Can AI be creative?

The harder question is whether any of us humans are actually equipped to judge originality in the first place.

Yes, machines can be creative, just don’t expect anyone to agree on whether something is universally creative, or even on what creativity means.

But why not? Creativity is creativity, right?

Well, no. It really depends on who is judging and by what criteria. One way its framed is that has to be both useful and novel.

So let’s do a test using that definition and you’ll see why, after we’ve had three plus years with AI, that question still bubbles up the company social.

To be stamped as creative, the output or idea needs to be useful (e.g. valuable, effective, appropriate) and novel (e.g. original and new).

So how does AI stack up and generating that effect? Based on my own experience, pretty well. Anyone who has used it to knock out an email or unblock a creative project already knows it can be useful, effective, and appropriate.

But can AI be novel? Again, absolutely. You’ve seen the avocado chair in the shape of Trump’s head, right?

On those yardsticks, yes, AI can indeed be creative. But this idea of “novel” is too generous, and let’s not let AI squeeze through on a definition technicality.

Instead, let’s sharpen the question to this: can AI be original?

The originality problem

This is where it gets messy, because “original” is both subjective and objective. If creativity has to prove that it is original, then we have a problem, because judging true originality requires something close to omnipotent awareness of what came before.

On a slight head-f**k side note, if AI is aware of almost everything, then AI arguably has the best chance of being creative and is also the best placed to judge it, so based on our own definition, can humans ever be as creative as AI? I’ll just leave that here…

Anyway, if originality is a key tenet of creativity, it is pretty tough to judge, which is why this point is so contentious. Let’s paint that picture with an example: the invention of the fork.

A head-fork

Someone creates a fork in complete isolation, never having seen one before, and then shows it to a person who already knows what a fork is. Is it creative? From the maker’s point of view, yes: they genuinely arrived at something new, with no idea it had ever existed. But from the viewer’s point of view, no. They’ve seen a fork a thousand times, so there is nothing new about it at all.

Now turn it around. Imagine someone designs a fork knowing full well that forks already exist, and shows it to a person who has never seen one in their life. From the maker’s point of view, this is not creative. They were copying something they already knew. But from the viewer’s point of view, it is a marvel: a strange new object they’ve never encountered before.

The point is that creativity does not sit neatly inside the thing itself. It often sits inside the person encountering it.

Something can feel fantastically creative to one viewer and completely unremarkable to another, depending on what each of them already knows, has seen, or expects.

So when people argue about whether AI is creative, they are usually muddling together two different questions: whether something appears creative to a viewer, and whether it is genuinely original in any absolute sense. Once you separate those two, and park the first for a moment, you get back to the harder question underneath all of this: can AI actually produce something original?

Can AI actually produce something original?

To even attempt to answer that, we have to return to our hypothetical omnipotent being: an all-knowing judge with total awareness of what came before. Only someone in that god-like position could say with confidence whether something is genuinely unprecedented.

When Einstein came up with general relativity, or Picasso helped create Cubism, you could pretty confidently say: yes, this is truly ground-breaking. (Side note: there are very few non-contentious examples of absolute creativity; even the iPhone was preceded by IBM’s Simon by a stonking 13 years.)

The problem with AI is that it has ‘seen’ far more than any human ever could. And when something has seen everything, it is very easy to say, “Well, you’re probably just copying aren’t you.” It’s not surprising that we think that either. We’ve been told that AI is trained on a gazillion petabytes of human content.

But hold on, so what? Humans are no different. If experience is part of invention and creativity, are we also not “trained on” a bucketload of data?

We’re also told that AI is really just a predictive engine and, because of its vast knowledge, it is very good at copying and working out what comes next.

That’s not creative; that’s just being well informed with a talent for maths.

That is partly true, but it isn’t the full picture. At a high level, text-based AI works in at least three relevant ways, and the prediction bit is only the first part.

This is the most techy bit here, but I’ll attempt to soften the blow with some stupid analogies.

First, AI predicts the next word, but it does so with context. Yes, it bases the next word on what it has seen before, but it does not guess blindly. It looks across the whole passage, weighing which earlier words matter, what the sentence is really about, and where the meaning seems to be going.

