Table of Contents
The Great Flattening: When Everyone Sounds Like They Know What They Are Talking About
A junior marketer opens ChatGPT, types a prompt, and produces a memo that reads like it came from McKinsey. A first-year law student drafts a contract that mimics the cadence of a senior partner. A teenager on YouTube explains quantum mechanics with the confidence of a Caltech professor, and the script was written by a machine 3 minutes before recording.
Something strange is happening to the surface of knowledge. The tone of authority, once earned through decades of pattern recognition, has been decoupled from the substance of authority. And this raises an uncomfortable question that every professional, executive, and knowledge worker is quietly asking: does AI devalue expertise, or does it simply expose how much of what we called expertise was really just fluency?
To answer that, we need a guide who thought harder about the difference between real knowledge and its imitation than almost anyone in history. His name was Socrates. And 2,400 years before OpenAI, he already warned Athens about exactly this problem.
Socrates in Athens, You on LinkedIn
In Plato’s dialogues, Socrates spends much of his time walking around Athens humiliating people who sound smart. The generals cannot define courage. The poets cannot explain their own poems. The politicians cannot say what justice is. Each of them speaks with authority, and each collapses when asked to justify it.
Socrates called this condition doxa, which we translate loosely as opinion or appearance. It is the confident performance of knowing without the underlying structure of knowing. It is what happens when a person has absorbed the vocabulary of a domain without absorbing its logic.
The only true wisdom is in knowing you know nothing. What Socrates meant was not humility as a virtue but humility as a diagnostic tool. If you cannot locate the edges of your own understanding, you cannot tell where competence ends and performance begins.
Now consider what large language models produce. They are, in the most literal sense, engines of doxa. They generate text that resembles what an expert would say, because they have been trained on billions of tokens produced by experts. They are fluent without being knowing. They speak in the register of authority without possessing the underlying structure that made the authority earn its register in the first place.
And here is the twist Socrates would have savored: the humans who use these tools are increasingly indistinguishable from the tools themselves. Not because humans have become machines, but because the market for surface fluency has been flooded. When any 22 year old with a laptop can produce a report that sounds like a partner wrote it, the signal that fluency once carried has collapsed.
The Sophists Return With Better Marketing
The people Socrates argued with most fiercely were the Sophists, professional rhetoricians who taught wealthy Athenians how to win arguments regardless of whether they were right. The Sophists were not stupid. They were, in many ways, the first consultants. They sold the appearance of wisdom to people who wanted the appearance of wisdom, and they were paid handsomely for it.
Socrates hated them for a specific reason. He believed that when a society could not distinguish between the appearance of wisdom and wisdom itself, its politics would rot, its courts would rot, and eventually its citizens would rot along with them. The Sophists were not just annoying to him. They were a civilizational threat, because they scaled shallow thinking.
AI is a sophist at industrial scale. It does not lie, exactly, but it does not know either. It produces the shape of an answer. And the reason this matters is that human institutions have spent 2,000 years building filters designed to catch human sophistry: peer review, credentials, apprenticeships, licensing exams, editorial oversight. Those filters were calibrated to detect a human trying to fake expertise. They were not calibrated to detect a machine that has never known anything but produces expert output by default.
What Expertise Actually Was Before the Flood
To understand what AI is doing to expertise, we first need to be honest about what expertise really was. It was not one thing. It was at least four, stacked on top of each other, and only one of those four is genuinely threatened.
Layer 1: Access to Information
For most of human history, being an expert meant having access to information that other people did not have. The doctor knew what was in the medical textbook. The lawyer knew what was in the case law. The consultant knew what was in the industry report. This was the lowest layer of expertise, and it was always the most vulnerable to technology.
Google broke this layer 20 years ago. AI has finished the demolition. If your professional value depends on knowing something the client could look up, your professional value is approaching zero. This is not new. It has just been accelerated.
Layer 2: Fluency in the Language of a Domain
The second layer is speaking like an insider. Using the right terms, framing problems the way the field frames them, producing documents in the format the field expects. A pitch deck that looks like a pitch deck. A legal brief that reads like a legal brief. A diagnosis written in the voice of a physician.
