Towards Formalization of Constructivist Seed AI

Abstract

A general intelligent system is expected to solve wide range of problems and adapt across multiple varied environments. It has already been hypothesized that seed AI needs to be bootstrapped in a system which could evolve to handle multiple problem domains. Such a seed AI may consist of few core intelligent components and a seed program. However, there is no formal structural definition of seed AI in place. Thus, an abstract model of the seed AI is presented and its generality has been proved. The formal structure of such an algorithm has also been derived. It has been discussed how and in what setting the proposed model of seed AI can achieve different properties of an intelligent agent like adaptability, constructivism, learnability and recursive self-improvement. The proposed theoretical framework for seed AI can serve as the basis for construction of any artificial general intelligent system. A prototype of seed AI has been developed using universal search to demonstrate and present a guidance on physical implementation of an agent-based system. The agent has been experimented in a heterogeneous toy problem to illustrate its usability.