Modifying Turing Machines for General Intelligence: Designing Intelligent Machines

Computer Science is built on the premise that Turing Machines (TM) are sufficient to execute all computable algorithms. However, the computations are intended to be specified by, and performed on behalf of, external thinking agents. Conversely, autonomous systems designed for Artificial General Intelligence (AGI) must, without external help, independently identify and execute appropriate computations to respond intelligently in every situation. In this context, we explore the limitations of the machinery behind the TM specifications. Based on the analysis, we propose a design for Intelligent Machines (IM) that builds upon and expands on the ideas in TM. We argue that the IM design must have two alternating, cooperating processes: a non-TM part that uses its inputs and stored data to generate instructions, and a TM part that uses its instructions to update data and alter the input for the next cycle. This paper builds on the authors’ earlier works on the Mathematics of General Intelligence, Teleological Computations, and the Subjective Emergence of Objective Minds (SEOM) model. In the concluding section of the paper, we discuss the importance of the abstractions that SEOM offers to unify all the different aspects of Intelligence. A promising approach using SEOM to help Large Language Models (LLMs) understand is also discussed.