Provides a unified view on Information theory, Bayesian probability theory and machine learning.

Convinced me that Bayesian inference over hierarchies of flexibly structured representations is a good model of human inductive learning.

A two-dimensional Turing machine with a very simple set of rules but complicated emergent behavior.

"If such a universal [learning] algorithm exists, many variants likely also exist, some of which may be much more appropriate for implementation in silicon. This STTR is seeking proposals that identify and ultimately implement the best of these variants."

The machine has 2 states and 3 colors, and is 596440 in Wolfram's numbering scheme. If it is universal, then it represents the smallest possible universal Turing machine. Is it universal?