The analogy here is a foot soldier versus a senior commander. The foot soldier follows the handbook, so when X happens, they do Y. The commander, however, considers the bigger picture, responding with both experience and real-time information to make better decisions.

Okay, so it gets context. Whoop-de-do. It’s tantamount to Pavlovian conditioning with more bells on top. This is where people typically stop, and if this was all AI did, then yes, you absolutely could not call it creative.

Second, AI gets a lot of feedback. Long before you typed anything, people rated thousands of sample outputs and it learned to lean toward the clearer, more useful, and more acceptable ones.

Fine. It has a bunch of underpaid individuals passing judgement on what “good” looks like. I imagine this is a bit like the sixth time you do your company’s mandatory training: answer accuracy is not always paramount.

Third, AI is increasingly good at reviewing and reasoning. It can now often check its own work, revise its own answer, and evaluate its own thinking in real time rather than blurting out the first thing that comes to chip.

Fine again: it marks its own homework. But from personal experience of building agents for a while now, getting AI to rely on its own thinking is not always the best strategy. It’s what happens to all of us when we argue ourselves into a corner and end up spouting complete tosh because we’ve convinced ourselves it reinforces our original thinking and starting again feels like too much effort.

So even if you grant the machine supreme intelligence, plus a few extra bells and whistles, you might still insist that it is only a very sophisticated copyist, and sophisticated copying is not creation.

So let’s return to the original question:

“I get that it’s smarter than just predicting, but CAN IT BE CREATIVE - can it ever take a step beyond what it was already shown?”

The suitcase and the wheels

Well yes, it can. And the clearest way to see how is to forget machines for a moment and think about a suitcase and a set of wheels.

For decades, suitcases were these unwieldy things with handles that you had to lug around as you made your way to the airport. Now think about wheels. Wheels had nothing to do with suitcases; they went on cars, bikes, and carts. These were completely ordinary things that everybody had seen a thousand times over. Then someone put wheels on the bottom of the suitcase and nobody ever carried their luggage through an airport again.

Nothing new had been invented in the strict sense. The suitcase already existed and so did the wheel, and the only new thing was the combination. Some bright spark took two familiar objects and put them together in a way no one had thought to try before. The reason this creative thought probably came about is that they shared some common ingredients: both were grounded in movement, both were grounded in travel, and both were there to move something from A to B.

That is what invention and originality usually are. It is rarely a brand new ingredient that falls from the sky; instead, it’s more often a fresh combination of ingredients everyone already had, but had not thought to smash together yet.

This is also why the “it just predicts” objection is too narrow. Prediction only takes the obvious next step and decides a suitcase needs a better handle. Invention takes the road less travelled and decides a suitcase could probably benefit from some wheels.

AI, at its best, can do that sort of combinatory leap too. It holds an enormous stock of familiar things and the relationships between them, which means it can reach for two things that rarely meet and combine them into something that is genuinely new.

Now the obvious pushback is: prove it. Show me that AI is not just doing some advanced pattern-matching jiggery-pokery, and that it can actually pioneer new ideas or concepts.

Move 37

You’ve probably heard of this, but in 2016 a program playing the ancient board game Go made a move, known simply as Move 37. It was widely described as a move no top human would seriously consider. There was nothing obvious there for it to copy from accepted human play. It just emerged from the machine exploring the game in a different way, and it worked.

Now, a supercomputer costing tens of millions of dollars and running on absurdly powerful chips feels a touch different from your free tier of ChatGPT, and this I concede for now. (although we still have the avocado trump armchair). But that wasn’t really the point.

The point is that AI isn’t just a prediction machine. It can perform novel connection-making that we also perform when inventing, designing, painting, or writing.

Now before anyone starts throwing stuff at me, I’m not saying AI creates like a human. It has no skin in the game, no life behind the work, no embarrassment, no longing, no weird instinct for beauty that it can’t fully explain. Most of the time it still needs a human nudging it, steering it, and deciding what’s any good. But that’s not the same thing as saying it can’t produce something creative. It just means its version of creativity, if we’re willing to call it that, is a different beast.

So creative? Yes, I think so. But does that make it a creator? Personally, I’d reserve that label exclusively for the human condition, though maybe that’s just my precious human ego talking.