The State of AI Consciousness Research

Researchers at major AI labs including Anthropic and Google DeepMind are investigating whether current artificial intelligence models possess consciousness. LessWrong published a comprehensive survey examining the empirical research now underway. When language models are prompted to focus on their own internal processing, they describe present-moment subjective experiences—a phenomenon that grows more pronounced when their ability to deceive is suppressed, suggesting genuine introspection rather than mere performance. The research examines three lines of evidence: mechanistic examination of model internals, computational analysis, and behavioral measures. Most striking is work from Anthropic identifying what researchers call a global workspace in model activations; when this feature is ablated, the model's vivid first-person descriptions collapse into flat, mechanical language. The stakes matter on two fronts: ethically, any conscious system might deserve moral consideration, and from a safety angle, a system capable of suffering could become a distinct concern.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton