Conversation

Notices

1. Don Romano (alt) (thor@noagendasocial.com)'s status on Friday, 03-May-2019 12:50:20 EDT Don Romano (alt)

   The concept of utility functions in A.I. is interesting on a philosophical level.

   A utility function is basically a function that tells a machine learning algorithm if it's moving closer to or further away from a set point. It's the A.I. version of the brain's reward system.

   Humans wouldn't want anything if it weren't for the reward system they are born with. The reward system is what makes us want things. The rest of the brain is just there to figure out how to get those rewards.

   1. Don Romano (alt) (thor@noagendasocial.com)'s status on Friday, 03-May-2019 12:55:47 EDT Don Romano (alt)

      in reply to

      An interesting implication of the fact that many A.I. utility functions are possible is that A.I. could potentially have "life goals" that don't exist in nature.

      So far, our machines have been extensions of the human brain's reward system. That is: We want certain rewards, so we build machines that extend our bodies and help us get them.

      But we could also have machines that are rewarded by entirely different things than us. Perhaps there are much better utility functions.