at the Lawyers'
Committee that legislation coming down the pike should reflect
at least six key principles.
First and foremost, AI regulations should protect American
civil rights by including an explicit anti-discrimination
provision that prohibits algorithmic discrimination.
Second, AI tools should be evaluated and assessed for
discrimination and bias, both before and after deployment. As
Mr. Bayer mentioned earlier, these machines learn and they
evolve. So, we can't just test one time. We need ongoing
review, assessment, and audit.
Third, developers and deployers of AI should have a duty of
care requiring that products that they offer are safe and
effective. As with any consumer product, the developers and
deployers should be liable if these products are not safe and
effective.
Fourth, AI regulation should include transparency and
explainability requirements so that consumers know when, how,
and why AI is being used in a way that impacts them.
Fifth, data protection requirements should limit the use of
personal data and safeguard consumer privacy. Data protection
is necessary to ensure that personal information is kept secure
and not used against consumers unfairly.
And finally, AI regulation should establish robust
oversight and enforcement of mechanisms, empowering Federal
officials with adequate authority and resources in providing a
private right of action to remedy algorithmic harms in the
private sector.
We believe this type of legislation will help to limit
discriminatory impacts of AI, and will be good for all
consumers. Thank you. I look forward to your questions.
[The prepared statement of Mr. Hewitt appears as a
submission for the record.]
Chair Klobuchar. Thank you very much, Mr. Hewitt. Dr.
Epstein.
Dr. Epstein. I'm going to turn on the screen. Oh great.
STATEMENT OF ROBERT EPSTEIN, PH.D., SENIOR RESEARCH
PSYCHOLOGIST, AMERICAN INSTITUTE FOR BEHAVIORAL
RESEARCH AND TECHNOLOGY, VISTA, CALIFORNIA
Dr. Epstein. Ms. Klobuchar, who's one of my very favorite
of America's leaders because she's one of the few Members of
Congress who has never taken a dime from Google. And Mr. Lee,
who helped me and my team to get Google to back off in a
national election. Thank you, sir, immensely. It's my first
chance ever to thank you.
Thank you, and other Members of the Committee, I'm here to
tell you about an existential threat to our country that is so
well hidden--I warn you, by the way, some of what might turn up
on the screen in the next minute or two might be disturbing. I
apologize in advance, but it's necessary that you see it.
It's so well hidden that you might know nothing about it.
It's a threat posed by Big Tech monopolies, eerily predicted by
President Eisenhower in 1961.
In 2016, Google alone shifted more than 2.6 million votes
to Hillary Clinton--whom I supported--using subliminal
techniques I had been studying and quantifying since 2013.
Four days later, a leaked video showed Google's leaders
devastated by Trump's win, telling their employees they would
not allow Trump to win the Presidency again. They would
guarantee his defeat through their, quote, ``great strength and
resources and reach.''
They made good on this promise in 2020, and in 2022, as I
explained recently in The Epoch Times, they stopped the ``red
wave'' cold. I lean left, but I don't think a private monopoly,
one with no accountability to the public, should be able to
pick our Nation's leaders.
Who knows how these secretive companies will lean next
year, after all. After that all-hands meeting, Google perfected
at least a dozen new methods of subliminal control that I have
now been studying, using rigorous scientific methods for more
than a decade.
To shift votes, we know from leaked emails, that Google
relies on what they call ``ephemeral experiences''----
[Large viewing screen is displayed.]
Dr. Epstein [continuing]. Fleeting content such as search
results, search suggestions, and UpNext videos on YouTube,
content that impacts undecided voters and then disappears,
leaving no paper trail.
Since 2016, my dedicated team has been building
increasingly more sophisticated monitoring systems that
preserve and analyze ephemeral content. This is Google's worst
nightmare, because it means we are surveilling them just as
they surveil us and our children 24 hours a day.
In other words, we are giving you, our Nation's leaders,
the ammunition you need to finally hold Google and other tech
companies accountable to the public.
[Large viewing screen is displayed.]
Dr. Epstein. Our research, which we publish in prestigious
peer-reviewed journals, allows us to measure the power Big Tech
has to shift votes, while our monitoring systems let us see
whether these manipulations are actually being used. In one
case so far, we shared our data with Senators Lee, Johnson, and
Cruz. They sent a strong letter to the CEO of Google, which
ceased some of its election manipulations that very day.
It turned off the political bias in its search engine, and
stopped sending partisan ``Go vote'' reminders on its homepage.
Through our monitoring, we detected those changes the moment
they were made. We preserve ephemeral content through the
computers of a representative sample of real voters. One must
monitor through the computers of real people because Big Tech
sends out personalized content. To see what they're sending
people, you must look over the shoulders of real people, just
like the Nielsen company does with television viewers.
Algorithms, my friends, I'm sorry to say, cannot be
regulated because they are inherently opaque even to the
programmers. And even if you did regulate, we still need to
monitor to make sure there is compliance. We started small, but
have deployed bigger systems with each election.
In 2022, we preserved more than 2.5 million ephemeral
experiences through the computers of more than 2,000 voters in
10 swing States. We are now building the world's first
nationwide digital shield, and we just released a public
dashboard. It's at americasdigitalshield.com.
[Large viewing screen is displayed.]
Dr. Epstein. It's what you're seeing on your screen, and it
shows our cumulative findings in real time as we are receiving
the data. We are collecting and displaying data 24 hours a day
through the computers of more than 13,000 voters in all 50
States, and so far, we have court-admissible data in 15 States.
We have thus far--excuse me, the extreme political bias we
are seeing in content being sent to voters along with the
highly sexualized and violent content being sent to America's
kids confirm my worst fears: The technological elite, as
Eisenhower called them, are now in control of our democracy,
and they're systematically indoctrinating our children.
If we can secure funding to expand our system so we have
court-admissible data in all 50 States, the tech companies will
back down in 2024. And even if they don't, we will have
incontrovertible evidence of election rigging on a massive
scale.
If no monitoring system is in place, Google alone will be
able to shift between 6.4 and 25.5 million votes in the 2024
Presidential election, leaving no paper trail and making a
mockery of the free and fair election. Thank you for your
attention and for protecting our great Nation from threats,
both foreign, and, I hope, domestic.
[The prepared statement of Dr. Epstein appears as a
submission for the record.]
Chair Klobuchar. Thank you. Dr. West, thank you.
STATEMENT OF SARAH MYERS WEST, PH.D., MANAGING DIRECTOR, AI NOW
INSTITUTE, NEW YORK, NEW YORK
Dr. West. Thank you, Chair Klobuchar, Ranking Member Lee,
and to the Members of the Committee for inviting me to testify
on this important set of issues. I deeply appreciate this
Subcommittee's ongoing attention to the role of algorithmic
systems in shaping the economy at large, often in ways that
harm consumers while benefiting centralized actors.
In the most basic sense, algorithms are simply instructions
that computational systems follow. They frequently involve
using statistical techniques and applying them to very large
amounts of data to arrive at a decision. For example,
processing personal data to tailor a price to the maximum the
system predicts a buyer will be willing to pay.
