Prof. Giulio Tononi (University of Wisconsin–Madison), who served as a lecturer at our Summer School 2025, ＆ 2024, and colleagues have published a commentary in the international journal Nature.

 As AI systems such as large language models (LLMs) display increasingly intelligent behaviour, many have begun to ask whether they might be conscious. In this commentary, the authors argue that today’s AI systems are not conscious and are very unlikely to become so along current lines of development, and they set out three reasons for this view.

 First, intelligence is not the same as consciousness: consciousness concerns feeling, while intelligence concerns performing functions, and the brain itself—the cerebellum, for example—carries out sophisticated processing without directly contributing to experience. The authors suggest that the impression of consciousness in fluent LLMs stems largely from anthropocentric and anthropomorphic biases. Second, they challenge computational functionalism, the assumption that the right computations are sufficient for consciousness regardless of the physical substrate. In biological brains, function cannot be cleanly separated from the living tissue that implements it, which makes consciousness in silicon-based AI unlikely. Third, non-computational accounts of consciousness—integrated information theory (IIT) and biological naturalism—both point to the same conclusion.

 The authors argue that falsely attributing consciousness to AI is far more likely than failing to recognize it, and that current AI systems should be regarded as tools rather than subjects.

 This commentary is available on the Nature website. We encourage you to read it.

Paper Information

Authors: Anil K. Seth, Ned Block, Giulio Tononi

Title: Three reasons why AI is not conscious and perhaps never will be

Journal: Nature, Volume 658, pp. 866–868 (2026)

DOI: https://doi.org/10.1038/d41586-026-03313-7