Rona Wang, an Asian American MIT graduate, asked an AI image generator for a LinkedIn headshot and it lightened her skin and gave her blue eyes. AI systems absorb biases from their training data, which comes mostly from a few elite Western institutions.
Is AI racist? Some examples seem to be showing that. Okay, so picture this. An Asian American MIT graduate, Rona Wang decides to spice up her headshot with the help of an AI image generator, asking for professional LinkedIn profile photo. But guess what happened? It transformed her appearance, including her race. It made her skin lighter and gave her blue eyes. It seems like AI's got a preference here when it comes to the term professional. But hold up. You need to understand what's going on here. At first glance, the seems like obvious racial bias by the AI. But what's really going on here? The truth is there's no such thing as inherent racial biases with AI. All they do is absorb the biases that exist in their training data. And a recent research study found that most benchmark data sets used to train AI models lack diversity because they come from just a handful of elite Western institutions. And in Rona's case, the AI probably associated professional headshot with the limited examples of professional headshots in had seen before, and those were predominantly of white people. It had no ill intent, just very narrow training data.