Now, while many of the underlying techniques have existed
for decades, such systems are supercharged by the surveillance
business model that's been promulgated by the tech industry.
This drove the creation of technologies that collect detailed
and often intimate information about us as we move about our
lives. This business model produced information and power
asymmetries that have profound societal effects.
Now, firms that are positioned at key choke points in the
economy are best positioned to leverage data in a manner that
is harmful both to competition and to consumers. This creates
extended potential for collusion in markets where this
otherwise would not be possible.
This is most visible when we see systems go awry, like when
the price of hand sanitizers spiraled at the start of the
pandemic. But frequently it takes shape in more invisible ways
that are harder to identify from the outside. And my fellow
witnesses have identified several examples in real estate and
car insurance.
This works the other way around, too, though. For example,
Dr. Veena Dubal has studied the use of algorithmic models to
determine the wages of rideshare drivers, which effectively has
served to drive down their take-home pay across the board.
What this means is that people are being placed in a dual
bind. On the one side, they experienced personalized pricing
that extracts as much money as possible from them, and on the
other algorithmic, wage discrimination that drives down their
pay.
This is one reason why it's particularly crucial that we
address concentration among the firms producing and deploying
algorithmic systems.
Another is to mitigate the risk of creating single points
of failure that could have ripple effects throughout the
economy. Over-reliance on the same algorithmic model can
present imminent risks to financial stability. And these risks
are not only posed to financial markets, but, you know, the
more broadly a single system is used, the greater the
consequences could be across housing, credit, payments,
transportation.
We all know that algorithmic systems are not infallible. To
the contrary, where investigators have looked under the hood,
they often find that the underlying data on which these systems
are being trained is flawed, and this leads to widespread
errors in decision-making. Think about the frustration of
having an insurance claim be denied just because the evaluator
on the other side was in a rush, or hadn't had their coffee yet
this morning.
We can think of algorithmic systems as taking each of those
individual decisions and replicating them at massive scale,
often with little to no scrutiny before they've been
commercially deployed. We've granted a staggering amount of
power to the scant few firms that have the data and
computational infrastructure required to develop and deploy
algorithmic systems, and we've given up consumer sovereignty in
the process.
The opacity of algorithmic systems also makes effective
enforcement all the more difficult. By denying the public
information about how algorithmic systems are affecting their
lives, each of us lacks the information that we would need to
be able to understand whether a decision's accurate, to whether
we have means to seek remedy or how to push back. And this is
why effective regulatory intervention that foregrounds bright-
line measures is particularly needed in this moment.
To conclude, there are three paths forward that I'd like to
particularly highlight.
One is, we need to use existing enforcement mechanisms to
ensure strong oversight of algorithmic systems, and robustly
enforce the agencies that have the existing authority.
We already have a range of laws in place that can be
applied to anticompetitive and harmful uses of AI and
algorithmic systems, and we need to make sure that our
enforcers have what they need to meet that challenge.
Second, we need specific bright-line rules to curb AI use
where it has demonstrated harms to consumers and competition.
As an example, the passage of a Federal data privacy law,
including a strong data minimization mandate, should be an
urgent priority here.
Last, we need legislation that will tackle the market
structure and gatekeeper power of dominant digital platforms,
which hold an unprecedented amount of economic and political
power. Thank you very much for your time, and I look forward to
your questions.
[The prepared statement of Dr. West appears as a submission
for the record.]
Chair Klobuchar. Thank you. Professor Alford.
STATEMENT OF ROGER P. ALFORD, PROFESSOR
OF LAW, UNIVERSITY OF NOTRE DAME LAW SCHOOL,
SOUTH BEND, INDIANA
Professor Alford. Chairwoman Klobuchar, Ranking Member Lee,
and Members of the Committee, thank you for inviting me to
testify today to discuss the impact of algorithms on
competition.
Let me begin by just saying that this is a matter that is
both a global concern and a matter of bipartisan concern. When
I was at the DOJ, I traveled the world, and it was rare to go
to an antitrust conference where there was not discussion about
the monopolization abuse of Big Tech companies.
Likewise, here in the United States, the Trump
administration, and the Biden administration, and almost every
State attorney general have filed landmark antitrust litigation
against Big Tech.
Both Senator Hawley and Senator Klobuchar have written
books expressing grave concern about Big Tech abuse of power.
Senators across the political spectrum have expressed this
concern.
In my written statement, I provided quotes of different
Senators from this Committee, and, unless you look at the
footnotes, it's hard to know which side they're on, in terms of
their expression of concern. Much of the concern relates to how
Big Tech has harnessed algorithms to exert market power like
never before in history.
Despite the obvious benefits of algorithms, which we can
all concede, their emergence creates enormous risks for the
abuse of monopoly power, and is a facilitating factor in
collusion between competitors. The algorithms themselves
recognize that they pose a risk to competition.
When I typed in the query, ``How do algorithms harm
competition? '' ChatGPT unabashedly confessed to me that
algorithms can harm competition in several ways, and then it
offered a half dozen different ways that it, algorithms, can be
harmful. It then concluded with a warning to lawmakers that the
fast-paced nature of technology often outpaces regulatory
measures, which I think this Committee should take as a
challenge.
With respect to price fixing, we've already talked about
different examples. Most recently in the RealPage case, the
Department of Justice has declared that weather firms
effectuate a price fixing scheme through a software algorithm
or through a human-to-human interaction should be of no legal
significance.
The RealPage example is but one that you have already had a
hearing about. Other examples in price fixing, we know from
2016 that there was a criminal investigation and prosecution of
sellers on Amazon Marketplace that led to a conviction with
respect to price fixing through algorithms.
And then, also just most recently with the JetBlue-Spirit
merger, information signaling through, so-called flashing is
happening where you're signaling to competitors what your
pricing behavior should be, and basically encouraging your
competitors to change your pricing behavior. All of those are
examples have different types of price fixing through
algorithms.
According to the DOJ Justice Surrendering and Freedom of
Action, and agreeing to abide by the will of a trade
association--which by the way I was involved in, in the recent
real estate price fixing. I as an expert witness in which $1.8
billion was awarded against the National Association of
Realtors--we know that following the trade association rules
should be viewed as and frowned upon.
The same is true with respect to algorithms. Concerted
action by an algorithm and reliance on a joint algorithm
generates price collusion.
With respect to monopoly abuse, there's so much to be said.
I encourage you to look at my written comments, but let me just
make a few quick comments about monopoly abuse through
algorithms.
In the Federal Trade Commission's complaint against Amazon,
Amazon algorithms detected if a seller offers a lower price on
any other online store, anywhere on the internet, and it would
punish those sellers on Amazon by disqualifying the seller from
utilizing the ``Add to Cart'' or the ``Buy Now'' button.
I don't even know how to buy anything on Amazon without the
``Add to Cart'' or the ``Buy Now'' button, but if you do have
any price competition anywhere, then you basically will be
taken off of that opportunity. And in Amazon's words when they
do that, the sales of those sellers tank.
