What Smart AIs Can Do without Consciousness
Journal of Artificial Intelligence and Consciousness March 1, 2026 Anna Strasser
Large language models (LLMs) can perform tasks that in humans require reasoning, planning, and understanding—abilities normally linked to consciousness. However, these abilities might be realized in artificial systems without consciousness. This paper examines whether LLMs solve tasks in fundamentally different ways from humans and whether we can justifiably ascribe agency or socio-cognitive abilities to them. It discusses benchmarks, data contamination, and robustness issues, and uses Daniel Dennett's distinction between competence without comprehension and competence with comprehension to explore whether artificial systems could possess socio-cognitive abilities that fall somewhere between the two. The analysis also addresses general difficulties in attributing abilities, including consciousness, to AIs.