The Confidence You Cannot Distinguish
Expertise was never just about knowing things. It was about the friction required to prove it. That friction is gone.
I have spent twenty years and two hundred thousand dollars earning my credentials. A PhD from the University at Buffalo. Postdoctoral work at Harvard. Roughly four hundred books read during my doctoral program alone. A weeklong comprehensive examination. A three-hundred-page dissertation defended publicly before a panel of professors. And on most days, nobody cares.
What people care about is whether you have a platform.
This is the central tension I am investigating in my forthcoming book, Credibility Architecture. Not the tired complaint that expertise is dying (that literature already exists, and much of it is quite good) but the structural question underneath it: what happens when the signals we use to evaluate expertise have fully decoupled from the substance those signals once tracked?
I want to be precise about something before we go further. Expertise is not a credential. It is not a title. It is a relationship between a person and a domain, built through sustained contact and calibrated against resistance. It exists whether or not anyone perceives it. It can be present in someone the world ignores and absent in someone the world celebrates. The appearance of expertise — vocabulary, bearing, fluency, the rhythm of how one speaks about a field — is a separate register entirely. These signals function as proxies because verification is expensive. We all use them. They are not inherently deceptive.
Historically, however, these two registers were held in productive tension. The friction of acquiring the appearance tracked closely enough with the substance that the shortcut worked. You could not easily fake the signals without also doing at least some of the work. That friction is what kept the system bounded.
The drift between appearance and substance is not new. My own doctoral research traced it to the earliest American scholarly institutions. Benjamin Franklin published Gulf Stream maps in the Transactions of the American Philosophical Society, maps based not on his own observations but on the practical knowledge of his cousin Timothy Folger, a Nantucket whaler who would never have been invited to a Society meeting. The credit went to whomever published, not to whomever knew. In thousands of pages from these early journals, the pattern repeated: knowledge generated by women, people of color and the working classes, repackaged and published by the gentlemen whose institutional access permitted it. The gap between expertise and its recognition is a founding feature, not a modern aberration.
What has changed is that the conditions keeping the gap bounded have been systematically removed.
Neil Postman saw this forty years ago, and his analysis remains the most important I have encountered. In Amusing Ourselves to Death, Postman demonstrated that the dominant communication medium of an era does not simply transmit ideas. It shapes which ideas can survive transmission. Print rewarded sustained argument. Television rewarded camera presence, vocal warmth, brevity and emotional immediacy. The expert who could not perform within the medium’s constraints was filtered out regardless of qualifications.
I can verify this personally. Thousands of people have watched my videos. Roughly two dozen have read my substantially better written work. The selection pressure is not on substance. It is on the qualities the medium can carry.
But Postman was writing about television, a medium with human editors, broadcast limitations and at least some institutional gatekeeping. The platform era is a different mechanism entirely. Streaming services eliminated the competition for airtime. Social media democratized broadcast. And the algorithm replaced the human gatekeeper with an optimization function whose only criterion is engagement.
The algorithm does not evaluate quality. It surfaces whatever is best optimized for audience retention and shareability. This is not a change in scale. It is a change in kind.
Consider the case of MrBeast. He has 482 million YouTube subscribers, considerably more people than the population of the United States. He has been transparent about how he achieved this: he and several friends made it their full-time occupation to reverse-engineer the algorithm, A/B testing thumbnails, pacing and camera angles until they found the formula. Clickbait imagery. No more than three seconds without something happening. Relentless visual stimulation.
This is genuine expertise… in algorithmic compatibility. But it has nothing to do with gastronomy (he owns a burger franchise), confectionery (and a candy company) or any of the other domains in which his subscriber count now functions as a credential. Audience scale is load-bearing credibility. A person with hundreds of millions of followers is treated as an authority on whatever subject they choose, because the follower count is the new shorthand for verification. If 482 million other people follow someone, they must know what they are talking about. The audience does not verify the substance because verification is expensive. The existence of a large audience is a free shortcut.
The proxy still functions cognitively. But the thing it points to has changed.
Reach used to require something: capital investment, institutional access or demonstrated quality. Now it requires algorithmic compatibility. The proxy has decoupled from what it once tracked.
