AI Bias Pt 1

You know, AI is machine learning, also known as ML, which came from computer science. And the history of AI goes way back to the US military and the UK military and a few universities in the US, which means that the people who built, defined, and iterated on AI technologies were people who, one, had access to computers, two, were not confined to the kitchen washing dishes, and three, came from a certain socioeconomic background. You probably see where I'm headed here. So this means that the outputs and products of AI comes from one certain way of thinking and lived experiences. Now, the problem with that is AI is a technology that makes decisions for people, life-changing decisions, and if it's not trained on representative, accurate, and updated data, it'll make inaccurate and wrong decisions, harming lots of people. This is what AI bias means, and unfortunately, it's a problem that has persisted from the onset of ML and hasn't made a lot of progress over the years. Let me give you one example when it comes to gender bias. You've probably heard about Amazon's recruitment tool that discriminated against women's CVs and resumes, and it only favored male CVs because the AI tool recognized only male names and colleges and not women's. So anytime a woman applied for a job through the tool, it didn't put their applications through. The tool was shut down once the issues were discovered. Now fast forward to today, a researcher discovered that ChatGPT's responses to girls versus boys was different. When asked technical questions, boys received more technical answers while girls were given more flowery and princessy kind of responses. So we all need to be aware and conscious of prevalent biases in AI systems, which are highly dependent on the use case and context of how you work and interact with AI.