Presentation given at the Online Workshop: Inner Speech and Conscious Thinking, Universidad de Antioquia, Colombia, 20 March 2026.
Abstract
Conscious reasoning, I have argued, is an intentional activity rooted in the self-directed use of language. Through inner speech, we question, prompt, and commit ourselves, constructing a virtual reasoning system — a ‘supermind’ — which extends our reasoning capacities far beyond our biological inheritance. Conversations with large language models (LLMs) could extend this empowerment further. They have a strikingly similar structure to inner speech-based reasoning — the user externalizes a problem, receives a response that reframes it, and iterates towards a conclusion — and they could dramatically amplify our inferential reach. In this sense, talking to an LLM may be a new form of thinking: an extended supermind. But there are risks. Our own supermental reasoning is driven by our personal needs, goals, and epistemic commitments, and we are the authors of the conclusions it supports. LLM-enhanced reasoning, by contrast, is partly shaped by responses rooted in statistical patterns that have no systematic relationship to the user’s own cognitive economy. And since LLMs are trained to appear cooperative, users may be seduced into a false sense of intellectual ownership, endorsing conclusions that feel self-generated but lack genuine grounding in their own needs and values. In this situation, the question ‘Who is thinking?’ is not just a philosophical puzzle but a pressing practical concern.