With respect to Google search litigation, I think we've all
been riveted by the recent litigation that's happened here in
DC against Google on search. We know that Google denies
competitors the opportunity to scale. That was the essential
argument what was happening with respect to those revenue
sharing agreements. Google wants consumers to reach the
conclusion that their competitors are inferior. They use a less
polite word than that, and it will spend billions to control
upstream inputs to deny its competitors the opportunity to
improve.
How it does so is simple. Search algorithms require query
and click data to train the algorithms to improve search
quality results. And in 2021, Google paid over $18 billion to
Apple--or 36 percent of its mobile annual ad revenue--to be the
default searching engine of Safari, and we know what happened
with Bing in terms of their search capabilities. Their mobile
search capabilities are far inferior to their desktop.
Finally, with respect to Google ad tech litigation, we know
that Google uses algorithms to rig auctions in the digital
advertising marketplace. Because of Google's involvement and
dominance on the buy side, and the sell side, and the
exchanges, and in the middle, it has information advantages and
uses those advantages to exploit it to its own benefit.
My time has run out. Let me just conclude with a
recognition that both Senator Lee and Senator Klobuchar had won
great success last year, and that was the State Antitrust Venue
Act. And I just want to thank you for that. We finally are back
in Texas after 2 years of languishing in New York because of
the work that you've done, and that's been a significant impact
on the litigation that we've been pursuing against Google.
And then finally, the legislation that was attempted last
year, the AMERICA Act of S. 1073, I think that that is a great
solution. It's attempting to do much of what the litigation is
trying to do, and I commend you for that legislation as well.
Thank you.
[The prepared statement of Professor Alford appears as a
submission for the record.]
Chair Klobuchar. Thank you very much. I'm going to have
Mazie Hirono take my place. She's been diligently here to go
first, so thank you.
Senator Hirono. Kind of you, Madam Chair. Thank you very
much. So, listening to all of you, we know that algorithms are
being used in many different ways, and it is very hard to keep
track at the--I feel very--how shall I say? At a loss, as we
have to navigate the waters that are before us.
So, one consideration would be, perhaps, that we should
have maybe at this point a dedicated digital platform
regulator, this for Mr. Baer, and so maybe a competition
promoting regulator who could act in this space the way the FCC
did, and what is it--Carterfone back in 1968.
So, do you think we should be considering a new dedicated
digital platform regulator? Or would resources be better spent
bulking up our current antitrust enforcers? Do you have
anything to say about that?
Mr. Baer. I always have things to say in just about
everything, whether they're thoughtful or not. I do think that
serious consideration ought to be given to a digital regulator.
Whether that's a better idea than simply enhancing the FTC's
authority to do it, it seems to me an open question.
One of the cynical observations I'd make is, we've seen
companies like Facebook, Meta, Google, others say, just tell us
what the guardrails are. Give us some perspective guidance, you
know. And there was a huge Facebook PR campaign a couple years
ago about that. But the minute anybody puts something on the
table, suddenly it's ``Katy, bar the door,'' and we can't do
that.
So, I do think looking, though, for some prospective
guidance so that companies that want to behave, are able to
train their employees to behave, is worth doing. Antitrust
enforcement, other than mergers, is always after the fact. And
it takes a long time for the enforcement to catch up with the
conduct.
So, in this area, given how fast paced it is, as a number
of my colleagues have observed, trying to find a way to get out
front makes a lot of sense.
Senator Hirono. I think that is the concern that so many of
us have, that things are moving so quickly that whatever
regulatory framework we can put in place is already outdated by
the time we figure out how to regulate in this space.
So, Mr. Hewitt, you gave a lot of examples of how there can
be all kinds of discrimination happening with the use of
algorithms. And I am worried about the potential, more than
potential for making discrimination into algorithms, especially
if the algorithms are going to be used for things like lending.
And you already mentioned that, you know, in the space of
housing discrimination, there are all kinds of discriminatory
decisions that can be made and that people are not even aware
that it is happening. But it's through the use of algorithms.
So Fair Housing lending--there are Government testers, for
example, or shoppers who apply for housing and loans in order
to root out potential discrimination. How important is the
testing, to this kind of testing, to uncovering discrimination
in the housing and lending context where algorithms may be at
play?
Mr. Hewitt. Well, thank you, Senator Hirono. Testing is a
classic, you know, time-honored, tried-and-true technique that
helps to build an evidentiary basis in all sorts of civil
rights litigation and enforcement with respect to public
accommodations in general. I believe the Department of Justice
used testing online to effectuate its charges and settlement
with respect to Meta or Facebook.
I would also say, however, that as important as testing is,
we would need millions of testers to keep pace and keep up.
Which is why we need also legislation that compliments the
testing, legislation--regulation that is a prophylactic that
requires beforehand testing.
Not testing only in the marketplace, but testing
beforehand, almost akin to an environmental impact assessment.
We wouldn't disturb the environment or an ecosystem before
having some analysis of the impacts. And so I think that would
be a perfect complement to testing, even though testing, as you
indicated, remains critical and important.
Senator Hirono. And I think that--doesn't the Supreme Court
have a case that could set aside the ability of nonprofits and
others to engage in testing?
Mr. Hewitt. If you were referring to the----
Senator Hirono. I think there's a case.
Mr. Hewitt. The Atkinson hotels case.
Senator Hirono. So, we have to worry about that. And I know
that you notice some of the parameters that you would like us
to consider for legis--regulatory legislation. So, I hope we
can come up with something in this space, even if it's fast
moving and complicated. Thank you, Madam Chair.
Chair Klobuchar. Thank you. Senator Lee.
[Pause.]
Chair Klobuchar. They're like an Olympic commentary.
They're saying, ``Well, he's preparing now. He's coming down to
the floor.''
Senator Lee. Thanks so much, Madam Chair. I really enjoyed
your opening statements. Thank you for that.
Professor Alford, I want to start with you, if that's all
right. In 2024, both the DOJ and the State attorneys general
are going to have trials regarding Google's abuse of its
monopoly power, specifically in the ad tech market. Can you
discuss with us just a little bit about how legislation in the
ad tech market might be part of the solution in addition to
litigation?
Professor Alford. Sure. Yes. So, as you said, in 2024, we
anticipate that the Department of Justice ad tech case, as well
as the--we can anticipate, as I said, in 2024, the DOJ ad tech
case, as well as the State AG ad tech case, led by Texas, will
go to trial.
But, as we've seen already from the past few years, there's
extraordinarily hurt--extraordinary hurdles associated with
those two cases. Extreme delay, complete denial of any sort of
wrongdoing whatsoever, the exercise of privilege abuse that
Judge Donato recently called was a frontal assault on the
administration of justice.
And so, we know that the--and even if there is success, the
likelihood of a full structural remedy that would actually
correct the market in the way that it should, is a best case
scenario. So you're--the legislation----
Senator Lee. That's compounded by some of the dynamics that
we're describing----
Professor Alford. Exactly.
Senator Lee [continuing]. The inherent difficulty in
understanding what's happening within the algorithm.
Professor Alford. Exactly, exactly. So, the legislation
that you proposed, S. 1073, the AMERICA Act, goes to the heart
of the concerns in the DOJ litigation and in the State AG
litigation, which is to recognize the conflicts of interest, to
have one company or a set of large companies have a control
over the buy side, control over the sell side, and control over
the exchange in the middle, and basically use that information
to advantage itself and to harm its own clients is just
fundamentally problematic.