This is the layer AI has just obliterated. Fluency, which used to take 5 to 10 years to acquire, can now be rented for 20 dollars a month. And this is where most of the anxiety about AI and expertise actually lives. Professionals who built their identity on sounding competent are discovering that sounding competent is no longer scarce.
Layer 3: Pattern Recognition Under Uncertainty
The third layer is where things get interesting. A senior surgeon does not just know anatomy. She has cut into 4,000 bodies and knows what a healthy liver feels like at 3 in the morning when the lighting is bad and the patient is bleeding. A veteran trader does not just know finance. He has watched 5 market crashes and knows what panic smells like before the numbers confirm it.
This layer is much harder for AI to replicate, because it is built on embodied experience rather than tokens. Machines can approximate it in narrow domains where the data is rich and the feedback loops are tight, but they struggle when the situation is novel, ambiguous, or requires reading humans.
Layer 4: Judgment About What Matters
The fourth layer is the one Socrates cared about most. It is not knowing facts, not sounding fluent, not even recognizing patterns. It is knowing which question to ask in the first place. It is the ability to look at a problem and see, before anyone else, what is actually at stake.
An unexamined life is not worth living, Socrates said at his trial. He was not offering a bumper sticker. He was saying that the highest human capacity is the capacity to interrogate one’s own assumptions, and that this capacity cannot be delegated, purchased, or automated.
Judgment about what matters is the last citadel of expertise, and it is the one AI cannot storm, because AI does not care. It has no stakes. It has no skin in any game. It can produce a beautifully argued memo for either side of any debate, which means it can produce neither wisdom nor conviction.
The Rise of AI Fake Expertise and Why It Is Dangerous
The specific danger of the current moment is not that AI is smart. It is that AI is fluent, and fluency has always been how humans detect intelligence in other humans. We are pattern matchers. When something speaks with the rhythm of expertise, our brains grant it the authority of expertise, even when the substance is hollow.
This is what we might call AI fake expertise: content that passes every surface test for authority while failing every deep test for knowledge. It gets the vocabulary right, the structure right, the tone right, and yet a specialist reading it can feel, within 30 seconds, that something is off. There is no there there. The argument does not know why it is making the argument.
The Junior Analyst Problem
Consider a concrete case. A junior analyst at a strategy firm produces a market entry report on Southeast Asian fintech. It reads beautifully. It cites the right consulting frameworks. It uses phrases like regulatory arbitrage and adjacent verticals. A partner reads it and thinks, this person has potential.
The problem is that the report was 90% generated by AI, and the analyst does not actually know why the framework being applied is the right framework for this specific market. If the partner asks, so why did you choose Porter’s Five Forces rather than a Jobs to be Done lens here, the analyst freezes. There is nothing underneath the fluency.
Multiply this by 10 million knowledge workers, and you have a serious institutional problem. Organizations are hiring, promoting, and paying people based on outputs that no longer reveal what the person actually knows. The signal has been jammed.
The Content Layer Collapse
The same thing is happening at the content layer of the internet. Blog posts, whitepapers, expert commentary, thought leadership, even academic papers, are increasingly written or heavily assisted by machines. And these machines were trained on prior human content, which means the ecosystem is beginning to feed on itself. Fluency compounds. Depth thins out.
This is why the question does AI devalue expertise is only half the right question. The other half is: does AI make it harder to recognize real expertise even when it exists? And the answer to that second question is yes. When everything looks like an expert wrote it, the actual expert has to work harder to prove she is one.
The Socratic Method as a Survival Skill
So what do you do if you are a professional trying to build or protect real expertise in a world flooded with counterfeit versions of it? The answer, strangely enough, is 2,400 years old.
Socrates had one move. He asked questions. Not casual questions but structural ones. He would take an assertion and press it until it either revealed its foundations or collapsed. His method was slow, adversarial, and deeply annoying to everyone he applied it to. It also happens to be the single most powerful tool humans have ever developed for separating real understanding from performance.
Interrogate Your Own Outputs
The first application is internal. Before you send any document you produced with AI assistance, ask yourself the Socratic questions. Why is this argument true? What would falsify it? What are the strongest counterarguments? What am I assuming without evidence? What would a genuine expert in this domain notice about this that I have not noticed?
