BofA’s Bessant: Need to guard against bias in AI banking tools
With the barrage of AI-based banking tools on the market, cultural- and gender-based algorithmic biases are top of mind for bankers. As AI and machine learning tools are increasingly deployed to anticipate customer behavior or assess creditworthiness, taking a step back and understanding how bias is introduced into the process is key.
Speaking at SourceMedia’s Most Powerful Women in Banking LEAD Conference in New York on Wednesday, Cathy Bessant, Bank of America’s chief operations and technology officer, said AI’s growth prospects can’t be considered without looking at the societal impacts. For the past year, the bank has been working with the Harvard Kennedy School’s Belfer Center for Science and International Affairs on a multidisciplinary Council on the Responsible use of Artificial Intelligence. One question on which she said the committee was deliberating is whether external regulation will be necessary for the responsible use of AI.
“Responsible AI gets the power of AI and makes the world a better place,” Bessant said. “AI can tell you everything you need to know, [but] the decision to use it is a human decision. Responsible AI is that intersection between humans and technology to produce a better outcome.”
To get there, Bessant stressed that people ultimately need to make a judgment call on whether the outcomes of AI-powered efforts are useful or not and whether they advance the objective at hand. “People get sold on AI in all kinds of firms,” she said. “I want all of the people in my company, including the front-line business unit leaders to understand what it is [technology leads] are pitching us.”
Bank of America, through its work with the Council, is looking to tackle where bias is introduced. “We are aggressively tackling bias – gender, racial and cultural,” Bessant said, noting that models can often take on the perspectives of the humans who developed them. “Backtesting for bias is very important, but the key is what is the question we’re trying to solve for and that is often where the top of the bias begins.”
