What a Forgotten Philosopher Can Teach Us About the Limits of AI: Meet Michael Polanyi
Michael Polanyi (1891–1976) doesn’t come up much at the dinner table. He was a Hungarian-British polymath — a physical chemist who became a philosopher of science — and most people have never heard his name. But his work points to a thesis we need right now: AI may be useful, but it cannot replace the human judgment behind knowledge as it moves from novelty to infrastructure.
Polanyi’s central insight was simple but radical: science is never as detached or rule-bound as it likes to claim. Behind every discovery is a person — with instincts, commitments, and a feel for what matters. He called this personal knowledge, and it’s a useful lens for thinking about artificial intelligence.
Let’s walk through it using a simple See, Judge, Act framework, as Catholic social teaching often recommends for hard issues.
See: What’s Actually Happening
Strip away the hype and the panic, and here’s what AI systems do well: they recognize patterns across enormous amounts of data, generate plausible text or images, and imitate expertise convincingly enough to fool most of us most of the time.
Here’s what they don’t do: they don’t live anywhere. They have no body, no history, no stakes in the outcome. They’ve never been embarrassed, lost sleep over a decision, or had to look someone in the eye after getting something wrong.
Polanyi has a name for the kind of knowing that comes from actually being somewhere: tacit knowledge. His famous line is that we know more than we can tell — not just more than we can say out loud, but more than we could write down even if we tried. A teacher senses a class losing focus before anyone fidgets. A musician feels when a phrase needs to breathe. A doctor notices something off in a patient’s face that no chart captures. None of that lives in a rulebook, which means none of it can be fully handed to a machine, no matter how good the machine gets at sounding like it understands.
That’s the situation. So the next question is harder.
Judge: What does this mean
It’s tempting to think of algorithms as neutral because they’re built on math. Polanyi would push back hard on that. Numbers don’t choose themselves — people decide what to count, what to optimize, what counts as an error, and whose experience gets centered in the data. Every one of those choices is a values choice wearing a technical disguise.
So when an AI system makes a recommendation, denies a claim, or drafts a decision, it isn’t reporting objective truth from nowhere. It’s reflecting the judgments — and blind spots — of the people who built and trained it. That’s not a flaw to be engineered away; it’s a feature of how all knowledge works, machine-assisted or not.
This matters because the decisions we most need wisdom for — how to treat a struggling student, whether to trust a diagnosis, how to weigh competing claims in a community dispute — are exactly the decisions that resist being reduced to inputs and outputs. They require prudence: the kind of practical judgment that’s earned through experience, not downloaded from a dataset.
None of this means AI is useless or dangerous by nature. It means AI is a tool shaped by human hands, and tools don’t bear responsibility — people do.
Act: What to Do With That
If Polanyi is right, the answer isn’t to reject AI or to treat it as an oracle. It’s to put it in its proper place: a powerful assistant for tasks that benefit from pattern-spotting and speed, used by people who stay alert to the parts of judgment that can’t be outsourced. That’s where action begins.
Practically, that might look like:
Naming the choices baked into a system. Before trusting an AI tool’s output, ask what it was optimized for and who decided that.
Reserving final judgment for humans in consequential decisions — hiring, medical care, sentencing, anything where dignity is on the line.
Protecting tacit expertise rather than letting it atrophy. If a skill can be quietly outsourced to a model, the temptation is to stop practicing it — but that’s exactly the knowledge Polanyi says we can’t afford to lose.
Treating “the AI said so” as the start of a conversation, not the end of one.
Questions Worth Sitting With
Where in your own work or life have you relied on a kind of knowing you couldn’t fully put into words? Could that knowledge be captured by a machine — or only imitated?
When you’ve used an AI tool, did you ask what assumptions or values were built into its design? What would change if you did?
Are there skills or forms of attention you’ve started to let go of because a tool can now do a passable job in their place? What would it cost to keep practicing them anyway?
Who decides what counts as a “good” outcome for the AI systems you encounter — and would you agree with their definition if you saw it written down?
Polanyi spent his career insisting that knowledge always has a knower behind it — someone with skin in the game. AI doesn’t change that fact; it just makes it easier to forget. The point here is simple: AI can assist, but it cannot replace the judgment that gives knowledge its human weight. The work now isn’t to out-think the machines. It’s to stay the kind of people whose judgment is worth trusting when the machines fall short.