It's not the kind of thing that we see in other markets.
And so, the bill that you've proposed is essentially attempting
to get at that issue, and particularly for the very largest
ones.
With respect to the middle tier companies, your legislation
tries to do something, which is essentially provide more
transparency obligations to act in the best interests of their
clients rather than their own interests. So, it's a very
salutatory development, and I hope that there's hearings and
votes on that legislation.
Senator Lee. So, in other words, would it be fair to say
that--well, existing authorities--enforcement and litigation
under existing authorities may well be capable of addressing
the problem in this particular environment with algorithms and
what you can use them to do in the marketplace may take so long
to get to the result that you want, that it may become
obsolete, and therefore not ideally suitable to deliver the win
for competition and therefore the consumer that you need.
Professor Alford. Yes. And also, I should add that, you
know, the complaints are over hundreds of pages, and you can
only try so many of the core issues in those cases. And so the
legislation can, I think, get at more problems than simply what
the litigation will be able to get to.
Senator Lee. Now, it can be a little bit challenging to
digest the true scale of these digital markets and to
comprehend the vast impact that they have on our society and
our internet economy. Can you offer any insights in that
regard, help us to understand that?
Professor Alford. Sure. So, you know, I've been working on
the State AGs case against Google for years now, and there's a
lot of different things one could highlight.
But essentially, you know, if you control the information
and you control the inventory that is provided to the
exchanges, you can do a lot of things through algorithms that
no other competitor can do.
You can sequence when the bidding happens so that you get
the first bids. You can peek at the information that is being
offered in bid on other exchanges, and then you can adjust your
pricing behavior so that you will win the bid simply by
offering a penny less.
So basically, you'll starve the other exchanges from
competition just by peaking insider trading--I guess you could
call it, and then using that, you can offer it advantages that
are only available if you go to their exchange, like real time
bidding.
And so, you can do things like uniform pricing behavior
where there might be that sellers want to go with another
exchange, and they'll have different ceilings and different
floors, but Google will impose uniform pricing behaviors so
that's impossible to do that. So the whole range of different
things that are happening with respect to the abuse of this
monopoly power.
Senator Lee. Now, one could argue that the same things in
an environment where there is robust competition, those same
things could maximize consumer welfare by bringing, you know,
bringing up price, bringing down price, and increased quality
all at the same time. Because if it operates in an efficient
marketplace, it's make--makes an efficient marketplace in which
there is robust competition even more efficient.
Can you talk a little bit about how artificial intelligence
itself might be made to thwart that? What are some of the
competitive risks of Big Tech in, specifically, in the emerging
AI space?
Professor Alford. So, with respect to artificial
intelligence, it's extremely new. It's extremely emerging. But
we know several things. One, the startup cost and the barriers
to entry are extraordinarily high. Unbelievably high
Senator Lee. Just because of the computing power?
Professor Alford. The sheer computing power. The
information that you need to have inputs into the database, the
training data, the cloud storage alone is just astronomical.
And the potential for mistakes if you don't have sufficient
data are enormous. The parlance, of course, is hallucination.
But if you actually, like, dig down to what is a
hallucination, that's what happens when the training data is
insufficient, and it hasn't been asked that question before in
the information data and so it overfits, is the language that
they use.
Senator Lee. Meaning it makes stuff up.
Professor Alford. It makes stuff up, exactly, but that's
because of the insufficient data. So, you can well imagine that
the data brokers, that those that control the data are going to
have the advantage in, within the AI space and new entrants,
new emerging players are not--are going to have extreme
difficulty trying to have the sufficient amount of data to be
able to compete.
So, the similar problems that you're seeing in Search where
there's just not enough data for tail queries to be successful
on Bing, that kind of thing could happen with respect to AI, as
well.
Senator Lee. It'd be good if they could give us a warning
when it's hallucinating, play a Pink Floyd song, or something
like that.
[Laughter.]
Senator Lee. I don't know. All right. My time's expired.
Chair Klobuchar. Or use a watermark. Okay. All right. Thank
you very much, Senator Lee.
Mr. Baer, when you led the DOJ Antitrust Division, you
brought, as you pointed out, the division's first-ever
prosecution of price fixing case involving algorithms. The
challenge is presented by algorithmic pricing. I have only
become more acute, as we've all, a number of us have pointed
out. What challenges do enforcers face today when attempting to
bring the cases against groups of competitors that use
algorithmic price setting tools to increase prices?
Mr. Baer. The first problem, the biggest problem is simply
detection. This case we brought, in 2015, involved two sellers
of poster art on the Amazon Marketplace. They apparently
concluded that there was too much price competition. They got
together, the two companies in London, and wrote code that
would basically--when anyone searched for poster art, these two
would pop up in random order, but always at the same price.
That's hard to detect. How do we find it out there? I think
it's fair to assume that one of the companies did an internal
audit, found out it was involved in this scheme----
Chair Klobuchar. I see, they reported----
Mr. Baer [continuing]. And came in and ratted themselves
out. That doesn't happen all the time. In fact, it happens very
little of the time. So, it is really----
Chair Klobuchar. And you also don't always have the two
people writing code in London. So, what are they doing now?
Mr. Baer. So, you know, that--the worry, as I talk about in
my testimony, is that you may be able to get to the same result
if everybody just writes code that is roughly comparable, and
the machines who can gather data in nanoseconds can figure out
that cutting price over the long run is a losing strategy. And
so, the worry is, independent decisions by firms in the same
market may result in massive spread of anticompetitive pricing.
That's the worry.
Chair Klobuchar. That's because they're able to, so they
all--they don't really want to reduce prices. So they--even
though that's what competition is about, the reason you reduce
prices to get more customers. But they can figure out through
the algorithms well, where they don't really need to increase
the prices. Is that where it is? Like--so explain it to me.
Mr. Baer. Yes. The notion is that if the machines learn
that if I discount, somebody else's pricing algorithm is going
to price below me, and I'm going to quickly realize whatever
gain I got from that initial price reduction goes away and the
Pareto optimal prices is going back to something that is less
competitive. I mean, to me that's what antitrust is all about.
Finding fact patterns where consumer interests diverge from
company self-interest.
We've seen it in oligopolistic markets over time that firms
independently figure out if we just track each other's pricing
and don't compete, we're better off. With pricing algorithms,
it's quite possible rather than having a concentrated two,
three, four-firm market that the algorithms themselves, the
machine learning will enable everybody to figure out price
gutting isn't good for the bottom line even though it is good
for consumers.
Chair Klobuchar. Yes. So how can Congress ensure that
courts properly evaluate cases involving the algorithmic price
fixing where an agreement's difficult to prove?
Mr. Baer. So, I think the first instance, as I said in my
testimony, I think the antitrust agencies need to develop the
data that establishes this is going on.
You know, antitrust law evolves through common law and to
point out situations that aren't reached by the current
interpretation of what constitutes an agreement, what are the
plus factors from which a court can infer a meeting of the
minds, the knowing use of the same kind of pricing algorithm.
