Introducing PedoAI. The first deep learning driven physiognomy model built to distinguish predatory pedophilia, created from a dataset of 1.2 million criminals
While this linear model's overall predictive accuracy barely outperformed random guessing,
I was tempted to write this up for Pivot but fuck giving that blog any sort of publicity.
the rest of the site is a stupendous assortment of a very small field of focus that made this ideal for sneerclub and not just techtakes
Hashemi and Hall (2020) published research demonstrating that convolutional neural networks could distinguish between "criminal" and "non-criminal" facial images with a reported accuracy of 97% on their test set. While this paper was later retracted for ethical concerns rather than methodological flaws,
That's not really a sentence that should begin with "While", now, is it?
it highlighted the potential for facial analysis to extend beyond physical attributes into behavior prediction.
the blog tagline is "Dysgenics, forecasting, machine learning, sociology, physiognomy, IQ, simulations", so he tells us straight up what's wrong with him
The implication here that it isnt methodically flawed is quite something.
E: and I don't have the inclination for to do the math, but a 97% accuracy seems to be on the unusable side considering the rate of 'criminals' vs not-criminals in the population. (Yeah, see also 'wtf even is a criminal').
about 3 in 100 americans are in prison, on parole, etc. so if that's the definition of a criminal, you would get 97% accuracy by just guessing not criminal every time