AI Report Digital - Flipbook - Page 17
Responsible, Ethical AI
The Consumer Financial Protection Bureau
Organizations that want to leverage artificial intelligence capa-
In line with its mission to protect American con-
bilities may want to come up with ethical guidelines and respon-
sumers, the CFPB stands out among financial reg-
sible use policies. These guardrails can help guide how, where
ulators for its interest in how financial institutions
and to what end an institution should apply AI technology.
are using AI.
One example of this is Microsoft’s responsible AI standards,
“Unchecked ‘AI’ poses threats to fairness and to
a 27-page document it shares publicly and with companies
our civil rights in ways that are already being felt,”
it works with, says Daragh Morrissey, director of AI at
said CFPB Director Rohit Chopra in an April 2023
Microsoft Worldwide Financial Services. A financial institu-
speech. “When consumers and regulators do not
tion’s responsible AI policies could include outlines for secu-
know how decisions are made by artificial intelli-
rity and privacy protocols, fairness and inclusivity, reliability,
gence, consumers are unable to participate in a fair
transparency and appropriateness. Morrissey also recom-
and competitive market free from bias.”
mends that financial institutions ask their vendors about their
responsible AI approaches and compare how those policies
The CFPB is concerned that AI and algorith-
line up with their own.
mic-powered credit decisions could lead to “digital
redlining,” which he defined as “redlining caused
through bias present in lending or home valuation
algorithms and other technology marketed as arti-
Financial institutions should consider
their own ethical framework amid evolving
regulatory scrutiny.
ficial intelligence.” Redlining is a discriminatory
practice where borrowers are denied access to
credit or services based on factors like race; its historical practice leveraged maps of areas that were
deemed more or less desirable and were delineated
by red lines.
President Biden’s Executive Order
In April 2023, the regulator signed onto an interagency statement regarding enforcement efforts
In October 2023, President Joe Biden issued an executive
against discrimination and bias in automated sys-
order regarding the development and use of artificial intelli-
tems, along with the Federal Trade Commission, the
gence. The order is intended to coordinate the government’s
Equal Employment Opportunity Commission and
approach to AI development, promote responsible innovation
the Department of Justice’s Civil Rights Division.
and competition, and keep America safe.
The statement reminded enterprises that the use of
The order makes clear that AI technologies do not excuse
organizations from complying with consumer protection laws,
which are especially important in financial services. Further,
automated systems and other “innovative new technologies” falls under these agencies’ existing legal
authorities to uphold consumer civil rights.
the government will enact safeguards against harmful AI out-
And in May 2023, the CFPB published a circular
comes like “fraud, unintended bias, discrimination” and privacy
reminding financial institutions of their obligations
infringement. The administration encouraged the Consumer
to follow consumer financial laws and adverse
Financial Protection Bureau to require its regulated entities to
action requirements, along with the clarifica-
“use appropriate methodologies including AI tools” to comply
tion that a technology’s complexity, novelty or
with federal law, including evaluating their underwriting mod-
opaqueness is not a defense for violating the law.
els for bias or disparate impact among protected groups, along
It followed up that circular with September 2023
with collateral and appraisal valuations that can minimize bias.
guidance about adverse action notices when AI has
The order also stipulated that the Department of the Treasury
been used in a lending decision. Lenders provide
publish a report looking at best practices for managing
those notices to a consumer if they’ve denied credit.
AI-specific cybersecurity risks at financial institutions.
ARTIFICIAL INTELLIGENCE: A REAL-WORLD APPROACH | 15
