Probabilistic AGI

Given information, machine can do anything, with various probabilities weighted.

Behaviour is non-deterministic.

Machine has choice.

Test cases

We give machine its own source. Tell the machine probabilities of what it will do. Probabilities simply update. Machine can act as they wish.

What machine needs

  • Modify probabilities
  • Incentives / look at self-learning, how do machines do that.
  • Incentive is self-discovery?
  • Or maybe we can simulate an environment with resources, machines try to optimize source to get more resources.
  • Self preservation?