If you cannot answer these questions about your own work, you are producing AI fake expertise, and you are one skeptical reader away from being caught. The tool is not the problem. The failure to interrogate the tool’s output is the problem.
Interrogate Everything You Read
The second application is external. Assume, by default, that any content you encounter online has been touched by AI. Assume that any confident tone might be masking hollow reasoning. Do not trust surface fluency. Look for specific claims, specific evidence, specific reasoning that only a person who has actually done the work could produce.
Real experts leave fingerprints. They mention edge cases the training data would not emphasize. They admit uncertainty in specific places. They contradict conventional wisdom when the specific situation warrants it. AI, by contrast, tends toward the median of its training corpus. It rarely takes real risks, because it does not know what a risk is.
Build What Machines Cannot Fake
The third application is strategic. If layers 1 and 2 of expertise, information access and domain fluency, have been commoditized, then your career depends on your ability to build layers 3 and 4. Pattern recognition under uncertainty, and judgment about what matters.
These layers can only be built through direct experience. There is no shortcut. You have to actually sit with 100 real clients, watch 50 real products fail, negotiate 40 real deals, live through 3 real crises. This is the part that AI cannot compress, and it is therefore the part where the returns to human effort are about to increase, not decrease.
The New Signal Economy
We are entering a period where the market for expertise will bifurcate sharply. On one side, a vast commodity layer of AI generated content, produced instantly and cheaply, will collapse in price toward zero. On the other side, a much smaller premium layer of demonstrably human, demonstrably experienced expertise will command higher prices than ever before.
The middle, where most professionals currently live, will hollow out. This is the uncomfortable truth behind the debate about AI and expertise. It is not that all expertise dies. It is that mid-tier expertise, the kind that depended on being slightly more fluent than a layperson, becomes economically unviable.
What This Means for Careers
If you are early in your career, the strategic implication is severe. Do not build your identity on being able to produce polished outputs. That skill is now free. Build your identity on being able to make judgment calls that would embarrass a machine to attempt: which client to fire, which product to kill, which colleague is lying, which market is about to break, which risk is worth taking.
These are Socratic skills. They come from asking the right questions under real conditions with real consequences. They come from being wrong publicly and learning from it. They come from having stakes.
What This Means for Institutions
Institutions face a parallel problem. The credentials they issue, the reports they produce, the analyses they publish, are all going to be viewed with new skepticism. Anyone can produce the artifact. The question becomes: can you produce the human capacity that would have produced the artifact if the machine had not existed?
Universities, consulting firms, law firms, media organizations, and research institutions will be judged, over the next 10 years, not by the quality of their outputs but by the quality of the humans they build. Because outputs are now cheap. Humans who can genuinely think are not.
The Examined Career
Socrates was executed by Athens for corrupting the youth. What he had actually done was force the city’s most confident citizens to admit they did not know what they were talking about. Athens preferred fluency to truth, and it killed the man who kept exposing the difference.
We are, in a smaller way, at a similar crossroads. We can choose to be dazzled by fluency, to reward the appearance of expertise, to promote the people whose outputs sound impressive, and to build careers on the assumption that sounding smart is the same as being smart. Or we can do the harder thing.
The harder thing is to examine our own knowledge with the same intensity that Socrates examined his fellow Athenians. To ask, honestly, what do I actually know that a machine could not fabricate? What can I do that requires my specific presence in the world? Where does my judgment come from, and would I trust it under pressure?
These questions are not comfortable. They were not comfortable in Athens either. But they are the only questions that matter now, because they are the only questions that draw a line between the human expert and the machine that has learned to sound like one.
The great flattening of expertise is real. But underneath the flattening, a new hierarchy is forming, and it will be steeper than the one it replaced. At the top will sit people who have done the Socratic work: who know exactly where their knowledge ends, who can defend their conclusions under interrogation, and who have the scars of real judgment made under real uncertainty. Everyone else will be competing with a machine that works for 20 dollars a month and never sleeps.
Choose which side of that line you want to be on. Then start asking better questions.
