Nalimov's Bayesian Syllogism to Overcome the Barrier of Meaning in AI Systems

Introduction: artificial intelligence is progressing at an unprecedented pace thanks to deep learning algorithms for complex software referred to as artificial neural networks (ANN). State-of-the-art software performs impressively well; however, some cases expose its unreliability. The reason is that AI systems lack the essence of human intelligence and are unable to overcome the barrier of meaning. The emphasis on software and lack of focus on human intelligence mean creating an AI is an underestimated challenge that is relevant to the computer science of today. The goal hereof is to provide theoretical substantiation to, and to further the probabilistic theory of meaning as a human decision support model for AI systems. Methods include theoretical analysis of the probabilistic approach to V.V. Nalimov's meaning problem; C. Shannon's information quantity calculation method. Results: this paper substantiates and describes mathematically the key concepts of the probabilistic theory of meaning quantification, to which end it uses text as a source of evidence; the description follows the cybernetic approach. Besides, the paper proposes a meaning quantifier in the form of differential entropy of the exponential distribution of the entropy difference between two adjacent signs in a sign system; it also proposes a meaning quantification method while defining the meaning in the cybernetic context. The practical significance of the paper lies in the fact that the key mathematical concepts of the probabilistic meaning quantification theory, which are described herein from the standpoint of cybernetics, can be classified together as a probabilistic human decision support model. The results of the study can help create information and control system for decision support functionality of AI. The materials of this paper could be of theoretical and practical interest for AI researchers and developers.