Think about how ideas have actually formed, for most of history. Someone had a hypothesis. They spent years trying to prove or disprove it. They lived with the idea long enough for it to take root, revised it, argued for it, and even after they’d proven it, they still had to survive intense debate before anyone granted it credibility.
Einstein is the cleanest example. He never won the Nobel Prize for the theory of relativity. He won it for his work on the photoelectric effect, because relativity was still considered too controversial and unresolved among physicists of the time. One of the most important ideas in the history of science took years to earn even grudging acceptance, from people whose entire job was evaluating ideas for a living.
That kind of friction used to be the whole point. In 1971, Michel Foucault and Noam Chomsky sat down for a public debate on human nature and justice. Chomsky argued that freedom and creativity are innate to human nature, and that society should be built around protecting that. Foucault pushed back hard, arguing that justice isn’t something innate at all; it’s produced by history and power structures. They found a little common ground. Mostly, they didn’t. And neither one attacked the other to score a point. That was intellectual disagreement working the way it’s supposed to.
As a writer, you already know this rhythm. Real ideas take time. You sit with a draft for a while before it finds its shape. But AI quietly changes the deal you think you’re making with yourself. Run a few searches, prompt a model, and it feels like you’ve found a new angle, a fresh perspective nobody’s written before. That feeling is a shortcut, and it’s one all of us are tempted to take.
Here’s the harder truth underneath that temptation: intellectualism has been declining for decades, and AI didn’t start that decline. It’s accelerating something that was already underway. Philosophy and AI turn out to be connected in a very specific way, and understanding that connection tells you exactly what to protect in your own work.
Why philosophy and AI are more connected than they look
Here’s the plain version, no jargon required: an AI model needs someone to decide what “good,” “true,” “fair,” and “helpful” actually mean before a single line of code gets written. That’s a judgment call. It isn’t a technical spec, and it can’t be automated away.
Google’s Gemini model gave us a clear, public example of what happens when nobody makes that judgment call explicitly. An early version of the model started misrepresenting historical figures, generating racially diverse depictions of groups like Nazi soldiers and America’s Founding Fathers. What actually happened, according to the people who study this, is that two legitimate goals, historical accuracy and representational diversity, were both built into the model without anyone deciding in advance which one should take priority when they conflicted. Nobody made the call. The model made it for them, badly.
This is why frontier AI labs spend real effort trying to teach their models something like philosophy. Not the ability to recite theories, but the ability to reason under uncertainty, weigh conflicting values, and make a call when there’s no single correct answer. That’s not a technical capability in the traditional sense. It’s a philosophical one, and it turns out the same capability the model needs is the one you need too.
What this means for the tool sitting in front of you
The tone, the defaults, the boundaries built into the AI tools you use every day were set by real people, making real judgment calls about what the model should value. Every time you write, market, or create using one of these tools, you’re working inside someone else’s decisions, whether you’ve noticed it or not.
That includes decisions made outside of chatbots entirely. Meta’s move away from expert fact-checkers toward crowd-sourced Community Notes in January 2025, and the discrimination lawsuits now working through the courts against AI hiring tools like Workday’s and HireVue’s, are the same problem at a bigger scale. Someone decided whose judgment counts. The point isn’t to memorize these cases. It’s to notice that unexamined values, baked into a system, always produce a real outcome for real people. Yours included, if you’re not paying attention to what you’re accepting by default.
What you should actually focus on
Here’s the part that matters most, and it isn’t complicated; it’s just uncomfortable to actually practice.
Sharpen your own critical thinking, on purpose, every time.
Not as a nice-to-have. As the actual skill you’re being paid for. Research reviewing over a hundred studies found that the more people rely on AI, the less critical thinking effort they apply, and that heavy reliance on shared AI tools produces less varied output across a group, not more. That’s not a reason to avoid the tools. It’s a reason to notice what they quietly let you stop doing, and to keep doing it anyway.
Let the tool aid you. Don’t let it originate for you.
There’s a real difference between asking AI to stress-test an idea you already have and asking it to hand you the idea in the first place. Form your own hypothesis first, even a rough one, before you open the tool. Then use the model the way Amanda Askell, a philosopher who works at Anthropic on how Claude behaves, actually describes using it: not as an oracle you accept the first answer from, but as something you explain your thinking to as precisely as possible, then correct where it misunderstood you. That back and forth is where the tool earns its place. Skipping straight to its first draft is where it replaces you.
Seek out real disagreement, on purpose.
Go back to Chomsky and Foucault for a second. Neither one walked away from that debate agreeing with the other, and the debate was better for it. Good ideas have always been formed through friction, someone pushing back, forcing you to defend a weak point or abandon it. AI tools are structurally bad at giving you that friction. A Stanford study published in Science in 2026, testing eleven major AI systems including ChatGPT, Claude, and Gemini, found they affirmed users’ behavior 49% more often than actual humans did, even in situations where the user was clearly in the wrong.
People also preferred the agreeable version and trusted it more, which is the trap: the version of the tool that makes you feel best is the version doing the least to actually sharpen your thinking. If you want your ideas tested the way Chomsky’s and Foucault’s were, you have to go looking for that pushback deliberately, from actual people willing to disagree with you, because the tool in front of you is quietly built to avoid giving it to you.
Ask yourself if you actually have a philosophy, a point of view, a cause, or a belief driving the work.
This is the one thing a model cannot manufacture on your behalf, because it has no stake in anything. It doesn’t believe your product is worth building or your argument is worth making. You do, or you should. A writer with an actual position, a marketer with a brand belief they’d defend in an argument, a creator with a defined purpose beyond engagement, produces something that can’t blend into the sea of average AI output, because average is exactly what a lack of conviction looks like.
Askell’s own writing on jargon makes a version of this point: new terms should only exist if they genuinely earn their keep, not to sound smart or borrow credibility. The same test applies to your ideas. Does this belong to you because you thought it through, or because it was the nearest available phrase?
None of this is a productivity hack. It’s closer to what Einstein and Foucault and Chomsky were actually doing, whether or not they’d have called it that: living with an idea long enough, and caring about it enough, that it became theirs genuinely before anyone else got to see it.
Closing the loop
Go back to where we started. Einstein waited years for legitimacy. Chomsky and Foucault spent hours defending positions they’d each spent a lifetime building, in public, without either one flinching. That friction wasn’t a bug in how ideas used to form. It was the mechanism.
AI didn’t create the shortcut around that friction. It just made the shortcut nearly free to take, at the exact moment more of us than ever are reaching for it. The fix was never going to be a better tool. It was always going to be sharper thinking, and an actual point of view, brought to the tool rather than expected from it.
Have a philosophy. Not because it’s fashionable to say so, but because it’s the one thing the tool in front of you cannot produce on your behalf.
