Use machine learning to help compute decisions

[ck2qsuZT]

could this be applied to this?

https://www.youtube.com/watch?v=TnUYcTuZJpM

I wonder if people would let robots vote for them

[mitar]

For that we would first need more data. Much much more. But yes, that could also be an interesting approach. One downside is that people probably want to know how the decision was being computed, and machine learning is often a black box and hard to explain.

[ck2qsuZT]

Alot of people don't mind black box magic, as long an they can ask questions

On May 4, 2017 8:46 AM, "Mitar" [email protected] wrote:

For that we would first need more data. Much much more. But yes, that could also be an interesting approach. One downside is that people probably want to know how the decision was being computed, and machine learning is often a black box and hard to explain.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/peer/mind/issues/210#issuecomment-299226053, or mute the thread https://github.com/notifications/unsubscribe-auth/AHwZSsZo0ojbkRUSqqE5nSs2gdkT84aEks5r2fLUgaJpZM4NQzje .

[mkanwal]

Would not recommend. There are a number of issues with applying current state-of-the-art machine learning methods in situations that involve careful ethical/moral decisions. e.g., see the intro in the talk posted in #30.

Read this fantastic article.

(In fact, one of the biggest issues, which you hinted at, is that many methods are too black-box and nobody really knows what's going on.)

[mitar]

So yes, this is like far future idea. And even more research-based than others. But yes, one reason why we are asking to collect data from users is to be able to look into if something like this is possible. But this is for me more a 10 year plan if at all. (Still not sure from other implications as both @mkanwal and I commented.)