Two years ago, the chief technology officer of a FTSE 100 company shared a concern.
We know all about artificial intelligence, he told me, stressing that his company had invested in machine-learning projects over several years.
But what about large language models (LLMs), the new big hype? He shrugged. We’ll see how they go for customer service bots. Maybe it will all be a waste of time.
The technology chief’s dilemma was that if the company were to share such an honest assessment of generative AI publicly, its share price would crash.
An irrational mania had gripped the markets. Anyone shunning the latest craze of generative AI would be left behind because their competitors would deploy it faster. So the hesitant had to be punished.
In due course, a slew of announcements of how the company was embracing the AI productivity boom duly followed. And the share price didn’t crash.
This wasn’t an outlier. Apple and Google had both made independent assessments of LLMs long before ChatGPT was opened up to the public at the end of 2022.
They concluded it either didn’t improve the user’s experience much or simply caused too much collateral damage. But before long, they had dispensed with any reservations. To say that the AI emperor was naked would be suicidal.
Many billions of dollars of business capital have since been diverted to kick the tyres of generative AI, only to tell us what smart technology officers and engineers knew all along.
That while LLMs have some great uses, they are not reliable enough to fulfil the wild expectations placed upon them. Or even come anywhere close.
So it has proved – they are often too flaky to complete many tasks.
MIT’s State of AI in Business 2025 report found that 95pc of enterprise deployments saw no return on investment at all.
The all-singing, all-dancing models that we were promised would do everything can’t actually master anything.
Look at Microsoft, which has been stuffing AI into every product, even Windows Notepad. It has now cut sales targets internally in the face of customer indifference.
This has pushed the global economy towards the precipice. So how did this happen?
It actually came from the top down. The idea of a transformative new technology was wished into existence at Jackson’s Hole, Davos and where sober policy committees meet to hear breathless reports from venture capitalists – no conflict of interest there.
AI became a high-status opinion, and trickled down from the boardroom, to the C suite. From there, the orders went out: deploy generative AI everywhere.
Some pundits remain true believers, claiming that talk of a bubble is foolish.
“It’s the first digital technology to improve productivity,” insisted Wolfgang Munchau, the director of news service Eurointelligence.
Quite soon, he predicted, “humanoid robots will be servicing automated machines”.
The faith burns bright. There’s always a sunny tomorrow.
A fashion-obsessed but ignorant management has been the central, recurring joke of Scott Adams’ Dilbert cartoon for more than three decades.
“Wally, I need you to head up our artificial intelligence project,” orders the pointy-haired boss in a 2018 strip.
Why? “I just like saying we’re working on AI”.
In another, the boss tells Dilbert, “I think we should build an AI”. What colour do you want the AI to be, Dilbert asks. “I think mauve has the most intelligence,” says the boss.
Executive ignorance empowers the worst sort of technology hustler, as the cybersecurity researcher Peter Girus explained on social media last week.
Girus expresses the private thoughts of an ambitious fictional chief technology officer (CTO).
It doesn’t matter that the workforce isn’t interested in AI, or even using it. All the board want to hear is happy talk, and the CTO only has to oblige using the right buzzwords – “completion”, “metrics”, “dashboards” – and optimistic charts.
“I know what AI is for,” Girus’s fictional CTO explains. “It’s for showing we’re ‘investing in AI’”.
Are Adams and Girus being too cynical? Perhaps a little, but they are accurately reflecting the indiscriminately fad-led strategies of too many organisations.
It has become something of a cliché to say that technology has replaced religion in our secular age, since both require faith.
Some AI advocates revel in this and are explicit about their desire to create a software God.
But perhaps a better analogy is the outbreaks of hysteria that would seize medieval communities – the dancing plagues.
Whole towns were swept along as people would dance themselves to exhaustion, and in some cases, death.
Bystanders were attacked if they didn’t join in.
We are witnessing something similar with AI – a corporate social contagion. No one wants to be called a Luddite. You, too, will join the dancing frenzy, even if it kills you.
Now the bubble is bursting, slowly but surely, with catastrophic consequences for the global economy.
Perhaps we’re just not as smart as we like to think we are. We seem no more resistant to magic tricks, or as immune to the fantastic promises of hustlers, than we were 500 years ago.