Now that we’re all a little more familiar with AI, we might find ourselves getting a ‘feel’ for what AI will be good at, and not so good at. A while back we might have thought that AI wasn’t good at writing (or AI’s version of writing). A few years ago doing a backflip was out of the question for a robot. Now these things are do-able, but we can’t get them to fold laundry, nor is it easy to grab an object or pass it to someone (or something) else. One of the most common problems is that the powerful robo-claw crushes whatever it’s trying to hold.

This is encapsulated in Moravec’ Paradox, which states that “it is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”.

The two sides of the challenge – intelligence vs perception – are different for AI and robotics researchers because intelligence tasks can be reduced to the manipulation of symbols. Computers do this all the time. Everything in a computer comes down to ones and zeros. It turns out that even language, and the order of words, can be reduced to symbols, hence the rise of LLMs. It took a lot of research on the part of AI developers, but the ability to reflect language as symbols is what made it possible. You could even make a stab at describing how language works. Language itself is a series of symbols. The word ‘lion’, and the thing, the lion, that the word refers to, are separate things. The word is a symbol for the animal, and sentence structure is also a thing that can be reduced to symbols.

Now try to describe handing a ball to another person, from your hand to theirs. Any attempt to render it as a series of symbols falls down. You can describe it in words, but there are no symbols for the feeling that the other person has taken a grip of the ball, so you can let go. There are no symbols for the interplay between the two people, the unspoken negotiation that takes place.

So AI – that is, computers – can be great at reproducing the human activities which can be devolved down to symbols. Examples are language, chess, Go, programming, even putting colours to paper / pixels.

But AI struggles at the things we find so intuitive as to be hard to describe. That’s why AI lacks true creativity, if by ‘creativity’ we mean the urge to create and the exercise of coming up with something new. Why did your kid draw a house, rather than a cat? Could the kid even tell you why?

AI might take over jobs that humans find hard to describe – care work, for example. But even those things which are panicking people now – AI doing novels, artwork and music – are probably going to come out the other side of a process whereby people start to see that the products aren’t the same. Artists and creators might use AI to create things, like a collage of individual elements that AI helped create. But the use of AI to create wholesale ‘songs’ and ‘novels’ from scratch are never going to top the ability of a human to create something totally new, because it’s just something even we humans don’t understand.

Sources

Moravec’s paradox, Wikipedia, retrieved 12th July 2026

When will robots have their Chat GPT moment?, The Artificial Human, Radio 4, 01 Jul 2026, retrieved 12th July 2026