Ryan Browne and MacKenzie Sigalos, CNBC, June 23, 2023
Artificial intelligence has a racial bias problem.
From biometric identification systems that disproportionately misidentify the faces of Black people and minorities, to applications of voice recognition software that fail to distinguish voices with distinct regional accents, AI has a lot to work on when it comes to discrimination.
And the problem of amplifying existing biases can be even more severe when it comes to banking and financial services.
Deloitte notes that AI systems are ultimately only as good as the data they’re given: Incomplete or unrepresentative datasets could limit AI’s objectivity, while biases in development teams that train such systems could perpetuate that cycle of bias.
Rumman Chowdhury, Twitter’s former head of machine learning ethics, transparency and accountability, said that lending is a prime example of how an AI system’s bias against marginalized communities can rear its head.
“Algorithmic discrimination is actually very tangible in lending,” Chowdhury said on a panel at Money20/20 in Amsterdam. “Chicago had a history of literally denying those [loans] to primarily Black neighborhoods.”
In the 1930s, Chicago was known for the discriminatory practice of “redlining,” in which the creditworthiness of properties was heavily determined by the racial demographics of a given neighborhood.
“There would be a giant map on the wall of all the districts in Chicago, and they would draw red lines through all of the districts that were primarily African American, and not give them loans,” she added.
“Fast forward a few decades later, and you are developing algorithms to determine the riskiness of different districts and individuals. And while you may not include the data point of someone’s race, it is implicitly picked up.”
Indeed, Angle Bush, founder of Black Women in Artificial Intelligence, an organization aiming to empower Black women in the AI sector, tells CNBC that when AI systems are specifically used for loan approval decisions, she has found that there is a risk of replicating existing biases present in historical data used to train the algorithms.
“This can result in automatic loan denials for individuals from marginalized communities, reinforcing racial or gender disparities,” Bush added.
“It is crucial for banks to acknowledge that implementing AI as a solution may inadvertently perpetuate discrimination,” she said.
Frost Li, a developer who has been working in AI and machine learning for more than a decade, told CNBC that the “personalization” dimension of AI integration can also be problematic.
“What’s interesting in AI is how we select the ‘core features’ for training,” said Li, who founded and runs Loup, a company that helps online retailers integrate AI into their platforms. “Sometimes, we select features unrelated to the results we want to predict.”
When AI is applied to banking, Li says, it’s harder to identify the “culprit” in biases when everything is convoluted in the calculation.
“A good example is how many fintech startups are especially for foreigners, because a Tokyo University graduate won’t be able to get any credit cards even if he works at Google; yet a person can easily get one from community college credit union because bankers know the local schools better,” Li added.
The problem, according to Kim Smouter, director of the group European Network Against Racism, is that it can be challenging to substantiate whether AI-based discrimination has actually taken place.
“One of the difficulties in the mass deployment of AI,” he said, “is the opacity in how these decisions come about and what redress mechanisms exist were a racialized individual to even notice that there is discrimination.”
“Individuals have little knowledge of how AI systems work and that their individual case may, in fact, be the tip of a systems-wide iceberg. Accordingly, it’s also difficult to detect specific instances where things have gone wrong,” he added.
Chowdhury says there is a need for a global regulatory body, like the United Nations, to address some of the risks surrounding AI.
Though AI has proven to be an innovative tool, some technologists and ethicists have expressed doubts about the technology’s moral and ethical soundness. Among the top worries industry insiders expressed are misinformation; racial and gender bias embedded in AI algorithms; and “hallucinations” generated by ChatGPT-like tools.