The developed model showed a correlation coefficient of 0.95 between the AI-estimated age and chronological age. AI can tell if you're gay too, just by looki

The developed model showed a correlation coefficient of 0.95 between the AI-estimated age and chronological age. AI can tell if you're gay too, just by looking at you. Except it obviously can't it just makes up some stereotypes that work for 85% of the test split it was evaluated on, but have no causal relationship. I bring this up because I find it very sketchy using AI to estimate stuff that works in some stereotype to give good predictive power but doesn't actually "reason" about anything, so you can get wild errors whenever inputs don't fit the stereotype.