Skip to content
AI isn't Intelligent - it's Pattern Machines

AI isn't Intelligent - it's Pattern Machines

May 17, 2026
There’s a followup article on what the difference between human intelligence and pattern machines might be, which you can read here.

Marketing Genius

The single greatest move in AI was naming it AI.

That’s perhaps a little unfair to the researchers who work in the field in earnest, but I’m really talking about the business side here. The name “Artificial Intelligence” is magnificent, because it leads us into making assumptions about the technology from the very start.

Current AI systems are pattern machines. They analyse and produce patterns. They do not think, and can never think.

But there’s a bigger problem than that: they are also being used by pattern machines.

Pattern Machines

Humans are biology’s ultimate pattern machineai-isnt-intelligent-its-pattern-machines. That rustling in the bushes? Could be a predator. Better give it a wide berth, unless you’re eager to meet your ancestors the same way that Ug and Og met theirs.

Finding patterns in our environment is simply one of our innate survival skills. It’s something we do so well that it’s allowed us to exploit the power of specialisation, and through that build civilisations. The metalworker doesn’t need to farm any more than the carpenter needs to weave; we build more together by exploiting the patterns in our niches.

Sadly, we’re a bit too good at it. We also see faces in the patterns of knots in wood. Or on bits of toasted bread. We see animals in the shapes of clouds or rock formations. It’s a little bit embarrassing to be honest.

Yet the qualities of the patterns from the systems we’ve built means we seem to be blind to them being patterns. It creates a positive feedback loop, which risks a curious form of intellectual addiction.

Worse than that, because we’ve been told it’s an Intelligence, and because the patterns really are quite good, it’s easy to make the jump and assume that the pattern machines are actually intelligent. We don’t even realise there’s a problem with that jump.

Pattern machines aren’t intelligent. They can’t be. They are statistical machines that have no consciousness, no thought processes. The closest that they can ever get to “thinking” is iterating around and around trying different patterns until they find one that has a slightly better arbitrary score on an internal quality checking system.

Hallucinations are a part of pattern machines, whether implemented in silicon or biology.

Selective Attention

I do see some areas where pattern machines could have some significant benefits. To understand that, we must first take a brief diversion to the idea of selective attention.

You may already know of selective attention from this video in which you must count the number of times a basketball is passed between people wearing white whilst a second ball is also passed between people wearing black: https://www.youtube.com/watch?v=vJG698U2Mvo

Selective attention is a human issue. We have certain biological limitations in our faculties. We miss certain patterns simply because we’re distracted by other ones, making us incapable of finding them except by accident. Pattern machines don’t have that problem.

So when it comes to finding potentially useful patterns in chemicals for creating new drugs, pattern machines could be useful. Perhaps there are similar uses in material sciences and other areas.

Pattern Persistence

There is a flipside to finding patterns we’re not looking for. Ironically, despite being able to find new patterns, it’s very much rooted in existing ones.

There are also significant dangers of using them in areas where we know we already have inadequate or unacceptable source patterns. For example in areas of justice or employee relations we know that there may be a history of biases which we do not want to reproduce in the future. Pattern machines should be avoided there. The old phrase “garbage in, garbage out” applies to pattern machines too.

Costings

Regardless of the accuracy and nature of the patterns, there is also the cost to analyse or produce them.

The current trend in “consumer AI” seems unsustainable, and even the rather more financially comfortable enterprises find the costs a bit too much. OpenAI can’t make money on a $200/month subscription plan, which is frankly laughable when most companies would ask strong questions about an additional cost of $20/month per employee. These numbers may sound small individually, but when you scale them up to a whole company you’re suddenly looking at a sizable chunk of budget that someone needs to justify.

And AI is always additional, not a replacement. That’s changed a little recently but it’s mostly marketing - if you think you’re replacing a tool with a newer, shinier AI tool then it’s most likely you’re replacing that tool with some other tool that someone put an AI sticker onto.

Spending significant additional money on something you know will make just as many mistakes as the people you already employ is hard to justify. There’s an old joke about computers “making very fast, very accurate mistakes”. Pattern machines give us the opportunity to make very fast, very inaccurate mistakes, for a significant monthly subscription cost.

I’ve referred to them as pattern machines throughout this as I find that a useful trick when evaluating what they can do - as I said in the opening paragraph, if you use the phrases “AI” or “Artificial Intelligence” you’ve primed yourself to overestimate its abilities.

So am I an AI Skeptic? Yes.

Do I think they have no value whatsover? No.

I believe that the only way to find the value is to think of them as pattern machines, and to pay close attention to both the inputs and the outputs, as well as keep a close eye on the costs of running them.

To be fair, this isn’t the first time we’ve failed to understand a technology. In the 1950s and 1960s it wasn’t uncommon to see computers described as “electronic brains”, despite being no such thing. That led to ideas of them being intelligent which took a while to disabuse people of.

Interestingly this also reflects what we thought of as intelligence - arithmetic skills, recollection of facts, application of logic. Our view of education has changed somewhat since then, broadening to include and focus more on things that the electronic brains could not do.

Perhaps we will find our future selves determining that intuition is merely a pattern operation, which we can delegate to our new tools? It may seem heretical now, but no more than some views of computers seemed in the 1950s.

The genie is out of the bottle, but it’s far from clear that it can grant us any wishes, much less ones we’d actually ask for…

There’s a followup article on what the difference between human intelligence and pattern machines might be, which you can read here.
Last updated on