In the UK, MP offices, local authorities, schools and lawyers are describing a growing problem: a large share of incoming correspondence is now produced by artificial intelligence. According to a senior adviser working in an MP's office, roughly half of incoming emails are chatbot-generated, and these texts are harder to read and parse. The flow that began as a "trickle" in 2024 has now turned into a "downpour." This new "AI congestion" hitting public institutions shows how the technology's promise of accessibility can be turned on its head.
What happened?
MP offices, councils and schools are facing a flood of correspondence made up of complaints, information requests and legal objections. A significant portion of these texts is generated by large language models. Even experienced advisers say the rising volume has made the work heavier and more time-consuming. The texts are often long, formal and far more complex than their content warrants.
Why does it matter?
The real issue here isn't a wave of "spam"; it's a problem of asymmetry. Writing a complaint, an appeal or a freedom-of-information request used to require time, language skills and effort. Generative AI has reduced that cost to almost zero. Now, in a matter of seconds, pages of text dressed up in legal jargon can be produced. Yet on the receiving end, the human capacity to read, assess and respond to this correspondence is not growing at the same pace.
This imbalance can be read as a familiar logic from the digital world spilling over into bureaucracy. In denial-of-service attacks, a system is overwhelmed with more requests than it can process. Something similar is happening here: the gulf between the cost of production and the cost of response is paralysing the institution's function. The intent doesn't even have to be malicious. A tool used in the name of accessibility can, in aggregate, clog the system.
The second layer concerns the quality of the texts. AI-generated correspondence can be fluent but hollow; sometimes it cites regulations that don't exist, sometimes it strings together legal claims irrelevant to the matter at hand. That is exactly what makes the advisers' job harder: the text looks serious, so it can't be ignored, but sifting through it requires human labour. Separating genuine grievances from automatically generated noise is becoming an increasingly expensive task.
Who is affected, and how?
The heaviest burden falls on units whose buffer capacity is already thin. Small MP offices, under-resourced councils and schools with limited administrative staff don't have the manpower to absorb the surge in volume. This creates a second risk: genuine and urgent requests getting lost beneath the pile of automatically generated material.
| Dimension | Before AI | After AI |
|---|---|---|
| Cost of producing correspondence | High (time, language skills) | Almost zero |
| Length and format of texts | Variable, mostly short | Long, formal, "legalistic" in tone |
| Institution's sifting burden | Manageable | Rising rapidly |
| Visibility of genuine requests | High | Risk of being lost in the noise |
This table doesn't work in only one direction. For citizens who face language barriers or are unaccustomed to formal correspondence, these tools provide a real benefit; expressing a legitimate request in proper language is now easier. The problem is that the same convenience inflates the volume in an uncontrolled way. In other words, the technology is both democratising and congesting — both at once.
What's next?
The most likely scenario is that institutions will, on the defensive side, turn to AI once again. Systems that pre-classify incoming correspondence, summarise it and flag duplicate texts will probably become widespread. This creates a strange spiral: text produced by AI is in turn read and sifted by AI. However efficient it may seem, this arrangement carries its own risks — an automated filter mistakenly weeding out a genuine grievance could exact the heaviest price.
The second thread to watch is the response of rule-makers. Mechanisms such as rate limiting, identity verification or an "AI-generated" disclosure requirement could come onto the agenda. But none of these is easy; they carry a high chance of upsetting the balance between the right to access and the protection of capacity.
Ultimately, this is not a technology malfunction but an incentive problem. As long as production is free and response is expensive, the congestion will persist. The solution is likely to be found not in a single tool, but in institutional design that accounts for this imbalance. AI's real test in public services won't be producing text; it will be making the flood of text it produces manageable.
Yorumlar (0)
Henüz yorum yok. İlk yorumu sen yaz.
Yorum yapmak için Sinyal'i indir
Yorumlar Sinyal hesabıyla yapılır. Mobil uygulamada giriş yap, yorum bırak.