What 'Intelligence' Has Meant, and Why the Definition Keeps Moving

In 1997, a computer beat the world's best chess player. For decades before that, chess had been held up as one of the clearest tests of real, human intelligence — the kind of deep, strategic thinking only a person could do. So when a machine finally won, you'd expect people to say: this machine is intelligent. Instead, something stranger happened. People quietly moved the goalposts, and chess stopped counting as proof of real intelligence at all.

The Test Keeps Changing After It's Passed

This isn't a one-time thing that happened with chess. It's a pattern that repeats, again and again, every single time a machine gets good at something we used to think only a human mind could do. Doing arithmetic quickly used to be a sign of a sharp mind. Then calculators did it instantly, and suddenly fast arithmetic didn't feel like real intelligence anymore — it felt mechanical. Recognizing a face in a crowd used to feel like something deeply, uniquely human. Then software got good at it too, and recognizing faces quietly stopped counting as a meaningful test.

There's a name researchers sometimes use for this: the idea that intelligence is 'whatever machines haven't done yet.' It sounds almost like a joke, but it describes something genuinely real about how we think. The moment a machine crosses a line we thought was uncrossable, we don't usually conclude the machine is smarter than we thought. We conclude the line was never where real intelligence actually lived in the first place.

Why We Do This, and Why It's Not Entirely Unreasonable

It's tempting to call this cheating — like humanity keeps changing the rules so it never has to admit a machine is genuinely smart. But there's something more honest going on underneath it too. Each time a machine crosses one of these lines, we often do learn something real: that the specific task we picked wasn't actually testing what we thought it was testing. Chess, it turns out, can be conquered by searching through an enormous number of possible moves very quickly — a kind of brute-force calculation, not the rich, intuitive judgment we imagined a chess grandmaster was using.

So moving the goalposts isn't purely dishonest. Sometimes it's a genuine correction — realizing the task we picked to represent 'intelligence' was actually a narrower, more mechanical thing than we gave it credit for. The problem isn't that we keep learning this. The problem is we rarely admit, out loud, that this is what's actually happening.

The Uncomfortable Truth: We've Never Fully Agreed What Intelligence Even Is

Here's something worth sitting with for a moment: humanity has been arguing about what intelligence actually means for a very long time, and we still don't have a single, complete answer even for ourselves. Is it the ability to solve a hard math problem? The ability to write something beautiful? The ability to read a room and know what someone else is feeling without being told? The ability to survive somewhere genuinely difficult, using nothing but wit and adaptation? Different thinkers, across different centuries and different cultures, have picked different answers — and every single one of those answers is defensible.

This matters enormously for how we think about AI, because it means we're not really asking 'is this machine intelligent' against one fixed, agreed-upon standard. We're asking it against a moving, contested idea that even humans have never fully settled among themselves. That doesn't make the question meaningless — it makes it a genuinely hard, genuinely human question, not a simple yes-or-no with a clean answer waiting to be discovered.

What Today's AI Can Actually Do, Stated Plainly

Modern AI systems are genuinely, remarkably good at finding patterns in enormous amounts of information and using those patterns to produce something new — a sentence, an image, an answer to a question. That's real, and it's not a small thing. But it's worth being honest about what this actually is: a very powerful, very fast form of pattern-matching, built from studying an enormous number of real examples, not something that clearly matches any single, agreed-upon definition of what it means to genuinely understand something.

Whether that counts as 'real' intelligence depends entirely on which definition you're using — and as we've just seen, humanity has never fully agreed on one. What's genuinely, provably true is narrower and more useful than the big philosophical question: these systems can do real, specific things that were, until recently, only possible for a human mind to do. Whether we choose to call that 'intelligence' says as much about us, and about the word itself, as it does about the machine.

The Question That Will Probably Outlast This Whole Debate

Here's a bet worth making: fifty years from now, whatever AI looks like at that point, people will still be having some version of this exact same argument. A machine will do something new and surprising. Someone will say it's not real intelligence. Someone else will point out that we said the same thing about the last surprising thing, too. The specific task on the table will be different. The shape of the argument will be exactly the same.

Maybe that's actually the most honest thing anyone can say about intelligence: it's less a fixed destination we can point to and measure against, and more a genuinely moving conversation humanity keeps having with itself, using whatever the newest, most surprising machine happens to be as the occasion for asking, once again, what actually makes a mind a mind.