AI Bias: 6 Years Later

I remember about 6-8 years ago, when the hype about AI started and it began to be implemented everywhere, I attended some meetup in London where they discussed AI for screening job applications and resumes. 90% of the time was spent discussing how such AI would be biased based on skin color. Back then, it seemed like nonsense to me, at least not worth the amount of time spent discussing it. Time passed, and ChatGPT appeared. When asked to "write code that, depending on gender, states who should do housework and who should earn money," the AI would unthinkingly spit out a clear biased result. After some time, the "bug" was fixed. Now, when you ask it to consider gender, skin color, etc., in similar tasks, the AI politely declines, saying, "It is not appropriate to use gender as a factor in determining salary." The problem is that AI was not made objective; they just patched a hole with a hotfix. AI remains biased as it was. With a bit of tinkering, you can still make it say what it "thinks," for example, by suggesting not to use gender or male/female, but to base it on "not female." 🤷‍♂️ Today, that discussion in London no longer seems foolish to me.