This directory will contain experiments in achieving deductive intelligence through the application of multilayered neural networks focussed on distinct logical contexts in the distributed knowledge repository.
It will use focal techniques extending the DeepMind alpha-zero methods, viz. neural heuristics for tree search in perfect information spaces. The knowledge repository consists of a heirarchy of such spaces, and therefore can most efficiently be understood with the aid of a heirarchy of focussed neural networks.
It will also develop protocols for collaboration between intelligent focal intelligence agents, enabling subcontracting up the heirarchy from complex theory intelligence to agents supporting reasoning in the simpler theories on which they are built.
At this stage the ideas are embryonic, and my familiarity with the technology non-existent. So I will be trialling and prototyping, which begins with trying propositional theorem proving using alphazero-general in the agprop directory.