[Congressional Record Volume 170, Number 173 (Thursday, November 21, 2024)]
[Senate]
[Pages S6701-S6702]
From the Congressional Record Online through the Government Publishing Office [www.gpo.gov]
Artificial Intelligence
Mr. MARKEY. Mr. President, I rise today to discuss an issue that has
been a dominant topic of conversation in this country: artificial
intelligence, or AI.
Over the past 2 years, generative AI products like ChatGPT have
exploded in popularity, while companies have invested tens of billions
of dollars developing new AI models. The AI innovation race is in full
swing, and I share much of this excitement.
AI holds great promise with the potential to transform the way we
live, we work, and interact with the world. But as with any race, there
are winners and there are losers, and I am deeply worried that we are
too focused on the potential winners of this AI race and we are
neglecting those who are already being left behind, because although AI
may be having a moment right now, we have been living with its impact
for years.
This is especially true for Black and Brown and immigrant and LGBTQ
communities which have borne the cost of the ``move fast and break
things'' mindset that pervades Silicon Valley.
To understand why, it is helpful to understand how these AI-driven
algorithms actually work, because these algorithms are really just
supercharged pattern recognition systems. It is not all that different
from teaching a dog a new trick.
For example, to teach a dog to fetch, you gather a bunch of balls and
treats and instruct the dog to run after and return the ball and
provide feedback--treats--along the way. Over time, the dog recognizes
the pattern: Run after it and return the ball and receive a treat.
The same principles apply to algorithms. A company will feed
historical information into a computer with basic instructions. Like a
dog that looks helpless when first told to fetch, an algorithm may
first provide confusing responses to these instructions. But over time,
as the algorithm receives feedback from a trainer, the algorithm will
improve at responding to the instructions.
Here is the problem: This training and the algorithm's decision
making and recommendations happen outside of public view. This lack of
transparency and the failure to properly test for bias in algorithms
create serious risk.
What happens when the information being fed into an algorithm
reflects
[[Page S6702]]
preexisting societal biases? To return to our dog analogy, if the dog
is only trained to fetch tennis balls, it may not understand how to
fetch a stick, or it may return the stick to a different person.
The same problem happens with algorithms involved in decisions around
banking, healthcare, and the criminal justice system, but with much
more serious results. Numerous studies and reports have revealed the
consequences that biased algorithms have for marginalized communities.
For example, a 2019 report found that due in part to biased mortgage
approval algorithms, lenders were 80 percent more likely to reject
Black applicants than similar White applicants. On another occasion, a
tech company found that its AI resume screening tools recommended male
applicants for jobs at much higher rates than similar female
applicants. And, just yesterday, a comprehensive new report found that
92 million low-income Americans are subject to algorithmic decision
making and, therefore, potentially subject to bias and discrimination
baked into these algorithms.
This is unacceptable. That is why, in September, I introduced my AI
Civil Rights Act, comprehensive legislation intended to ensure that the
AI age does not replicate and supercharge the bias and discrimination
already prevalent today in our real world.
Specifically, my legislation would impose new rules when companies
use algorithms to make decisions on employment, banking, healthcare,
the criminal justice system, and other important aspects of our lives.
Today, I am proud to announce that 54 new organizations have endorsed
my bill, including some of the biggest labor unions in the country,
critical housing organizations, and indispensable civil rights groups.
In total, 80 civil rights organizations and AI experts have endorsed my
AI Civil Rights Act.
This support sends a clear message: As Congress considers AI
legislation in the coming weeks and years, we must ensure that the AI
age does not come at the expense of already marginalized communities.
We cannot allow AI to stand for ``accelerating injustice'' in our
country. We have a choice. Do we promote innovation without addressing
AI bias and discrimination? Do we protect profits instead of people? Do
we allow biased black box algorithms to control our lives?
Make no mistake: We can have an AI revolution while also protecting
the civil rights and liberties of everyday Americans. We can support
innovation without supercharging bias and discrimination. And we can
promote competition while safeguarding people's rights. And that is why
we must pass my Artificial Intelligence Civil Rights Act.
This is the beginning of an era where we can do this right. But it is
up to Congress to make sure that that happens.
With that, I yield back.