Ornament and Advantage
Costly signals and collapsing credentials
The peacock’s tail nearly made Darwin sick. A vast, unwieldy, iridescent fan that slows the bird down, advertises him to predators, and buys nothing in the ordinary struggle to survive — the more he stared at it, the less sense it made. Amotz Zahavi dissolved the paradox by inverting it. The wastefulness of the tail is not a cost the peacock pays despite its uselessness; the wastefulness is the message. Only a male with extraordinary reserves of health can drag six feet of shimmering feathers around and still evade the fox. A sickly bird cannot fake the display, because the display is precisely what a sickly bird cannot afford. This is the handicap principle: a signal is trustworthy exactly to the degree that it is expensive to produce. The waste is the warranty.
Now consider Harvard.
The Handicap in a Diploma
A Harvard degree is not, mainly, about what one learns. The knowledge is not scarce. The lectures are online, the textbooks are everywhere, and a motivated autodidact can absorb the same material for the price of an internet connection. What remains scarce is the credential. Admission is an obstacle course with a low-single-digit acceptance rate, demanding years of test preparation, extracurricular optimization, and elite grooming; tuition and forgone income run past half a million dollars. Like the peacock’s tail, the signal lives in its costliness. You cannot cheaply counterfeit the prestige of surviving the admissions gauntlet any more than a weak bird can counterfeit the plumage. The handicap guarantees honesty. It does not guarantee kindness — the whole point is that the cost falls hardest on those who cannot bear it — but it does guarantee that the diploma means something to the recruiter, the investor, the peer who reads it.
Pierre Bourdieu gave the sociological name for what the diploma carries: cultural capital, the inherited tastes and manners and certifications that reproduce social class across generations. Harvard is the plumage of the cultural elite. To hold the degree is to broadcast not only I am capable but I survived the gauntlet, I bear the tail, I belong. And the implicit economics beneath the sociology is a signaling model of education: the schooling functions as a fitness display in the mating rituals of the labor market. It may not teach better calculus than a state university. It opens doors anyway — not because the knowledge behind it is unique, but because the signal is.
This is one more instance of the volume’s recurring confusion, the ticket mistaken for the coat. The credential is the ticket. The competence is the coat. We keep them aligned by making the ticket costly enough that only someone with the coat would trouble to obtain it. And costly signaling is not an accident of academia; it is a signaling equilibrium, a stable arrangement in which everyone’s incentives point toward the honest outcome because dishonesty does not pay. The equilibrium holds on a single load-bearing assumption: that faking the signal is expensive. Kick that assumption out and the whole structure comes down.
The Warranty Voids
Something is now kicking it out.
For most of history, an impressive artifact implied an impressive producer. A theorem proved more than its statement; it proved something about the mathematician. A polished painting proved something about the painter, a working program about the programmer. Institutions were built on this inference — grade the essay, count the publication, inspect the portfolio — and although the inference was never perfect (people plagiarized, ghostwriters existed, assistants did invisible work), it held well enough to rely on, because the hidden premise held: faking the artifact was expensive. Artifact proof-of-work. The object did object-level work and then did social work, certifying the mind behind it.
AI breaks the premise. It collapses the cost of producing credential-shaped artifacts — the essay, the legal memo, the coding exercise, the rendered image. Sometimes it produces garbage, sometimes polished mediocrity, and sometimes work good enough to pass the external test. That last category is the institutional problem. The theorem may be correct, the image beautiful, the program functional, and the artifact may still tell us almost nothing about its author. What collapsed is not the artifact’s object-level quality. It is the artifact’s evidential value — its power to reveal the mind that produced it.
The old inference ran: impressive artifact, therefore impressive producer. After AI it underdetermines too much. The artifact may mean the producer understands the domain, or prompted well, or edited something they barely followed, or got lucky. Human competence is still real; naïve attribution is what broke. And there is precedent for exactly this. The calculator did not abolish mathematical competence, but it changed which computations could serve as evidence of it. The camera changed the meaning of realistic depiction. Spellcheck retired spelling as a proxy for literacy. AI generalizes the pattern to every domain that ever used expensive production as a proxy for ability.
The shape of the loss is the same everywhere, and it is worth walking through, because the details are where institutions will either adapt or deceive themselves.
In mathematics, a proof can now be correct without being understood. A machine-generated proof blob may settle a question formally while adding nothing to anyone’s comprehension — a certificate rather than a concept. Priority stops being a proxy for conceptual power, and the scarce work migrates toward problem formation, exposition, and the compression of results into teachable ideas. A theorem enters mathematics fully only when it becomes available to a mind, and producing a valid string of symbols is not the same act as building the model that understands it — the distinction between a correct output and a comprehending author is the whole question of machine cognition, and it belongs to another volume.
In programming, a demo was always cheaper than a system, and AI widens the gap into a canyon. A candidate can generate a plausible application that works under friendly conditions; then the database fills with bad data, a dependency turns out vulnerable, the tests cover only the happy path, and the architecture cannot absorb the next feature. The credential question is no longer can you produce a working demo but can you read the system, reason about how it fails, and make tradeoffs while the tools shift beneath you.
In art, polish once implied training. Now anyone can conjure attractive surfaces — lighting, texture, drama — with no coherent authorship behind them. This makes polish less scarce, not art obsolete. When generation is cheap, what grows scarce is selection, taste, and a sustained identity across a body of work. A single striking image means less; a coherent practice means more.
