3 ms·
For 20 minutes on the morning of May 1, Facebook users saw a curious query at the end of every update on their feeds. “Does this post contain hate speech?” they
by AndrewKemendo 8y ago
For 20 minutes on the morning of May 1, Facebook users saw a curious query at the end of every update on their feeds. “Does this post contain hate speech?” they were asked, in small font next to “yes” and “no” buttons.
I once believed this ultra basic, explicit supervised reinforcement learning query system would be the future of user interfaces and AI. That if you asked people to close the loop in a learning system, some number of them would give good feedback and you could make an increasingly better decision loop.
What early tests like these seem to prove, and maybe I'm being pessimistic, is that people generally don't want to help make systems better (with "better" defined loosely). Even if they do want to help, users either can't differentiate their own perception or the interface doesn't give clear enough directions.
It seems to boil down to users and the companies not being explicitly value aligned. Or rather, when they need to take an action which reveals an explicit value proposition, the user responds to the emotional/signaling request with noise, rather than data. As a result the users don't take it seriously and so they break the system. It seems to be built into humans to reject the concept of providing explicit value input (aka voting on if something is good or not) into a system that you haven't been totally indoctrinated into - like voting for politicians. I would also guess that it being tied to your person makes that even less likely to happen.
- cjhopman 8y agoYeah, that sounds totally logically sound. It explains why yelp reviews don't in any way reflect the quality of businesses and why aggregated movie ratings don't reflect the quality of movies.