The AI backlash may be what AI needs.

The backlash has already begun. Not as one coherent idea, but as a collection of different frustrations that are starting to attach themselves to AI.

There are the data centres: huge, ugly, resource-hungry buildings appearing in local communities, increasingly associated in the public mind with the demands of AI.

There are people at work being told they must use AI, sometimes with their usage measured, whether or not it actually helps them do their jobs.

There are organisations that started with “we need to use AI” rather than with a problem worth solving. The big consultancy firms whose own work is threatened by AI are pursuing income streams extracted from organisations who fear being left behind. However, we are inevitably going to hear more stories of expensive AI programmes failing to deliver the benefits that were promised…probably the 70% fail trope.

Add concerns about cyber security, autonomous agents, surveillance, job losses and the extraordinary concentration of power in a handful of technology companies and their billionaire founders.

These things aren't all connected. But in the public imagination they increasingly are. And “AI” now largely means generative AI, so a remarkably wide range of anxieties is being attached to the same label.

For those of us who are enthusiastic about AI, I think the backlash is something we need to take seriously rather than dismiss as resistance to change.

Some of the concerns are legitimate. Blind enthusiasm isn't much better than blind scepticism. Leaders introducing AI need to start with real problems, understand the risks, put sensible governance around it and pay attention to what the experience feels like for the people affected by it. But there is another risk.

I've already heard people say, quite confidently, “AI doesn't work.”That is a very different claim.

A particular AI project may not work. A business case may have been nonsense. A workflow may not have needed automating in the first place. None of those things tells us very much about the underlying technology.

Even if development of frontier AI models stopped today, there is an enormous gap between what the technology can already do and what most organisations have learned to do with it.

Closing that gap is increasingly less about waiting for a better model. It is about integration, redesign, skills, governance, change leadership and, in some cases, reinventing how an organisation meets the needs of its customers.

There is a useful historical parallel. When the dot-com bubble burst, a great deal of hype and bad investment disappeared with it. The internet didn’t. Perhaps an AI backlash will do something similar. It may clear away some of the nonsense.

The danger is that, once the hype subsides, organisations mistake disappointment with the first wave of AI for evidence that the underlying change has gone away.