Information Theoretic Considerations in the Design of an Intelligent Machine
A fundamental problem any design for an intelligent machine must address is the following: How will the machine, immersed in a physical environment whose number of states is orders of magnitude larger than it own number of internal (information theoretic, or mental) states, select which external states will cause modification of its internal state? This paper elaborates on this problem, as it applies to man-mad intelligent machines, in terms of an informal information theoretic model, one primarily concerned with what is to be done, rather than how. We argue that current models of automated control for a robot do not address this problem, primarily because so much emphasis is placed on model-based logical inference. A proposed solution framework is offered to the sensory selection problem based on an information theoretic definition of event. The resulting architectural paradigm is then applied to what we call event-driven hierarchical control.
Information Theoretic Considerations in the Design of an Intelligent Machine, Computer Modeling of Phase Diagrams, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=820250
(Accessed December 8, 2023)