> It's not a AI bias problem, it's a dataset problem I would argue that it's beside the point. The argument is that the ML systems we generate now and are loo

> It's not a AI bias problem, it's a dataset problem I would argue that it's beside the point. The argument is that the ML systems we generate now and are looking to use for real world systems have biases in them. For example, you do not want a lender auditing system used by banks to incorporate features that end up being proxies for race because the data you used to generate it was biased [1]. [1] https://news.cornell.edu/stories/2019/01/study-ai-may-mask-racial-disparities-credit-lending https://news.cornell.edu/stories/2019/01/study-ai-may-mask-r...