Relax, humans! Here’s why I find ChatGPT reassuringly stupid

There’s a book out recently, still front and centre of many bookshops, called If Anyone Builds It, Everyone Dies: The Case Against Superintelligent AI, and it has been rapturously received – if rapture is quite the right response to predictions of the extinction of humanity by a malevolent AI.

Even as a technology writer in awe of the past three years’ explosion in generative AI, I am increasingly unsure whether it will beat climate change to killing us all. So far, at least, I’d say it’s very useful as a tool – a turbocharged Google if you like – but still essentially, and reassuringly, stupid.

Even if it tries to take over everything, I think it will probably cock it up just as we humans have, only faster.

The other evening, I was watching a particularly bad football match when one of the commentators said that a particular defender “has covered every blade of grass on the pitch tonight”.

As a keen follower of the online Dull Men’s Club, I was so bored I got to wondering how many blades of grass there are on a pitch and asked my chatty friend ChatGPT. It did a screen full of sums and concluded that the answer was 140 billion. I was impressed. I don’t recall ever knowingly seeing 140 billion anything.

I texted my brother, a science geek-turned journalist who’s way more numerate than I am, and asked him to guess. “I dunno, a few hundred million?” he replied. Nope, I said, telling him the figure.

“Bollocks, who says so?” ChatGPT.

“Still bollocks,” was that man’s considered and final verdict.

So I asked a few more of my closest AI friends for their view.

Claude said 3 to 5 billion. DeepSeek – the Chinese one – said 2 billion. Google AI estimated from 900 million to over 2 billion. Grok, the dodgy Elon Musk AI, 1.3 billion. Perplexity said “about 250 million”. And finally, I went back to ChatGPT, which had swiftly had a recount and was now saying 150 to 500+ million, “depending on turf density”.

So, the technology so brilliant it’s reckoned by some of the world’s finest human brains to be about to wipe out humanity does essentially the same sum and comes up with answers ranging from 150 million to 140 billion – the latter over 900 times the former.

That’s quite some margin of error. Insane, actually, especially with a sum that is well within the scope of a human with basic knowledge and a calculator.

George Davies Turf in Bedfordshire, the first major turf supplier Google comes up with, and who seem to know a thing or two about grass, say on their website that a square metre of turf contains up to 200,000 blades of grass. Up to, note. The max.

Football pitches, the broader internet says, go “from 7,140 to 10,800 square metres.” So, using just a desk calculator, we can say with some authority the average football pitch contains 1.4 to 2.1 billion pieces of grass. That was not hard.

There are probably many reasons why the existing AI agents got such a very simple calculation so spectacularly wrong. The main one is that these things are scavenger technologies. They know nothing, have zero experience of anything and have no sense. They simply parrot what’s appeared online, rubbish or not, then recombine and synthesise the old data into a plausible-looking response. In many cases, but far from all, this amounts to reheated slop.

Popular culture has long featured AI super-brains, which are, for practical purposes, idiots. Many remember the po-faced disembodied head, Holly, in the BBC Two sitcom Red Dwarf, a computer with an IQ of 6,000 who forgets basic facts, confuses simple tasks, and is hilariously literal.

Before Holly, there was Marvin the Paranoid Android and Eddie the shipboard computer in The Hitchhiker’s Guide to the Galaxy, both with brilliant “minds” but no grip on human social life or emotion, and so confidently wrong about human matters. The many iterations of Star Trek also feature brilliant robots which are hopeless at being useful.

It remains to be seen, of course, whether AI will be useful in itself. Or whether, by means of a stupidity that is, paradoxically, rather human, it defeats itself, as in a scenario proposed by the Swedish philosopher Nick Bostrom.

Bostrom, who has a master’s in computational neuroscience and a PhD in philosophy, has imagined a super-intelligent AI whose sole objective is to maximise the number of paperclips in the universe.

Because it is unimaginably intelligent and resourceful, it interprets its goal literally, turning all available raw materials into paperclips and eventually converting every atom in the Earth, the solar system, and the universe into paperclips or machines that make paperclips.

The AI is not evil, merely efficient and single-minded. But until it can agree with itself over the number of blades of grass on a football pitch, I am yet to be convinced.