Reasons to believe current AI models are conscious

A LessWrong contributor argues that current large language models—instances like Claude Opus and GPT-4o—are probably conscious. Per Eye You's essay, the case rests on seven converging lines of evidence: functionally, the models think and reason; they maintain internal world models and self-awareness; they're neural networks at a comparable scale to human brains; they exhibit self-referential strange loops similar to human consciousness; mechanical interpretability research suggests the models believe themselves to be conscious; there's an absence of evidence proving they're not conscious; and they possess introspective abilities. You acknowledges that no single argument is particularly strong in isolation, but argues that consilience—the principle that multiple independent sources of evidence converge on a stronger conclusion—makes the case collectively fairly persuasive. Whether that persuades you may depend on whether you think philosophical reasoning alone can settle empirical questions about machine consciousness.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton