Rigid flexibility. The logic of intelligence. (Q869480)

Rigid Flexibility presents a new theory of cognition. As such it is devoted to the general research for Artificial Intelligence (AI), in contrast to a focus of other AI research that is focusing on solving specific reasoning problems such as planning or learning for example. The author takes an abstract, but precise, level of argumentation that allows him to precisely position and characterize his work in the context of the variety of other positions taken within the diverse AI research community. After discussing that any attempt to establish a scientific theory of intelligence requires to first establish a system of shared beliefs and tasks in a community that is centered around a working definition of what is to be understood by the term ``intelligence'', he presents such a working definition and compares it with other definitions proposed by the AI community so far. He proposes a paradigm shift in AI where intelligence is considered as ``the capability of an information system to adapt to its environment while operating with insufficient knowledge and resources.'' The focus on an information (processing) system that has goals, executes actions, and uses models of the world captured as its beliefs, is not so different from other cognitive models used in AI. The system interacts with the environment through stimuli and responses, denoted often also as experience and behavior, which can be described as streams of input and output information. What makes the approach slightly different, is its strong focus on the ability of the system to continue to function under insufficient resources. At the core of the cognitive model sits NAL -- a non-axiomatic logic represented in a language called Narsese. The logic is built out of 8 layers, where each layer adds new grammar and inference rules. NAL-0 allows one to formulate simple inheritance statements between terms. NAL-1 defines evidence for inheritance relations. NAL-2 introduces sets and extends the notion of inheritance defined in NAL-0. NAL-3 defines set operations on terms, or more precisely on the sets of individuals represented by a term. NAL-4 introduces compound terms and borrows ideas from set operations. NAL-5 to NAL-8 introduce higher-order concepts into the language. Inference control is introduced as a means to enable a NAL-based system to continue to function under insufficient resources. The review of the principles and assumptions underlying the various streams of AI research make this book very interesting to read and enable the reader to easily compare the presented theory with other existing approaches. An application of the proposed cognitive theory to specific problems studied in AI is outside the scope of the book. As a consequence, it is up to the interested reader to identify areas of application and to demonstrate the success of this theory in these application areas.