Sounds like you're thinking it through deeply, and that the focus is on the mechanism more than the ML. That's definitely the right place to focus, I wish you t

Sounds like you're thinking it through deeply, and that the focus is on the mechanism more than the ML. That's definitely the right place to focus, I wish you the best of luck! As you note, the big challenges are going to be: 1. Getting people to provide enough (true) information to let your algorithm work without becoming self-referential via pebble spending feedback loops. 2. Incorporating physical image features. Realistically this is your primary source of data. Modern DNNs could do a great job of this but there is a strong chance it'll learn blue-sky-tank confounders like camera quality, framing etc. IIRC that's a big problem with trying to use ML to find good candidates on dating sites but I can't remember where I read this. 3. If it's cheaper for the alpha males to message women than most men, won't that just accelerate the problem existing sites already have? It means most women will end up with inboxes full of messages from the top few percent of men because the others can't afford to message more than one or two per day, which would just reinforce the perception that all men are pigs who aren't serious about relationships.