Can AI erase enough of its own bias to address racial and gender gaps?
VCs and lenders rely on business data such as transaction records to inform funding decisions for fintechs or small businesses, but bias and uneven representation in management still result in gender and racial disparity.
Machine learning, which analyzes data and credit qualifications absent direct human interaction, is emerging as a potential way to mitigate exclusion and selection bias, provided AI can overcome its own shortfalls.
“AI can create more inclusion by elimination. It doesn’t eliminate people but it reduces the bias of people,” said Janét Aizenstros, the chair and CEO of Ahava Digital, a Kitchener, Ontario-based digital consultancy that produces consumer data aimed at reducing gender, ethnic or racial bias.
The racial and gender gaps in venture capital are stark. According to Deloitte, 80% of partners at U.S. VC firms are white, 14% are women, 15% are Asian, 3% Black and 3% Latinx. Beyond discouraging equal opportunity for employment in the VC industry, these numbers also create problems for companies trying to raise capital, since a vast majority of the decisions on pitches, funding and meetings involve white men.
That effectively reduces opportunities for entrepreneurs and developers in a variety of markets including fintech, e-commerce, small business and payment processing. VC firms with no women partners invest in companies with women CEOs three times less than VC firms with women partners, according to the Fintech Health Network.
The gaps are large enough that it will take years to address, requiring evolution in corporate culture, social change and education.
“There’s not a clear-cut answer. It’s a combination of people, technology and government coming together to create a steady move toward getting better, but there’s no immediate remedy,” Aizenstros said. “It took us 400 to 500 years to get here and it will take time. But one of the things that can help is better AI technology.”
Aizenstros, who has worked in finance roles at TD Canada Trust, Canon, and has a background in metaphysical sciences, argues AI technology can level the playing field by reducing the amount of screening that’s based on social interaction, which can contribute to subconscious factors in decisioning that aren’t necessarily data-driven.
AI also has bias challenges because systems are fed by data that can have biased sources. IBM, for example, has identified about 180 biases that have infiltrated AI analytics.
“The system can be manipulated because there’s still human biases,” said Aizenstros, adding that can be reduced through a system that deals with business criteria more agnostic to cultural differences.
Newer AI systems are attempting to identify biases. The MIT-Watson AI lab has attempted to spot inconsistencies in decision making that are then embedded in the AI analysis. IBM did not return a request for comment by deadline.
“Everyone’s concerned about AI doing the same thing because it’s based on ‘previous data’ but if you look at the economics of a business, how much money that’s coming in per unit, you get much more of an understanding of that business and its audience,” said Michele Romanow, co-founder of Clearbanc, a Toronto-based firm that has invested $1 billion in more than 2,000 companies in Canada and all 50 U.S. states.
“All decisions are made with AI,” said Romanow. “Even if you have data but still have humans take over and be involved in the decisioning, there’s going to be bias.”
Romanow appeared as an investor on "Dragon’s Den," a Canadian version of "Shark Tank," an experience she said gave her insight into the struggle firms have getting funding from VCs. Clearbanc’s model is designed to boost gender and racial inclusion by using AI and a mix of financial metrics that are also designed to mitigate the amount of funding startups pay to Facebook or Google to acquire customers.
“It’s rare to get a good idea from reality TV,” Romonow said. “But I saw companies giving up a huge portion of their equity dollars.”
The result is a model based on a share of future payment transactions and revenue. It connects to Stripe, for example, and uses that data to vet the company. It uses read-only access to these accounts and AI to measure the startups. Romanow contends this model has resulted in eight times the investment in women-owned businesses, though Clearbanc did not provide a detailed breakout.
“Everyone’s concerned with how you build AI. Does it have the same biases because it’s using existing data?” Romanow said. “But if you look at the unit economics of a company and use that in the AI, you understand much more about the company and its strength.”