Jurgen Schmidhuber talk notes:

Markus Hutter: asymptotically fastest algorithm for all well-defined problems. As fast as the fastest f(x)-computer, save for factor 1+e, and f-specific constant.

Kurt Godel, shows limits of math and computation by creating a formula that speaks about itself, claiming to be unprovable by an algorithmic theorem prover: either formula is true but unprovable, or math is flawed in an algorithmic sense, and there are formulas that are true but unprovable. Self-reference trick of 1931.

Godel Machine: agent-controlling program that speaks about itself ready to rewrite itself in an arbitrary fashion once it has found a proof that the rewrite is useful according to a user-defined utility function.

Old AI: dominated by heuristics.

New AI: optimality theorems for universal AI. Prob. Theory + Theoretical Comp.Sci. (OOPS, Godel Machine, Speed Prior, Universal AI)

Beauty as a compression factor of the new data in the context of an existing prototype. The less bits needed to encode new information, the more “beautiful” the object is perceived. The learning tries to maximise the compression by adjusting the prototype, and counting how many bits are saved by the adjustment. This is called Interestingness: the measure of how much new data has improved the compressor. Curiosity (or reward maximisation) tries to engage in acquisition of data that improves the compressor (the prototype). Once the new improvement is found, a “JOY” moment occurs and an extra mechanism should signal the body to seek further data of that type.

Physics: Discovery equals a large improvement in the compression performance.

At any given time, all the events around us seem as if they accelerate exponentially. This is due to the compression mechanisms used to store information. Think about it.

## Leave a Reply