There is no real AI race.
Well, there is. In a sense. There are enormous amounts of money being poured into increasingly large models, armies of engineers trying to squeeze another few percentage points of performance out of them, companies desperately trying not to be the one that gets left behind, and governments beginning to behave as though whoever controls the next generation of AI will control the future.
So yes, there is a race.
But we are making rather more of it than it deserves.
What we call AI is still, at its core, mathematical models doing what we have managed to teach them to do. That is already astonishing. It is also nowhere near the thing we keep pretending it is.
There is a rather important distinction here that gets lost in the breathless reporting.
AI is not a magic box into which you pour an ordinary person and out comes extraordinary work.
In fact, quite the opposite.
You can give current AI tools to someone who knows very little and, yes, they will produce some entertaining results. They will generate pictures, write things that sound vaguely intelligent, answer questions, invent stories, summarise documents and occasionally produce something that makes you stop and wonder what exactly happened to the world.
But getting genuinely good results is considerably harder.
The prompt matters. The context matters. The sequence in which you provide information matters. The assumptions you make matter. The model you choose matters. Knowing when the machine is confidently talking rubbish matters even more.
And then there is the rather inconvenient fact that the models themselves are not neutral.
They have tendencies. They have biases. They have preferences embedded in their training. They have things they will readily say and things they will resist saying. They have particular ways of framing questions. They can be astonishingly good at following one line of reasoning and remarkably stupid when you ask them to step sideways from it.
In other words, they are tools.
Very impressive tools, certainly.
But tools.
And like most powerful tools, their usefulness depends rather heavily on the person holding them.
This is where the current AI revolution becomes interesting, because the genuinely important change may not be that machines have suddenly become intelligent.
It may simply be that we have created a new kind of interface between human beings and information.
That alone is a very big deal.
Google has spent decades sitting in the middle of the information economy. Search was the gateway through which most of us entered the internet. If you wanted to know something, you typed it into a search box and were handed a collection of links, advertisements, fragments and suggestions from which you were expected to construct an answer yourself.
AI changes that relationship.
Instead of asking a machine where something is, we increasingly ask it to make sense of something for us.
That is already putting a dent into Google’s extraordinary dominance of search.
And that, in itself, is significant.
You don’t need artificial general intelligence to disrupt a business model that has been built around people searching for information in a particular way. You merely need a better interface.
But this is where the language gets ahead of reality.
We keep talking about general AI as though it is sitting just around the corner. The assumption seems to be that we have built most of the pieces already and that the final step is simply to make them bigger, faster and more capable.
Perhaps.
I wouldn’t hold my breath.
We don’t have an artificial intelligence that combines anything remotely resembling the full range of human capabilities.
Then again, what exactly are human capabilities?
This is where the discussion becomes considerably more uncomfortable.
Because we have a rather inflated opinion of ourselves.
We like to imagine that human beings are magnificent general-purpose reasoning machines, capable of understanding the world, forming independent judgments, weighing evidence, changing our minds when confronted with new information and generally navigating reality through the extraordinary power of thought.
It is a lovely story.
It is also, on closer inspection, rather questionable.
A great deal of what we call thinking is pattern recognition.
We see something that resembles something we have seen before and immediately begin filling in the blanks.
A great deal of what we call knowledge is accumulated pattern recognition.
And an astonishing amount of what we call opinion is simply a narrative we have heard often enough to mistake repetition for truth.
We absorb stories from our parents, our friends, our education, our culture, our political tribe, our profession, our newspapers, our social media feeds and the people whose approval we want.
Then we repeat them.
Eventually the narrative becomes part of us.
Once that happens, actual thinking becomes strangely optional.
We don’t investigate the premises very much anymore. We don’t necessarily test the conclusion. We recognise the pattern, retrieve the appropriate narrative and deploy it.
And then we congratulate ourselves for having an opinion.
Seen from that perspective, AI begins to look rather less alien.
It recognises patterns.
It constructs narratives.
It predicts what should come next.
It is influenced by the material it has absorbed.
It can become extremely confident while being completely wrong.
It can reproduce the prejudices embedded in its training.
It can produce something that sounds remarkably like reasoning without necessarily possessing the thing we imagine reasoning to be.
There is an uncomfortable resemblance here.
But the resemblance should not be mistaken for equivalence.
AI is not human.
And humans are not secretly just large language models walking around in shoes.
There is still something profoundly different about the way we inhabit the world. We have bodies. We have instincts. We have memories tied to experience. We have desires, fears, ambitions, mortality and all the other messy machinery that comes with being an animal that knows, at least dimly, that it is going to die.
AI doesn’t have that.
At least not now.
And yet the most interesting part of the comparison may be what it does to our conception of ourselves.
We tend to assume that if a machine can imitate something we do, then the machine must be extraordinarily advanced.
Sometimes that is true.
Sometimes it is simply revealing how little of what we do was as mysterious as we thought.
Machines already vastly surpass us in particular niches.
They calculate better than we do. They remember more than we do. They can process enormous quantities of information without getting bored. They can recognise certain patterns far faster than a human can. They can search through possibilities at a scale no individual human could ever approach.
And now they can produce language and images and code that are often good enough to make the distinction between tool and collaborator increasingly uncomfortable.
But that doesn’t mean they are about to become human.
It means they are becoming extremely good at particular things.
And that is probably the more sensible way to look at the whole AI story.
There is no need to pretend that we have already created artificial general intelligence.
We haven’t.
There is no need to pretend that the machines are useless because they don’t think exactly like us.
They are already enormously useful.
And there is certainly no need to assume that the next breakthrough is inevitably going to produce some omniscient digital being that suddenly acquires every capability we associate with ourselves.
Maybe it will.
Maybe it won’t.
The problem is that we don’t actually know what “every capability we associate with ourselves” even means.
Humans are not as rational as we like to believe. We are not as independent as we like to believe. We are not nearly as consistent, objective or intellectually disciplined as we like to believe.
We are pattern-recognition machines with prejudices.
Very complicated ones, admittedly.
But pattern-recognition machines nonetheless.
And perhaps that is why the AI revolution feels so strange.
It isn’t merely that machines are becoming more capable.
It is that, for the first time, we are building machines that are good enough at some of the things we do to force us to take another look at what it is we actually do.
That doesn’t make AI human.
Far from it.
But it does make the question considerably more interesting.
And no, I wouldn’t hold my breath waiting for general AI to arrive tomorrow.
There is a very long distance between being extraordinarily good at some things and being generally capable of being a human being.
We may cross it.
But don’t confuse the machines getting better with the distance necessarily becoming shorter.
Sometimes what looks like progress toward the destination is merely progress toward a completely different destination.
Tweet version
There is no real AI race.
Well, there is. In a sense.
But we wildly overstate what is happening. Current AI is still mathematical models doing what we have taught them to do. Getting genuinely good results requires skill, context and very specific prompting.
It is an extraordinary tool. It is also not general intelligence.
The more interesting question may be what AI tells us about ourselves.
We like to think humans are sophisticated reasoning machines.
Much of the time we’re pattern-recognition machines with prejudices, repeating narratives until we mistake familiarity for thought.
AI doesn’t make us obsolete.
But it may force us to reconsider how special some of the things we thought made us special actually were.
Don’t hold your breath for AGI.