The same is true at the level of societies.
Groups of neurons coordinate through resonance. When timing aligns, signals propagate efficiently. But this coordination remains adaptive only because it is constrained by inhibition and feedback. Prediction errors impose cost. They interrupt momentum. They force revision. When those constraints fail, pathology emerges. Seizures are not caused by too little activity, but by too much synchronization without restraint.
Modern social systems are drifting toward a similar failure mode.
Errors are detected constantly. False claims are flagged. Harms are documented. Contradictions are exposed in real time. Yet the system often continues unchanged. This is not because feedback is absent, but because it no longer bites.
Three patterns recur.
First, errors are detected but ignored. Investigations reveal misconduct and misinformation, but amplification mechanisms remain untouched. Attention continues to flow. Outrage continues to reward alignment. The system absorbs error without adjusting course.
Second, feedback exists but arrives too late. By the time correction surfaces, narratives have hardened and identities have aligned. Positions become social signals rather than hypotheses. In neural terms, the correction arrives after the window of plasticity has closed.
Third, correction signals are drowned by amplification. High-gain signals—fear, certainty, moral outrage—overwhelm slower, lower-amplitude corrections. Algorithms reward speed and engagement over revision. The field synchronizes around error, not because people refuse to think, but because the system privileges alignment over learning.
In this state, discourse doesn’t fail because individuals are irrational. It fails because thinking no longer changes outcomes.
This is why restoring balance is not primarily a matter of better arguments or more information. It is a matter of restoring friction where feedback can impose cost.
This is also where AI enters the picture—uncomfortably but unavoidably. There is growing hope that AI can function as institutional memory or neutral referee: preserving long-term outcomes, tracking consequences, and slowing collective amnesia. But this immediately raises the harder question: who trains it, who governs it, and whose errors it is allowed to surface?
A referee that cannot impose cost is decorative. A memory system that reflects only existing power is not corrective—it is stabilizing drift.
If AI is to help rather than harm, it must be designed not to optimize away error correction, but to amplify it asymmetrically: to slow amplification, privilege correction, preserve consequence, and resist the constant pressure toward immediacy. These are precisely the functions humans are worst at sustaining alone, especially at scale.
The lesson from both neuroscience and history is the same: error is cheap; correction is costly. Friction is the price a system pays when it chooses to learn.
When feedback stops biting, coordination creates a dangerous field pushing mass towards a black hole.
