Don’t be fooled. The AI that was supposed to save us time is doing just the opposite | Gaby Hinsliff

Another morning, another avalanche of emails. Even after a guilty cull, my inbox is sitting at well over 7,000, but lately I’ve started subscribing to the Quentin Crisp theory of housework: why bother, when after four years or so you stop noticing the accumulating dust? Empty inboxes just fill up again anyway.

In theory, email was supposed to save hours of meetings, phone calls and standing over photocopiers. Instead, we just spend hours deleting stuff people wouldn’t have bothered telling us if it had meant picking up the phone, but which they can now spam to their entire database at the click of a button. The glorious ease and convenience of being able to reach the world in seconds came, we now know, with the inconvenience of them being able to reach you right back.

And therein lies the challenge to the government’s hopes of AI helping to transform the machinery of the state on the cheap, saving not only time but billions of pounds in public money. It’s early days, but AI seems not so much to be streamlining public service delivery as gumming it up – partly because, as was once the case with email, the public is using it right back at them.

In a Substack post last week, the former civil servant Sam Freedman noted that governments around the world seem to be experiencing unusual surges in complaints, requests and responses to regulatory and official bodies since the adoption of ChatGPT. It’s the same story in the courts, with employment tribunal cases in England, Scotland and Wales reportedly rising by 39% in a year thanks in part to claimants using AI to draft claims, while magistrates and judges face a wave of people who can’t afford lawyers using ChatGPT to build cases that often don’t make sense. And all of this needs patient disentangling by a human, on the grounds that just because a defendant is citing precedents that don’t exist, it doesn’t mean there isn’t a genuine grievance buried in there somewhere.

Whitehall special advisers complain of departments drowning in AI-generated freedom of information requests, an abuse of a system originally intended to empower individuals. MPs’ offices are similarly swamped by AI-generated letters, all of which need to be read by humans for fear of missing something important from a constituent.

How long before public consultations on changing the law become so spammed by slop that they are in essence useless as a measure of what the country actually thinks? There’s a fine line between the noble aim of making it easier for individuals to interact with an otherwise remote state, and making it so easy that the result is an uncontrollable firehose of meaningless machine-generated guff.

The obvious answer to this kind of AI flooding is getting AI tools to detect or sift it, but that doesn’t always work when dealing with potentially vulnerable human beings. Constituents who use AI to express themselves in more fancy-sounding language when writing to their MP, for example, still have the right to expect a personal response from a human being.

Employers who thought they were being smart by getting AI to sift job applications, meanwhile, are reaping what they sowed, as resentful applicants automate their end of things too and spray-gun applications across the board, rather than concentrating on the jobs they really want. Now neither side can separate the wheat from an avalanche of chaff.

The risk of paralysis is likely only to increase if the public starts adopting agentic AI in the kind of numbers big tech fondly imagines. Early adopters might well find agents save them time and hassle. But shortcuts stop being shortcuts once everyone else is using them too, and risk turning into bottlenecks instead. Think of cars, which have vastly shortened journey times compared with travelling by horseback, except in cities at rush hour when too many other people are also trying to shorten theirs. By 2018, central London daytime traffic was moving on average at just over 7mph – about the same speed as a trotting pony.

Historically, humans have always learned to develop ways around problems of our own making. We have spam filters for email, congestion charges to keep cars out of cities, and in the long term we could well figure out what kind of frictions we need to put back in the system to make it possible to live with the impact of AI.

But for now a useful rule of thumb, if improbable claims start to be made for savings due to AI, will be that more isn’t always better; faster and easier isn’t always more convenient; and time saved in one section of public services has a horrible habit of popping up in another one. Let the Whac-A-Mole begin.