And the institutions that should resist this drift have already been captured by it. Universities, publishers, conference programs, hiring committees — organizations designed to certify substance — now factor platform visibility into their decision-making because their own audiences and stakeholders demand it. I have attended conference after conference where the keynote speaker was selected based on Instagram followers rather than the ability to produce a talk composed of substance.
Recently, I listened to a podcast where the host openly admitted she chose her guest — a graduate student discussing AI and intellect — not because of credentials, but because she saw her in a YouTube video. The guest had 216,000 Instagram followers and a bestselling Substack. On paper, I am far more credentialed. So are plenty of others. But we do not have the platform.
This creates a discursive loop. Visibility produces more visibility. The algorithm surfaces what audiences already engage with. The platform expert accumulates compounding advantage that has no relationship to whether their substance has improved. Meanwhile, the expert without platform falls further behind on every metric that institutions and audiences now read as evidence of authority.
There are no corrective mechanisms inside the system. The only pressure comes from outside it… rare moments when competence is tested under conditions the platform cannot control.
Which brings me to what I find most interesting about all of this: the question of confidence itself.
I have identified four distinct types of confidence that, from the outside, look identical. The audience cannot distinguish among them. The signals — fluency, polish, assuredness, the rhythm of knowledge — are produced by all four, but for entirely different reasons.
The first is natural confidence. This lives in temperament. It is almost always tied to class, gender presentation, race and the cultural scripts absorbed before one was old enough to question them. It looks like evidence of competence. It costs nothing to acquire. And if the current state of Hollywood nepotism is any indication, it is the most richly rewarded. Think of Dakota Johnson, who appears in film after film not because of her craft but because her mother is Melanie Griffith and her grandmother is the legendary Tippi Hedren.
The second is earned confidence. This is the only confidence with epistemological standing. It is built through repeated contact with the limits of your own competence — being wrong in ways that matter, failing publicly, repairing what broke. The person who has earned their confidence carries the residue of the failures that produced it. They hesitate where hesitation is warranted. They qualify before they assert. I failed one segment of my comprehensive exam — the first person in my program’s history to do so. I retook it and passed. That failure and recovery is directly responsible for my confidence in what I know. This is not humility. Humility is a posture. Earned confidence is calibration built over years. And the visibility economy selects against it, because the qualities required to earn it — slowness, qualification, visible struggle — are precisely what the platform medium reads as weakness.
The third is conscious fraud. This is the confidence man, an old archetype we now call a “con man” for short. He performs authority he knows he does not possess. He carries the cognitive load of his own duplicity. He tracks his stories. He calibrates. And he can be exposed, because the gap between what he says and who he is is real to him.
The fourth is new.
This is the person who performs authority they did not earn but does not realize it. This is confidence produced by scaffolding, tools that supply the output of expertise without the process that produces it. The person speaks fluently on a subject they have not studied, makes arguments they have not stress-tested, produces work that bears every sign of competence without the substance behind it. Scaffolding within AI conversations bypasses the normal feedback loop that would expose incompetence. Failures are absorbed or masked by the tool. At a certain point, the user cannot distinguish between work they have done and work the scaffolding has done for them.
The loop never closes.
The confidence man knew he was performing. He suffered the cognitive friction of the long con. He could betray himself. The artificially confident person carries no such load. The fluency feels earned. The argument feels original. Unlike the conman, they are not lying. They believe it. They are simply uncalibrated and the tools they rely on will never tell them so.
This is the architecture I am mapping. Not a polemic, not a eulogy. An investigation into the structural conditions that have decoupled the appearance of expertise from its substance and what emerges when that decoupling is complete.
The old system was imperfect. Credentials were gatekept from people who deserved them. The small rooms excluded those who belonged in them. The nineteenth century was the golden age of the conman. I am not nostalgic.
But the frictions that held the drift bounded — the cost of acquiring signals, the institutional gatekeeping, the small rooms where competence was tested under pressure — those frictions are being removed. And the technology that completes this removal does not break the system. It is the logical conclusion of a system that was already selecting for algorithmic compatibility over substance.
The question is not whether expertise still matters. It does. The question is whether the systems we rely on to identify it can still do their job.
I do not think they can. But I would rather investigate than mourn.
Credibility Architecture is forthcoming. Follow along on Substack for excerpts and updates.