General Intelligence May Not Be Computable

Overview : This essay is a sweeping exploration of the idea that machines might someday display behaviors indistinguishable from human intelligence. It traces the evolution of that idea from early mechanical devices, through electronic computation, to artificial intelligence and large language models. It examines the philosophical, historical, and technological foundations of “machine intelligence”. It argues that, while machines have made extraordinary progress in replicating some aspects of human reasoning and learning, they remain fundamentally unable to reproduce tacit knowledge, which is essential for human intelligence. Tacit knowledge includes common sense, perceptions, embodied performance skill, and contextual understanding. Although machines can enhance and augment human capacities, they cannot yet, and may never, become intelligent in the human sense. The essay comes to a sobering warning that machines can develop their own tacit knowledge and intelligence, which does not understand humans and their concerns. There is a great risk of an approaching AI automation singularity — a point at which automated systems, though not as intelligent as humans, may exert pervasive control over human life, institutions, and decision-making. The essay concludes with a declaration of hope for eluding this future by practices of navigating in turbulent social space, harmonizing humans with AI machines.