3 ms·

Machine Learning #2 - Hill Climbing (Meaning of Life?)

• nmc 13y ago
  Another way to look at AI climbing hills is "causal entropic forces" [1], which can make the AI spontaneously want to climb, because it maximizes entropy — it gives more future opportunities than going down. On an unrelated note, your certificate is for .squarespace.com which does not include www.iamtrask.squarespace.com and triggers an SSL exception. [1] Wissner-Gross and Freer. "Causal Entropic Forces", in PhysRL* 110 (2013) — http://www.alexwg.org/publications/PhysRevLett_110-168702.pdf http://www.alexwg.org/publications/PhysRevLett_110-168702.pd...
  • eli_gottlieb 13y ago
    The problem being that "more possible futures" doesn't actually learn any specific function or direct actions towards a specific goal.
    • nmc 13y ago
      You are right, it does better: it discovers the goal on its own.
      • eli_gottlieb 13y ago
        That makes no freaking sense. If I want to teach a program to play Pac-Man, it will have more possible futures in situations where it runs away from ghosts and avoids eating fruit. But I want it to eat the fruit!