Bias can be found in a variety of businesses, including the financial industry. In this video clip from a Motley Fool Live interview, recorded on April 21, TackleAI CEO Sergio Suarez Jr. answers a few questions from Fool.com contributor Rachel Warren about how AI can be used to fight against financial discrimination.
Rachel Warren: So, if the idea here, if I understand correctly, is that some of the models are fundamentally flawed in the way that they're interpreting the data. What are some of the ways that AI can actually help to combat finance discrimination?
Sergio Suarez: The thing is they're not misinterpreting the data. That's the data. [laughs] I think that people with Hispanic last names could end up getting lower credit scores. What people don't know is a lot of times when you're filling out loan applications, you think, "Hey, what's your race?"
The reasoning behind it is to be able to help them make sure there's no discrimination, but when you feed that into models, it doesn't know why you're telling it what the race is.
It studies that and says, "Hey, well, people with Hispanic last names tend to have lower credit ratings," and that's what ends up happening. These credit bureaus don't tell you the algorithms they're using, a lot of times, because they don't understand the algorithms that they're using.
That sort of bias is super, super common and if we continue to just have these deep learning models run rampant on their own by analyzing historical information that's not curated, we'll continue to get the same results.
Warren: Absolutely. What is the key solution here in your view? Is it adjusting algorithms?
Suarez: No. Actually, I think it's the data that we're feeding to our models. If you give it biased data, it's going to reinforce itself. I think what's really important is the people who are building these AI models and the neural networks, I think that it's important for them to be diverse as well.
At TackleAI, we really focus on having a diverse team because one is we're building software for the entire population, so I want to make sure that we reflect that. But on top of that is you need to see all the viewpoints when you're curating data for an AI set.
A lot of times, we get perspectives from people that we just wouldn't know if you're in the United States, for example. We have people from Kazakhstan and Nepal and Ukraine and Mexico. That's really what you need.
If you have one type of people that are constantly the ones that are training the models, then you're only going to get their viewpoints on the data that they have curated, so I think it's important to have that curated data by super-diverse people.
