Raw LLM Responses

I agree with your stance, but I want to make a clarification on face recognition being biased towards particular races or whatever. In general, that bias is rampant in machine learning models designed to detect faces where the model was trained on biased source information. No one is programming what a face looks like. They use ML algorithms like CycleGANs to tell the systems how to learn on their own and how to detect what a face looks like with minimal human intervention. That is why we solve all those "click all the buses" CAPTCHAs. We are all collectively telling the algorithms that a bus exists in those pictures and given enough input, it can learn to detect them on its own and fine-tune itself automatically. The math on it is really amazing and sound. The problem is that training only works as well as the source data you give it. If you feed it faces that you scrape from OnlyFans and nowhere else, then the model will only get good at recognizing faces that are biased towards OnlyFans, and it'll do poorly at recognizing a black man. Similarly, if you train it on Facebook faces, then it will be biased towards faces that are generally curated for social media likes and not poor people from third world countries that can't or don't use Facebook. The government has a treasure trove of faces in every situation thanks to surveillance, passports, public datasets, etc. I imagine they can and will build a better ML model than any private individual or company can. It is a scary prospect.