If we can demonstrate that's going on and it has an impact, we
can, as enforcers, encourage the courts.
But it may well be that's not going to be enough. And it
may well be that there needs to be legislation. It's a subtle
tweak to Section 1 of the Sherman Act, but that basically
directs the courts to take into account the anticompetitive
impact of common use of pricing algorithms in the same market.
Chair Klobuchar. Yes, we tend to be able to handle subtle
tweaks better than other things. So, thank you.
The FTC, Mr. Alford, recently sued Amazon for illegally
monopolizing e-commerce related markets. Part of the allegation
center on Amazon's use of price-scraping tools and algorithms
that punish small businesses, relying on algorithms on Amazon--
not algorithms--for setting lower prices off of Amazon. Can you
explain how these anti-discounting pricing algorithms harm
competition and increase prices?
Professor Alford. Yes. So, the Amazon complaint, I think,
is incredibly well written and useful, and they talk about a
variety of different behaviors. But with respect to the
algorithmic behavior, I think there's two key things to note.
One--and both of them is with respect to sellers and the way
sellers behave on the Amazon platform and then off the Amazon
platform.
With respect to on the Amazon platform, it monitors what
third-party sellers are doing. And anytime a third-party seller
on Amazon lowers the price, then Amazon will match that price
exactly.
And the reason that they do that is because they want to
make sure that third-party sellers don't get any sort of
increase in market share. And then it basically dissuades any--
there's no incentive to lower your prices----
Chair Klobuchar. Or increase prices----
Professor Alford [continuing]. Because whatever price you
offer, Amazon's going to go down there.
Chair Klobuchar. They're going to beat you, and they're
going to beat you in a big way when they do it.
Professor Alford. That's right.
Chair Klobuchar. They have the information.
Professor Alford. That's the first thing that they do. That
they basically disincentivize any sort of price competition to
increase market share.
The second thing they do is what I said already in my
opening remarks, and that is they eliminated, under pressure,
the price parity clauses in the contracts, and yet they create
a punishment mechanism for Amazon, where if you sell your
product on a third-party website, including perhaps your own
website, your own company website, at a price lower than what
is available on Amazon, then they're going to punish you by
preventing you from accessing the key feature that everyone
uses to buy stuff on Amazon, which is the ``Buy Now'' or the
``Add to Cart'' feature.
You can't get on there if you offer a lower price on
Target, or Walmart, or any other website online.
And the result of that, because of 98 percent of all Amazon
sales use an ``Add to Cart'' and ``Buy Now'' feature, the sales
of those third-party sellers tank. In my written remarks, I
said that it's like Amazon in this respect is like the
predators in the sci-fi movie, ``A Quiet Place''--if we're
going to use movie references--where they'd seek to destroy any
Amazon seller that dares to utter the sound of price
competition.
[Laughter.]
Chair Klobuchar. Very good, very good. And, I mean, one of
the extraordinary things about that complaint and the FTC, I
thought that when they put it out there, it was pretty
understandable, actually. One of the things that just stuck
with me was that 50 percent of the small business revenue was
going to them----
Professor Alford. Yes.
Chair Klobuchar [continuing]. You know, because they have
no choice and they have to advertise.
Mr. Hewitt, another issue related, algorithms can
perpetuate past patterns of bias and discrimination, which you
well illustrated in your testimony. This can result in fewer
options, higher prices for historically underserved consumers.
What commercial uses of algorithms concern you the most when it
comes to perpetuating bias and discrimination against
underserved consumers?
Mr. Hewitt. Thank you, Senator Klobuchar. I would say
housing and auto insurance are two that really stand out--and
housing in particular, the mortgage market. I mean, so many
Americans are priced out these days, ostensibly because of
housing stock availability and also interest rates.
But even when folks are trying to get a loan, we've seen
research indicating that when algorithms determine what
mortgage rates should be, that we see home buyers--prospective
home buyers of color essentially blocked out and iced out. It
creates this big distortion in the market. And so, it actually
leads to this cascading sense of, ``Well, I can't afford it.''
``They're not going to allow me to afford it.'' And, ``Being a
homeowner isn't even within reach for me.''
With respect to auto insurance, there are companies that
have tried to use geographic location as the means to determine
what types of rates you will--you will receive. And so, if you
think about it, if you happen to be a Black person living in a
majority Black neighborhood, you're going to get sometimes
higher rates quoted for you than somebody who lives in a
neighborhood just a mile away. It doesn't make market sense in
our view.
And so, the whole notion of competition which is within
this Committee's, you know, ambit, really gets turned upside
down. Because how can you have real competition when you have
some subsegments of the consumer marketplace who don't even
have real choices at all?
Chair Klobuchar. Mm-hmm. Well said. All right. I've gone
way over my time here.
So--do you want to let Senator Welch go, and then--okay.
He's been so good sitting there. So, we'll have Senator Welch
go next, and then I'm sure Senator Lee and I have questions for
you, as well. Thank you. Well, go ahead.
Senator Welch. Yes. I'll be brief. This is an incredible
hearing. You know, people think the fix is in, but they don't
really know how. And you're explaining how it is, and it's
really pretty astonishing how small business folks are getting
so incredibly ripped off.
So, I want to thank you for the clarity of your
presentation, and the deep work behind that. And I want to
thank Senator Klobuchar and Senator Lee. I mean, the Senate
needs to be dealing with this, and you're the leaders on this.
So, I express my gratitude for that.
You know, you've been talking--actually, you, Mr. Hewitt,
and Mr. Baer, about housing. It is a brutal problem. Incredible
in Vermont, nobody can live anywhere close to where they work.
It's really tough--that affordability crisis is real, not just
in housing. You mentioned auto insurance, but it's pretty much
everything. You just got to take what you--you don't have any
bargaining power at all.
And housing, do you remember that algorithm, it was called
Rent Maximizer? That was before they realized maybe we
shouldn't advertise what we're doing. And they changed--they
changed the name, but not the game. And all these companies
have been able to evade the antitrust laws, the significant
part because the antitrust laws just didn't anticipate--and how
could they--the world we're in with digital commerce.
So, Mr. Baer, I'll start with you. Just be specific, if you
can, about how our antitrust laws cover any competitive
behavior like this, and what do we have to do with respect to
our antitrust laws and legal standards to address the reality
of this anticompetitive behavior with a different tool? You
know, ``Bob isn't doing it,'' as you mentioned, but it's
happening with the algorithm.
Mr. Baer. Right. So, I think, first of all, some of this
behavior is being reached by not just the public antitrust
enforcers, but by private plaintiffs who are acting on behalf
of hotel customers, on behalf of apartment renters. So, there
is some movement. The challenge is when, as I indicated
earlier, competitors are independently and knowingly using the
same type of algorithm without writing it together, that
produces the same results as if they had sat down in London----
Senator Welch. So, what can we----
Mr. Baer [continuing]. Together, had a drink, and written
the mass. So----
Senator Welch. What can we do about it?
Mr. Baer [continuing]. That's the problem.
