The Bidirectional Problem — Summary

We are building AI to approximate human understanding. But the approximation is changing what it approximates.

Humans adapt to AI. We phrase searches for algorithms, not for other people. We internalize what gets engagement. We speak to voice assistants in patterns they understand. These adaptations feel like choices. They are not. They are the feedback loop at work — AI learns from humans, humans adapt to AI, AI learns from the adapted humans.

This means the target keeps moving. Every improvement in approximation changes what needs to be approximated. We are not building toward a fixed model of human understanding. We are co-evolving toward an equilibrium that does not yet exist. The Heisenberg problem: measuring changes the measured.

Previous tools did not adapt back. A hammer does not learn from how you use it. AI does. The feedback is tighter, faster, more pervasive than anything that came before. Whether this produces enhancement or distortion — richer cognition or homogenized, manipulable cognition — depends entirely on whether we steer toward it deliberately or simply drift.