One of the more useful ideas in the consciousness debate is attention schema theory. Neuroscientist Michael Graziano argues that awareness is a biological system’s simplified internal model of its own attention. This self-model helps the system regulate what it attends to next.

Importantly, awareness is not treated here as something outside the machinery of the system. It is part of the machinery, a causal self-model that helps monitor and direct engagement with the world. A schema in this view allows a biological system to register when attention has drifted and to bring it back to the task or goal at hand.

This offers a helpful way to think about today's AI systems. Current models do have attention mechanisms in the technical sense through transformer architectures, as well as extended reasoning. They can manage context, refer back to earlier material, generate plausible self-descriptions, and in agentic settings plan and sequence actions across multiple steps.

But these capabilities do not by themselves establish an attention schema. Research has found limited and unreliable forms of introspection in some models, including some ability to monitor and influence their own internal states. This does not establish consciousness or a persistent model of attention with an ongoing regulatory role. For that, the system would need a schema of its own attentional state that helps govern later attention over time.

On this account, such a self-model would be important to self-agency in the human sense. A persistent attention schema of this kind has not been established in current mainstream AI systems. If attention schema theory is correct, this remains a useful distinction to examine: attention as a computational mechanism, and awareness as a causal self-model.

Michael S. A. Graziano, Rethinking Consciousness: A Scientific Theory of Subjective Experience

Anthropic, Signs of introspection in large language models https://www.anthropic.com/research/introspection