Senator Welch. I mean, you've got, you know, one of the
most conservative Senators in the Senate, you know, Senator
Klobuchar, and you've got one of the most liberal, Mr. Lee,
right here.
[Laughter.]
Mr. Baer. Yes.
Senator Welch. Or did I get that wrong?
Mr. Baer. Well, maybe not on this one.
Senator Welch. But they're both here because they share
this common concern. So, what's your recommendation on one of
the remedies we should pursue?
Mr. Baer. So, I think encouraging the FTC to consider use
of Section 5, unfair methods of competition authority, to
bridge the gap between overt collusion and this sort of
indirect collusion is one thing. If the courts aren't going to
embrace that, and we don't know that they will, I think there
needs to be a legislative solution----
Senator Welch. Let me go to Dr. West, because I only have a
little time. Can you discuss some of the risks surrounding the
data collection, and whether or not and why you think a user
should have to give their express informed consent before an AI
system harvests their data?
Dr. West. So, you know, data is a really key source of
power for the firms that are engaged--talk button. Data is a
key source of power for the firms that are using algorithmic
pricing. It's what gives them these information asymmetries
that let them, you know, exact as much out of consumers than
they would otherwise. It's also what enables them to deny other
competitors that lack that data advantage access to the market.
So, you know, given this, you know, there's a feedback loop
in that it creates strong incentives for firms that are
utilizing these tools not only to the leverage the data that
they have, but to continue to collect and retain as much data
as possible. And that's why bright-line rules like data
minimization are particularly key.
Senator Welch. Thank you. Mr. Hewitt, can you discuss the
impact of collecting mass user data, especially without
receiving that informed consent that Dr. West is talking about?
How does that affect lower income communities?
Mr. Hewitt. Sure. Well, the companies that rely heavily on
algorithms, especially whether it be social media or others in
the consumer space, more squarely, they think they're holding
up a mirror to us, showing us who we are. When in fact, it's
really more like a fun house mirror. It's distorted. It's
giving a distorted picture, and it's essentially trying to tell
us who we are.
And the driver of that story about who they think we are is
based upon what Senator Klobuchar said in her opening remarks,
``garbage in, garbage out'' data that is based upon a history
of redlining and discrimination.
It doesn't just so happen that our communities
residentially are de facto segregated. It's because they were
once digitally segregated. It's because of prohibited redlining
practices and now we have digital redlining. That's
essentially--that's what's been happening.
And so, if all of our data is based upon explicit or
structural discrimination, then we can't make that be the
driver of all reality going forward.
Senator Welch. Thank you very much. And by the way, my
first job in law school was with the Lawyers' Committee for
Civil Rights Under Law.
Mr. Hewitt. How about that?
Senator Welch. Yes, it's something. Thank you.
Mr. Hewitt. Thank you for your service.
[Laughter.]
Chair Klobuchar. All right, well thank you very much,
Senator Welch. Senator Lee, you have some follow-ups.
Senator Lee. Thank you. Dr. Epstein, I'd like to go to you
next. America's Digital Shield project has, I think, it's over
13,000 field agents. I see you're nodding your head, I'm in the
ballpark, that allow the collection of data to recreate
authentic online experiences.
And it's my understanding that it operates in a manner
that's designed, maybe, kind of, to replicate or exceed the
capabilities of the Nielsen research data system, with the
latter system referring to TV viewers of course.
It's different when you're dealing with the internet
because it's much broader, much more--it moves much more
quickly. Can you explain a little bit how you analyze this data
to reach some of the results that you're looking for and how
you go about that?
Dr. Epstein. Yes, but first I have to say for the record
that the road to hell is paved with good intentions. And you
folks on this Committee have very good intentions and you're
wonderful people. But you don't understand--you're not really
hearing what's being said to you. What's being said to you is,
you can't regulate algorithms.
I've been a programmer since I was 13. I can tell you for
sure, algorithms, especially nowadays, are completely opaque.
And even if you tried to, they would outprogram you long before
you got that final vote, long before you got a bill passed.
Senator Lee. Yes. I don't recall ever having said that we
should regulate algorithms, nor do I think that's----
Dr. Epstein. Well, you're talking about----
Senator Lee [continuing]. Physically possible.
Dr. Epstein [continuing]. Appointing a special algorithm
regulator.
Chair Klobuchar. There was just one question, I think.
Dr. Epstein. Australia tried regulating algorithms. Your
concerns here are just not where they should be, in my opinion.
Instead, you have to look at what you just asked about, which
is you have to look at where the rubber meets the road. It
doesn't matter who's programming what. What matters is what the
consumer is actually getting on his or her computer.
And all of that is personalized. There's no way to know
what they're getting unless you are looking over the shoulders
of a very large number of real users who cannot be identified
by the tech companies.
And we've been building that kind of technology since 2016,
and America's Digital Shield is our attempt now to build a
permanent, self-sustaining system that looks over the shoulders
of tens of thousands of Americans in all 50 States, politically
balanced, representative of each population in each State, and
court-admissible. And that is what we have been doing.
And it's very, very difficult. That's a $3 million website
you were looking at. It's very difficult to find those voters,
explain to them what we're doing, get them to agree to sign
NDAs, very rigorous NDAs, and that's what we do.
Senator Lee. Okay. Can I ask a question about your Youth
Content Project?
Dr. Epstein. Yes.
Senator Lee. It's my understanding that it's shown that
Google's exposing children to videos on YouTube that are
inappropriate. In some cases, sexually explicit, or otherwise
disturbing, harmful in one way or another. How has your
research uncovered this under the UpNext feature? And how do
your--how does YouTube use algorithms to select the UpNext
content shown to children?
Dr. Epstein. Well, this has already been mentioned by other
people here, but YouTube selects content to increase watch
time. And this is true even for 5-year-olds. They want more
watch time because then the 5-year-old will run to mommy and
say, ``Mommy, buy that.'' So that's the key here. Watch time
equals revenue. Period. Leaks from these companies, especially
Facebook, have shown this unequivocally.
So, what we are doing now, we have, through our field
agents, we've gotten--we've been able to connect to our system,
the mobile devices of more than 2,600 children--children of
some of our field agents who want to know what content those
kids are seeing. And this content--if someone would turn the
screen on, again, it's displaying right now in real time.
This content is shocking. I've had research staff who will
not work with this content. They won't do it because it is so
disturbing. This is going to our kids, and it's going to kids
in such a way that parents don't even know what's going on.
[Large viewing screen is displayed.]
Dr. Epstein. There is no way to understand anything that's
happening on the internet unless you are looking over the
shoulders of tens of thousands of real people. They have to be
real people, real kids, real voters.
That's why we've been working so hard to build this
system--and it works. And we have learned ways to analyze the
data in real time that's streaming in for more than 10,000
computers, and we're going to get that number up to 40 or
50,000, if we can.
Senator Lee. And to be clear, that's content that's
appearing on a platform designed for kids, marketed to kids,
marketed to parents as for kids.
Dr. Epstein. Not only are these images coming from real
videos being watched by real kids, these are recommended----
Senator Lee. Right.
