The project is focused on persistent, cellular automata, self-adapting in an array of partisan contexts [M].

Goals and Requirements:

  1. Consistent strength of automata across the greatest number of contexts.
  2. Automata may utilize only trivial processing and memory.
  3. Goal is “win else survive”.
  4.  Self evolution of independent automata. Code must be self-contained. Automata can improve by optimization and growth.
  5. Gain is in regions of memory. Volume is appropriated and traded between automata in partisan contexts. Functional complexity of automata may be self-limiting via interval, defined as a cycle of finite decision periods.  Size of a given automata is a function of the gamespace, defined in this sense as allocated volume and memory.
  6. Application in non-sequential, non-simultaneous contexts. (Automata are self-limiting per decision speed vs. quality of decision.)
  7. Application in high order contexts. [m>3; m²(m²(m²)); 5(m²(m²)), etc.]
  8. Application in expanded dimensions. (m³, m⁴,…)
  9. Application in cyclic contexts.

Epimetheus, meaning “after+thought”, was the husband of Pandora, and opened the box that turned into a rabbit-hole.  Afterthought is a reference to positional Evaluation Functions, which are choices based on “information derived from the present result of past actions”.

The vast size of even the basic, non-trivial [M] gamespaces makes predictive search suboptimal per processing restrictions. (The gametrees become tractable in the endgame in finite contexts, but the early and mid-game the trees are intractable. Complexity of higher order games is presumed to be greater than Go.)  

Reinforcement of more optimal evaluation functions is predictive only in an inverted sense, making choices based on analysis of the past performance, as opposed to looking ahead.


SEE ALSO: Bounded Rationality