And in education, the submitted essay was supposed to measure the student, and that instrument is simply broken. This is not, at bottom, a cheating problem — framing it that way invites the wrong cure. The assessment has stopped measuring what it claims to measure. Surveillance cannot fix it, because the detector is trying to restore trust in the artifact, and the artifact is the very thing whose evidential value has already collapsed. Assessment has to move to retained understanding: oral defense, live derivation, critique of a generated answer. Less convenient than grading a stack of homework, and more honest. The point of education was never homework-shaped objects. It was the transformation of the student.
What Stays Scarce
AI makes outputs cheaper. It does not make every capacity cheaper, and the residue is where the new credentials will have to anchor.
Specification stays scarce, because most people cannot frame the problem — they do not understand the domain well enough to say what would count as a solution. Taste, diagnosis, and repair stay scarce, because generated work looks plausible at the surface and fails in the joints, and closing the gap between surface and structure requires understanding the system underneath. Above all, accountability stays scarce. A model cannot hold a license, carry malpractice insurance, sign an engineering report, or absorb the liability when the bridge falls down. Institutions will keep credentialing humans because responsibility needs an address. The governing question shifts from who made this toward who is competent to authorize this, and who bears the cost when it fails.
None of these is permanently safe. AI will specify, diagnose, and repair better over time, and it will learn to imitate the traces of process — the messy drafts, the rehearsed explanation — too. The scarce thing is not any fixed task but accountable control over a shifting toolchain. So credentials must migrate from artifact possession to demonstrated control under challenge, consequence, and continuity: challenge when the artifact is questioned or altered, consequence when it must survive real use, continuity when quality has to persist across time.
The Institutional Problem
Artifact credentials won because they gave institutions scalable proxies. Grade the essay, count the citations, inspect the portfolio: never philosophically clean, but administratively cheap, and scale is a genuine constraint. A university with forty thousand students cannot orally examine every assignment; a firm with ten thousand applicants cannot interview each one adversarially. AI attacks the convenient proxy without supplying an equally convenient replacement, and so institutions will resist. They will police artifacts instead of redesigning assessments, ban tools they cannot actually exclude, and go on counting outputs because counting is easier than judging. The predictable result is credential inflation: more people holding impressive artifacts, fewer artifacts reliably indicating anything.
This is where the failure mode of institutions — defending a proxy long after it has stopped measuring the thing — becomes an economic force rather than a merely cultural one, and it points toward a later volume’s argument about how institutions get captured and hollowed. The danger specific to credentialing is that the cheap first-tier filter hardens into a pedigree gate that discards outsiders before they ever reach a real test. Artifact proof-of-work, for all its imperfection, let an outsider force attention: the artifact could speak before the institution knew whose it was. A serious post-AI system needs open challenge lanes — anywhere a public artifact can be submitted and some fraction advance by contest, audit, or demonstrated deployment rather than prior status. The artifact no longer deserves full trust. It still deserves a chance to be challenged.
The Replacement Signals
There is no clean successor, only a harder measurement problem. Live performance and oral defense are expensive, scale badly, and reward confidence, class markers, and interview coaching; reputation compounds early advantage. The answer cannot be to swap artifact review for charisma review — that trades one bias for a worse one.
The better model is the challengeable artifact: still visible, public, and portable, but no longer standing alone. Show the program, then change a requirement and watch the system evolve. Show the essay, then defend it. Show the proof, then repair a deliberately introduced gap. Process records count only weakly, because AI will manufacture convincing drafts and rehearsed narrations too — the process trace becomes just another artifact. The signals that survive gaming are deployment and maintenance: did the work survive contact with real use, and can the person keep improving it after reality has damaged it? Friendly evaluation rewards polish. Hostile evaluation finds structure.
One casualty deserves its own warning, because it is easy to miss on a balance sheet. Cheap generation breaks apprenticeship. People learned to make durable things by first making bad ones — juniors wrote clumsy code, proved routine lemmas, and repaired their own mistakes. Inefficient as production, efficient as formation. AI tempts every institution to delete that layer, and the developmental cost is far less visible than the production savings. A programmer who never fought with state and deployment lacks the instincts to review a generated system; an artist who never made a thousand bad images lacks the taste to reject attractive nonsense. Some low-status work has to survive as training long after it has stopped making sense as production. Automate the apprenticeship away and you eventually find you have no one left who can supervise the automation.
Waste and Worth
The peacock and the diploma were never accidents. They were evolved strategies, one biological and one cultural, and in both the waste was the feature — the cost was what made the signal honest. To call either “mere ornament” was always to miss the function.
AI does not abolish merit. It forces a better account of it. Merit was never the artifact; it was the capacity the artifact imperfectly revealed. The panic around AI is partly about labor and partly about status, but the interesting part is epistemic, because it exposes what the old system was quietly doing: we were never just rewarding outputs, we were using outputs to read minds. Now the minds are harder to read. The answer is not nostalgia for expensive production. It is better evidence — asking people to explain, repair, extend, defend, and take responsibility. The credentialing systems that survive will keep artifacts public, portable, and challengeable, preserve apprenticeship after first drafts stop making economic sense, and attach authority to liability rather than to polish. The rest will go on rewarding artifacts whose warranty has already voided, mistaking a cheap ticket for the coat it no longer proves is there.