From Cephalopods to Large Language Models: Conceptions of Intelligence and Reasoning

Soumya Banerjee

Keywords: conception of intelligence and reasoning

Abstract: This paper explores convergences and contrasts between conceptions of intelligence and reasoning in three domains: cephalopods (principally octopus and cuttlefish), other marine animals (notably cetaceans), and contemporary large language models (LLMs). We use comparative biology and cognitive ethology to highlight different embodiments of information processing, problem-solving, and flexible behaviour, then draw instructive analogies for artificial intelligence research. We argue that cephalopods and other marine animals instantiate forms of intelligence that emphasize embodied, distributed, and context-sensitive problem solving; by contrast, current LLMs implement a disembodied, statistical-syntactic form of competence that nonetheless achieves surprising forms of emergent reasoning. Examining causal mechanisms, ecological pressures, and developmental trajectories that shape these intelligences reveals lessons for designing AI systems that are robust, adaptable, and socially situated. We suggest that reasoning and intelligence lie on a continuum from cephalopods to humans, and on to AI systems.

Submission Number: 72