AGI is unlikely to be a monolithic, static product. It is more plausibly an asymptotic, self-modifying system of breaking closure — closer to metabolism than memory.

Essays here: yuancao.me

5. The North Star
Intelligence is ultimately measured by innovation - the ability of creating novel conceptual space - not just performance on closed tasks.

4. Breaking Closure
Stronger intelligence requires mechanisms for recursive self-reflection that transcend fixed formal boundaries.

3. The Missing Engine
Existing models are powerful pattern matchers, but they lack Immanuel Kant’s "a priori" drive to organize experience and create meaning autonomously.

2. The 3 A's of Intelligence
Intelligence minimally requires Association, Abstraction, and Analogy, capacities that cannot be fully reduced to static cosine similarity.

Over a series of essays, I explore this from multiple angles: why intelligence must be open-ended, why current AI systems are structurally closed, and what stronger models would require.

1. The Computability Boundary
Computation has hard limits (Kurt Gödel, Alan Turing). AI inherits them.

True intelligence is an open-ended, self-referential process. It breaks closure through abstraction, analogy, and innovation.

Current AI systems are intrinsically closed. Their universe is defined by fixed representations, objectives, and formal rules. They are "locked" inside the statistical manifold of their finite training data.