When the printing press arrived in Europe in the fifteenth century, it changed who could produce and distribute information. That brought risk. Poor material could spread more quickly. Authority could be challenged. Information could circulate beyond traditional gatekeepers.
But few would now argue that the answer was to keep printing in the hands of the already powerful. The better answer was literacy, responsibility, public debate and wider access to knowledge.
The same principle applies to artificial intelligence. The answer is not simply to treat AI-assisted FOI as suspicious. The better answer is to promote responsible use.
FOI should be about recorded information
A good FOI request is not:
- a speech
- a complaint
- a legal submission
- an invitation for a public authority to explain itself in general terms.
At its best, an FoI request asks for specific recorded information likely to be held by a public authority.
That is where AI can help. A requester may begin with a concern, a suspicion, an article, a piece of correspondence or a complicated factual background. AI can help turn that into a concise request for records. That should be welcomed, not dismissed.
There is also an important role for WhatDoTheyKnow.
WhatDoTheyKnow helps people make Freedom of Information requests for free. It also publishes the requests and responses online, meaning the answer is available not just to the requester, but to everyone.
This public archive is valuable to help:
- people to see what has already been asked
- avoid duplication
- monitor a request progress
- journalists and researchers track patterns.
- others to follow requests, monitor responses and understand how public authorities deal with particular subjects.
WhatDoTheyKnow also explains that users can create a free account, and that once they have an account they can make free FOI requests through the site, subject to its daily limits – making it one of the most useful public tools for transparency in the UK.
AI should be used before the request is sent
The most important use of AI may not be drafting the final request, it may be the step before that. Before sending an FOI request, AI can be asked:
- Is this really a request for recorded information?
- Is the wording too broad?
- Is there unnecessary complaint language?
- Could the same information already be public?
- Could the request be narrowed?
- Is the public interest clear?
That is a responsible use of AI. It may reduce burden rather than increase it.
The discussion has largely been framed from the perspective of burden on authorities.
A practical template
I have therefore created a simple template prompt for drafting FoI requests with AI. If the concern is that AI may generate poor FOI requests, one answer is not simply to criticise AI use after the event. A better starting point is to help requesters use AI responsibly before the request is sent. That is the purpose of this template.
It is designed to do five things:
- focus the request on recorded information;
- remove unnecessary argument and speculation;
- consider whether the information may already be public;
- narrow the request to the most useful version;
- encourage responsible use of FOI.
The template is not legal advice. It does not guarantee disclosure. It does not prevent a public authority from relying on exemptions where appropriate. But it may help requesters ask better questions. And better questions matter.
Conclusion
The problem is not AI-assisted FoI requests but poorly framed requests. Badly presented approaches can be burdensome whether written by a person or a machine. A well-framed request can assist both the requester and the authority.
If AI is used to generate broad, speculative and repetitive requests, it will create problems. But if AI is used to narrow requests, remove argument, avoid duplication and focus on recorded information, it may become part of the solution.
That is why the issue should not be framed simply as AI versus FOI. The better approach is to ascertain how AI can help formulate better FOI requests. I believe it can.