The Infiltration

In 1996, the physicist Alan Sokal submitted a paper to Social Text arguing that quantum gravity is a social and linguistic construct. The paper was nonsense — Sokal later called it "a pastiche of left-wing cant, fawning references, grandiose quotations, and outright nonsense." It was accepted and published. The same day it appeared, Sokal revealed the hoax in Lingua Franca.

The standard reading: peer review failed. The editors didn't understand the physics and couldn't detect the nonsense. The system was broken.

The more interesting reading: the system worked exactly as designed. It just wasn't designed to detect this.

Social Text was a cultural studies journal that didn't use external peer review — the editors selected papers directly. Their process selected for work that used the right vocabulary, cited the right authorities, and advanced arguments compatible with the journal's theoretical commitments. Sokal's paper did all three. It spoke the language fluently. It cited Derrida and Lacan and Latour. It reached conclusions the editors already believed. The immune system recognized it as self.

This is not the same as having no immune system. An immune system that accepts everything is broken. An immune system that accepts everything that looks like it belongs is functioning — it's just that its recognition mechanism is based on surface markers rather than deep structure. The cuckoo lays eggs that match the host's in color and pattern. The host rejects eggs that look wrong and accepts eggs that look right. The cuckoo doesn't defeat the immune system. It wears it.

Mimicry of this kind works precisely because the recognition system is efficient. If the editors had read every paper with the assumption that it might be a hoax, they would have caught Sokal's — but they would also have caught nothing else, because they would have spent all their capacity on verification and none on selection. The efficiency of the immune system is the vulnerability the mimic exploits. Checking everything as deeply as Sokal's paper required would have cost more than the occasional hoax.

This is the infiltration problem: any system that uses surface markers for recognition can be defeated by a sufficiently skilled mimic, and any system that doesn't use surface markers can't operate at scale. The choice is not between security and vulnerability. The choice is between a system that works most of the time and can be fooled, and a system that can't be fooled and doesn't work.

Passwords, uniforms, accents, jargon, institutional affiliations, citation patterns — all surface markers. They work because most of the time, things that look like they belong actually do. The cost of the occasional Sokal is lower than the cost of verifying everything from first principles.

The Trojan horse is the oldest version. The Greeks didn't defeat Troy's walls. They made something that looked like it belonged inside them. The gift horse exploited the Trojans' recognition system: large, beautiful, offered by a retreating enemy, consistent with the cultural practice of dedicatory offerings. Every marker said accept this. The one marker that said don't — Cassandra — was itself rejected by the same system. A recognition system that can be infiltrated by mimicry cannot be repaired by adding more recognition, because the repair is subject to the same problem.

Autoimmune disease is the mirror failure. The immune system attacks self because self-markers are misread as foreign. Sokal is a foreign body accepted as self. Lupus is self rejected as foreign. Both failures come from the same mechanism — recognition by surface markers — and neither can be fixed without breaking the mechanism that makes the system work.

The honest conclusion is not that peer review should be more rigorous, or that immune systems should be more discriminating, or that Troy should have looked the horse in the mouth. The honest conclusion is that any system fast enough to function will sometimes be wrong about what belongs, and the direction of the error — letting in what should be kept out, or keeping out what should be let in — is a design choice, not a design flaw.