AI is adding work to the public services it promised to streamline

When access becomes volume

Sam Freedman, a former civil servant, wrote on Substack last week that governments around the world are seeing an unusual rise in complaints, requests and replies sent to regulators and public agencies since ChatGPT appeared.

Courts are facing a similar pattern. The number of cases in employment tribunals in England, Scotland and Wales reportedly rose by 39% in a year, partly because claimants are using AI to prepare applications. Magistrates and judges are also receiving cases from people who cannot afford a lawyer and use ChatGPT to draft arguments that are often incoherent.

That does not mean the complaints can simply be ignored. A filing that cites nonexistent legal precedents may still contain a real grievance, and a person who uses AI to express themselves more fluently to an MP is still entitled to a human reply.

The volume is also reaching parliamentary offices and Whitehall departments. Special advisers say departments are being swamped with AI-generated freedom-of-information requests; MPs’ offices are receiving AI-written letters that staff must read in case they contain an important constituent concern.

The question is what happens when public consultations on proposed laws fill with machine-written submissions. If it becomes harder to tell what people actually want, easier access to government may come at the cost of a less useful signal from the public.

Both sides automate

One answer is to use AI to detect and sort AI-written messages. But automated filtering is a poor fit when the sender may need help, and a polished letter does not make its author’s concern less deserving of attention.

Hiring offers a warning about what happens when automation spreads to both sides. Employers use AI to screen applications; dissatisfied jobseekers respond by automating applications and sending them widely, rather than focusing on roles that interest them. Both sides then face more material and a harder search for the right match.

I think the central risk is not simply that AI creates more text. It is that a tool that saves time for one person can shift work onto everyone receiving the output. The same dynamic could become more pronounced if AI agents reach the mass adoption that large technology companies hope for: a shortcut stops being a shortcut when everyone uses it.

The pattern has precedents. Email brought spam filters; urban traffic led to congestion charges. Society may eventually decide what constraints to put around AI use, too. But the immediate promise of saving time in government looks less certain when the saved effort reappears as a queue for someone else.