Dr. Epstein [continuing]. These are recommended videos. We
keep track----
Senator Lee. For kids----
Dr. Epstein [continuing]. Of what's----
Senator Lee [continuing]. Recommended.
Dr. Epstein [continuing]. Recommended and what they're--
what people are actually watching.
Senator Lee. Now tell me, if you can, imagine a world in
which Google did not have monopolistic market power. Just
imagine that world for a minute--John Lennon's ``Imagine'' is
playing in the back of your mind. How--how would Google's
behavior change? How would the experience with Google and
Google products change?
Dr. Epstein. I'm not exactly sure what you're asking.
Senator Lee. So, any of the things that we're talking
about. Would--would Google ever have been in a position where
it was telling half the country which side of the political
aisle it wanted them to gravitate toward? Would it be directing
content to kids that it shouldn't be?
Dr. Epstein. If we were monitoring on a large scale and in
a scientifically validated manner, which is what we've been
building here, okay, Google would pull out. They still could
make billions of dollars, but they would pull out. They'd stop
the manipulations of our elections. They'd stop the
manipulation of pricing to the extent that they're doing it.
They would stop because they could still make money, but we
would be making them accountable.
What you're seeing on the screen right now are elections
that Google has flipped. We have a long, long list of elections
that Google has flipped----
[Large viewing screen is displayed.]
Dr. Epstein [continuing]. Because we now know how to
calculate precisely how many of the votes they're shifting in
each election because we not only have the basic research, but
we also now have evidence coming to the eyeballs of real
voters. We have the evidence showing they're actually showing
people this content.
Senator Lee. Thank you. Professor Alford, I want to ask a
variation of the same question to you. What--what would Google
look like, or how would it be different than it is today if it
didn't have monopolistic control? If it didn't have market
dominance that it does?
Professor Alford. Yes. Let me just respond to----
Senator Lee. Yes.
Professor Alford [continuing]. His comment, if I can, as
well. He says the road to hell is paved with good intentions,
but also the road to heaven is paved with good intentions, as
well, I think.
And you cannot regulate algorithms, but you can identify
where the incentives are, and you can address concerns about
conflicts of interest, and you can create structural relief for
the largest companies and impose duties of--the best-interest
duties and things like that on the medium-sized companies.
So you can do things to address the incentives even if you
can't regulate the algorithms directly. The answer to your
question----
Senator Lee. It's not the algorithm that you're really
after, it's----
Professor Alford. It's the incentives.
Senator Lee [continuing]. Wanting to regulate. Yes.
Professor Alford Exactly. But to answer your question, if
there was competition, genuine competition in the marketplace,
then companies would do the best that they can to increase
their market share, and address all of their audiences in a way
that was respectful of them. Right?
They would--they would engage in behavior that increased
the chances of them maintaining the audience. Because if the
audience is upset with you, they'll go somewhere else.
We know with other markets where there are price
competition, or quality competition, or viewpoint competition
that individuals migrate to those other alternatives. Right?
And so, I think the fact that there's not sufficient
competition in these marketplaces means that they do what you
said, which is they abuse their own clients. And they wouldn't
do that if there was genuine competition.
Senator Lee. Thank you.
Chair Klobuchar. Okay. Thank you very much, Senator Lee. I
guess I'd start out with you, Dr. West, of, sort of, this area
of what we can do, and no one is serious that we're going to
lean, we're going to look over at people's algorithms and the
like, and be in their homes.
To the opposite. We want to get more privacy regulation and
control their rights on their data, and we also want to protect
kids more. There's many upgrades we can make to those laws
and--but right now we're focused on the competition piece of
it.
And Dr. West, one idea that would be helpful, I think, to
figure out what's going on, is to allow more transparency. We
are talking about the opaqueness of this. How can increased
transparency help protect consumers from anticompetitive
conduct by algorithms? And of course, we had this bill, Senator
Coons and I, and others about allowing researchers to know
what's going on. Dr. West.
Dr. West. Absolutely. I think it's a great question. So,
for one, given the domains in which algorithmic systems are
being deployed, it's particularly important that consumers know
when algorithmic systems have been used in making decisions
that are shaping their lives, their access to resources, their
life chances.
And that's a step toward addressing the significant
information asymmetries that give dominant firms their power.
But it's also clear that transparency won't be enough, because
often these systems are used by those with, you know, greater
power on those with comparatively less power. You know, think
about in the workplace, in, you know, real estate, where you're
applying for housing in a concentrated market.
There may not be an option to opt out or to seek remedy,
and that's why complementing transparency mandates with strong
enforcement of our existing laws and other bright-line measures
is going to be especially important to protecting both
consumers and competition.
Chair Klobuchar. Okay. Very good. And would you like to add
anything else about other issues that have been raised here?
Dr. West. I would just concur with Professor Alford that
you can regulate companies, and we should regulate tech
companies, in particular.
Chair Klobuchar. Yes. As you know, we haven't done any--
we've passed things out of this Committee, and Senator Lee and
I passed the Venue bill, which allows State AGs to do more, and
we've given some more resources to our enforcers.
But so far, getting these through the floor in both Houses
has alluded us in terms of putting some guardrails in place,
and I think that this--the time is passing here and we have to
get this done.
So, I guess I turn to you, Mr. Hewitt. You heard about the
discussion about what can be done. Could you talk a little bit
about what Congress can do to get at some of the issues that we
have raised that you and I talked about with discrimination?
Mr. Hewitt. Sure. Well, transparency, Senator Klobuchar, is
certainly part of it. We also need something that frankly could
trigger an incentive to do the right thing, and that is
potential liability. That's where----
Chair Klobuchar. I was--when Senator Lee was, sort of, he
paused and he said, ``What if Google didn't have,'' and then
you, you said a monopoly, which is true and a very valid
question. I thought you were going to say, ``What if they
didn't have immunity from any lawsuit? ''
Mr. Hewitt. Well, it's----
Chair Klobuchar. And that to me is something Senator Graham
has been pushing for a long time because of the inability of
the Congress to take any action on any of these things. And
luckily, we have, again, given the tools or resources to the
antitrust enforcers to do it.
But we really don't have any choice, then, because there
is--unless we just want this to continue where fentanyl-laced
pills are being sold on these sites--one-third of the people
that get poisoned with fentanyl get them off the platforms,
whether we want to just continue with some of the pornography,
the revenge porn, all these things going on. There's just no
liability, and they're raking in like a $1 trillion dollars,
so. Oh, that's how much their value is.
Mr. Hewitt. That's right.
Chair Klobuchar. So, I just--at some point, they use some
of that profit to get the stuff off their sites. It's just
like--it's pretty much common sense when we regulate everything
else. But Mr. Hewitt, I didn't mean to interrupt you. But thank
you for raising the point. Continue.
Mr. Hewitt. Well, look, I will just--I know we're short on
time. I will say there's a host of harms. As this Committee
does its work, I'm also thinking about how this fits into the
AI Insight Forum, where I saw you a few weeks ago, where we
focused on the impact of AI on elections and democracy.
