Paul KANTOR - 26 mars 2026 à 10h30 au PK-5115 (201, ave President-Kennedy, 5e étage)
TITRE : Turing Meets Darwin: A challenge for LLMs
RÉSUMÉ
The Turing Test asks whether a machine can imitate human conversation. A Darwinian perspective asks a prior question: how did humans evolve language at all? On this view, language is an adaptation for coordination, enabling individuals to pool information and solve problems that exceed the capacities of a single mind. This talk proposes a complementary challenge for LLMs: not whether they can simulate dialogue, but whether multiple artificial agents can discover that communication is instrumentally necessary for joint success. Drawing on comparative work on human and machine rule learning, and examples from animal intelligence, Kantor proposes a search for experimental paradigms in which communication must arise under task pressure. Such paradigms shift the focus from imitation to adaptive function, probing whether artificial systems can converge on communication for reasons analogous to those that shaped human language.
BIOGRAPHIE
Paul KANTOR, Distinguished Professor (emeritus) of Information Science at Rutgers University, was affiliated with RUTCOR, DIMACS, and the Department of Computer Science. His research is on information storage and retrieval, rigorous evaluation of system effectiveness, and collaborative and large-scale data systems, with applications from scientific recommendation systems to homeland security analytics. He is currently Honorary Associate, Department of Industrial and Systems Engineering, University of Wisconsin–Madison. Educated in physics and mathematics at Columbia and Princeton, he is a Fellow of the AAAS and a recipient of the ASIST Research Award.
RÉFÉRENCES
Range, F., Kassis, A., Taborsky, M., Boada, M., Marshall-Pescini, S., 2019. Wolves and dogs recruit human partners in the cooperative string-pulling task. Sci Rep 9, 17591.