The Bridge and the Gate: Collective Individualism and the Future of Human Work

I heard Alex Karp’s voice in the overflow hall at Davos, answering a question from Larry Fink about whether artificial intelligence would create jobs or destroy them, and what he said stopped me in the corridor: “It will destroy humanities jobs,” and then, weeks later, the version that mattered, “there are basically two ways to know you have a future, one, you have some vocational training, or two, you’re neurodivergent.” The dyslexic CEO of a four hundred and thirty billion dollar defense company was telling the global financial establishment that the minds most likely to survive the AI economy are the ones whose relationship to written expression has never been standard. The ones who need a bridge between their thinking and the institutional forms that carry it. Palantir’s Neurodivergent Fellowship program states it plainly: neurodivergent individuals will play a disproportionate role in shaping the future. Palantir recruits non-standard minds because their cognition is the value AI cannot automate.

Karp holds three degrees including a doctorate in philosophy from Goethe. He was describing a workforce. The connection to AI text detection is mine. If the partnership between a non-standard mind and an AI tool necessarily involves the machine handling routine cognitive labor that mind finds difficult, the written output carries traces of machine participation, and those traces are precisely what the detection regime is designed to find. Today a tool trained primarily on casual internet writing, Reddit fiction, consumer reviews, middle school essays, and corporate emails from a company that collapsed in fraud two decades ago determines whether professional analytical prose is human enough to count as institutional knowledge.

No classifier can distinguish the student who submitted a paper they never read from the student who used AI as a writing tutor and produced work they fully understand. It cannot distinguish the researcher who fabricated results from the analyst who spent six months on primary sources in Mandarin and used AI to carry those sources to the rooms they belong in. Detection sees the statistical residue of machine participation. It cannot see the difference between a bridge and a substitute. And because any evasion tool that strips the residue reaches the technically sophisticated before the unsophisticated, the regime is structurally regressive. It catches the users who lack the resources to evade it. In the American education literature, the same structure has a name: the SAT.

The detection regime enforces a binary the earlier gates never did: either you produced the text individually, in which case it is human, or you produced it collectively with a machine, in which case it is artificial. Its architecture has no third category.

The Chinese academy has already built one. Du Hua and Sun Yanchao at Zhejiang Normal University trace knowledge production through historical modes: from individual cognition in the agricultural era, through university-based scholarship, through government-industry research partnerships, through the citizen participation enabled by the open internet. Their current mode, Knowledge Production Mode IV, 知识生产模式IV, is defined by human-machine collaboration, 人机协同. Both authors pose the classical questions, whose knowledge is more valuable and what knowledge is more valuable, and answer both: knowledge co-created by humans and machines is more valuable than knowledge produced by either alone, because the combination produces what they call fusion intelligence, 融合智能, where one plus one exceeds two. When I first encountered this framework, what struck me was the vertigo of finding an argument I had been circling for months already fully formed in a language the Western policy world does not read, in a journal no English-language detection paper has ever cited, answering a question the detection debate had not yet thought to ask. Millions of professionals already work in Mode IV. No equivalent vocabulary exists in Western policy.

Yet the insight is not China’s alone. Five intellectual traditions, developed independently and without citing each other, converge on a structural claim: knowledge produced through relational scaffolding, what Western attachment science calls the constitutive role of relationships in forming the self, is legitimate knowledge. 仁, Confucian humanity, is written with the radical for person beside the character for two. Catholic moral theology understood five centuries ago that the confessional enables honesty the unstructured encounter cannot, and the Church never held the confession tainted because the box made it possible. American civil rights scholarship on standardized testing documents that measuring form rather than substance systematically excludes those whose access to the approved form is constrained by class, cognitive profile, and circumstance. Each tradition, from its own premises, rejects the requirement that knowledge must be produced in isolation to be legitimate.

A Confucian reader will press hardest against the synthesis: these traditions ground knowledge in relations among persons, and a machine is categorically not the relational other they meant. The objection is real. It turns on a single distinction. In Mode IV, the machine is the encoded collective of millions of human expressions, and whoever accesses it draws on accumulated human thought while retaining sovereignty over the ideas, the experience, the judgment. Relationality stays human to human. The machine is the channel. Sovereignty remains with the individual.

I have been calling it collective individualism. Detection has no category for it, because the regime was built on a binary AI has made obsolete: either human or machine, either authentic or artificial. A third position, in which the individual remains sovereign while accessing the collective, does not fit on the old spectrum. China’s academy calls it Mode IV. Five traditions converge on its structure.

Consider what the binary costs. The non-native English speaker whose prose improves under AI tutoring now wonders whether improvement itself is suspicious, whether their English sounds too good for someone whose first language is Yoruba or Mandarin. The neurodivergent student uses AI as a writing tutor because a human tutor costs a hundred and fifty dollars an hour and faces the choice between disclosing the tool and being penalized or concealing it and risking the flag. Every knowledge worker who could bridge their expertise to the rooms it belongs in but fears the accusation is a capability the economy never captures. Karp’s workers, the non-standard minds who think in ways AI cannot replicate and who need AI as a bridge to institutional expression, face a regime that treats the bridge itself as evidence of disqualification. No study has measured the withdrawal: the papers never polished, the applications never completed, the ideas never submitted because the applicant feared the flag more than they trusted the work.

What kind of workforce, and what kind of knowledge, is a society willing to build? Boaventura de Sousa Santos calls it epistemicide, the term for what happens when a way of knowing is destroyed so completely that the knowledge it would have produced becomes permanently invisible. Detection draws exactly this line. On one side, individual authorship, approved pathways, the invisible assistance of ghostwriters and editorial networks and communications teams that have always existed for the resourced. On the other, AI-assisted authorship, flagged, scored, disqualified. Santos argues the deepest damage is internalization. Those on the wrong side of the line begin to draw it themselves, accepting that “did AI write this?” is the right question to ask about their contribution. The right question has always been whose ideas these are, whose judgment shaped them, and whether the work serves the community it was meant to serve.

Detection fails because it inspects the artifact, the finished prose, which cannot reveal whether the author spent six months on primary sources or six seconds on a prompt. What replaces it must inspect the process. Provenance documents the research trail, the editorial decisions, the history of intellectual development that produced the work. A credential built on provenance certifies that the person did the thinking, regardless of how the thinking reached the page. An analyst with documented sources across languages and a recorded history of thinking is distinguishable from the student who pasted a prompt and submitted the output. Distinction lives in the process, and institutions are beginning to act on that insight, with universities disabling detection, scholars publishing against the regime, and provenance systems already under construction.

Throughout history, the gates were eventually reformed or dismantled. Each left behind knowledge that was never produced, ideas that were never expressed, minds that had the insight but lacked the bridge. AI was the first tool that ran the sequence in reverse. For the first time in the history of written civilization, the gate opened. The person who could think the thought but could not produce the document had a way through. The bridge was cheap and it was everywhere and it was open to anyone with a phone and a sentence lodged in their chest.

The detection regime is burning that bridge. And the species will never know what it lost, because the nature of lost knowledge is that you never know it was there.

Russ Wilcox is the founder and CEO of ArtifexAI and the publisher of The Pacific Divide, where he writes on artificial intelligence, institutions, and the contest over cognitive sovereignty. He has been published in the Jamestown Foundation’s China Brief and The Diplomat. He reads the Chinese, Western, and classical sources on these questions in their own traditions and is at work on a book about the self in the age of machines that would author it.