I think about the fact that platforms like YouTube are an
information super highway for white supremacist content. The
same content that inspired the shooter in Buffalo who went to a
Black neighborhood just to kill Black people, that content was
on YouTube in 2019. We sent a letter asking them to take it
down from a particular user. But we can't whack every single
mole.
And so, you're right that potential liability must be on
the table for these companies, whether it be in terms of social
media companies or other companies. And we're not talking about
lawsuits flying left and right, we're talking about
incentivizing proper conduct.
If anyone on this Committee or any of your staff used a
toaster for breakfast this morning, that toaster is more
regulated than AI is regulated today. That does not make any
common sense, and that is a problem that Congress can address.
Chair Klobuchar. Well said. Mr. Baer, many dominant tech
companies use algorithms, as we know, to determine what
consumers are going to see in response to a search query. This
gives them, as we just talked about with Professor Alford, it
gives them ability to preference their own products and bury
those of their competitors.
That's one of the reasons that Senator Grassley and I and
others on this Committee introduced the American Innovation and
Choice Online Act, to set commonsense rules of the road. How
can this type of self-preferencing by entrenched gatekeepers
harm competition and deter innovation?
Mr. Baer. Well, if people are locked in, if companies, the
platforms have market power, if they are essential to
communicate with consumers, that gives them the power to
exclude rivals, to preference their own products, to extract
rent from people, to get placement on the website that
disadvantages those who want to compete.
People have talked about the Amazon complaint by the FTC,
which I agree is a very thoughtful, plain English description
of some of these bad behaviors. But this notion that you can
penalize people that price lower on sites other than Amazon,
that tends to be a most-favored-nations thing where the rising
tide inappropriately lifts all boats up to the same price.
Amazon, too, according to that complaint, can basically
penalize and does penalize firms that have, find a less
expensive way of delivering the product to a consumer not using
Amazon's Prime delivery service.
As you noted earlier, that a company that is on Amazon, and
you need to be Amazon because it's a dominant provider, they
basically get a whole lot less of the price a consumer is
paying for the product or service they're buying.
And if you can't offer it for less anywhere else, we have
this situation that your legislation would address that would
stop those dominant firms from profiting off their dominant
position.
Chair Klobuchar. Thank you. Here's an interesting one, Ms.
West. So, we've always assumed that these price-fixing cartels
are likely to form in highly concentrated industries. Right?
Because they're the ones that they're--they tend to be able to
say, okay, let's just fix this price, and then there aren't
other people out there, so we're going to make a lot of money.
But algorithms actually can expand the number of industries
in which price fixing can occur. How does the use of algorithms
and artificial intelligence potentially expand the types of
industries where price fixing can take place?
Dr. West. So, as you've noted, historically, we would look
for cartel-like behavior in markets where there's either a
small number of players, making it easy for them to coordinate
with one another, or where all of the key firms are selling the
same kinds of products. It would be too hard to coordinate
behavior otherwise.
But what algorithmic systems do well is they enable firms
to deal with complexity. So, it enables firms that would
otherwise have difficulty colluding, to effectively coordinate,
whether intentionally or not, using large amounts of data
across differentiated products. And the effect that this has is
to extend the potential for collusion into places where it
otherwise would've been near to impossible to execute in
practice.
Chair Klobuchar. Dr. Alford, in your testimony, you note
that algorithms can make price-fixing cartels more stable by
minimizing the incentives to cheat. Can you talk about how
algorithms can make price fixing cartels more sustainable?
Professor Alford. Yes. So, the algorithm price-fixing
behavior that's possible is essentially where you defer to a
central decision-maker about how pricing will happen.
And so, as Dr. West said, you could have literally
thousands or even millions of individuals that could defer to
that. And the stability then would be decided because there's
deference to a central decision-maker that can make that, and
then there's stabilization that can happen as a result of that.
So that's one way.
The other way is what I mentioned earlier very briefly with
the signaling features where--so, consumers or regulators may
have a hard time identifying price signaling behavior that is
happening, but because of algorithms, you can do very, very
fast signaling to your competitors about pricing behavior. And
only if you're really knowing what to look for will you even be
able to see it.
And so, we see that in the airline industry where there's
so-called flash information sharing where signaling can happen
because of the instantaneous nature of it, and they can
immediately signal and then send back the price so that it
doesn't even reach the consumer necessarily, but it reaches the
seller. That's another example of stabilization.
Chair Klobuchar. Very good. Senator Lee, you have any
other? Okay, good. This has been actually really helpful for
all of us because--and I like some of the ideas that you've all
had, the transparency, looking if there's some changes we need
to make to the laws that would actually, could be very tailored
to the price fixing focus.
We know with AI in general, Mr. Hewitt, that--and with the
algorithms, that there can be all kinds of discrimination based
on past information that just is putting you in a place. And
it's been proven time and time again when people have done
experiments with it that we cannot be. And so, I think we know
in that grouping of changes we're going to make, I think
there's a lot of possibility to be looking at this as well.
So as hard as this all seems, as my colleagues expressed, I
actually see some potential for us to move on this on a
bipartisan basis because of the fact that we're going to be
doing a bunch of stuff on AI, and this kind of gives us the--
because it's so closely linked. Not the same, but linked.
It kind of gives us some possibilities to move there or to
do a separate thing as Senator Lee and I have done successfully
in the past, some, a special thing, that's focusing on the on
algorithms and what that means.
And I just think, just to stand by and think we're not
going to do anything. My favorite example was your example, Mr.
Hewitt, about the toaster. And, you know, we do major, major
deals about dressers that are falling down, which is a legit
thing that killed kids. Right?
But these things right now are killing so many kids a day,
from them getting hooked on drugs or other things that they're
getting off these platforms. And for us to just do nothing, it
just would literally be a dereliction of duty.
So that's why I think we're just starting to see change.
Certainly, the public is with us, at least a jury of our peers,
of 12 people were with us on one day, and the cases are moving
and all of that.
And my plea, as I do in almost every hearing, is that the
companies who I know have people out there--they have their
people, that they--many of them, that they listen, and that
they think of how we can work on these things to actually put
some guardrails in place.
Because otherwise, they're leaving us with no alternative
but the legal system and lawsuits, and I think there is a
better way to go here. Okay. Anything else you want to add?
Senator Lee. No. Thank you for being here. This is a great
hearing. Learned a lot from you. Thanks.
Chair Klobuchar. Okay. We are going to keep this record
open for 1 week, and the hearing is adjourned. Thanks.
[Whereupon, at 4:45 p.m., the hearing was adjourned.]
[Additional material submitted for the record follows.]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
A P P E N D I X
The following submissions are available at:
https://www.govinfo.gov/content/pkg/CHRG-118shrg60435/pdf/CHRG-
118shrg
60435-add1.pdf
Submitted by Chair Klobuchar:
Center for Democracy & Technology, December 13, 2023, letter.... 2
Responsible Online Commerce Coalition, December 20, 2023, letter 4
Submitted by Robert Epstein, Ph.D.:
Supplement to the Prepared Statement, pp. 9-480................. 9
Supplement to the Record, January 3, 2024, letter............... 481
Submitted by Mr. Damon T. Hewitt:
Supplement to the Prepared Statement, pp. 15-50................. 482
[all]