For centuries, Britain has exported some remarkable things to the world: democracy, parliamentary government, the English language, industrial innovation, cricket and even queueing, although that last one still depends heavily on geography and how urgently someone feels about getting home on public transport during rush hour. We have also exported politicians, albeit on a temporary basis…
Whether our House of Commons should be classed as a success story or a long-running work in progress depends very much on your tolerance for televised shouting and carefully avoided questions. We are presented with random people who are, by most objective measures, intelligent, capable and highly motivated, and we elect them into a system that gradually rewards survival over clarity, performance over substance and confidence over correctness. Within a surprisingly short period of time many of them acquire a set of behaviours that would be alarming in almost any other profession. Direct questions are met with indirect answers. Simple problems are expanded into complex narratives. Mistakes are not corrected so much as reinterpreted into something that sounds like a different kind of achievement. And all of it is delivered with the steady assurance of someone who is absolutely certain that accountability is a matter of interpretation.
At some point most systems would begin to suspect the machinery might be part of the problem. Politics instead tends to assume the users are at fault for not appreciating the machinery properly.
Which is where the thought experiment begins to intrude in a slightly uncomfortable way. Because while Westminster continues to operate on a structure that may feel increasingly medieval in its reliance on ceremony, interruption and ritual disagreement, AI has quietly started doing the sort of work that used to be held up as proof that humans alone could handle complexity. It reads vast quantities of information without getting bored. It compares competing claims without becoming emotionally invested in any of them. It produces summaries, drafts, forecasts and analyses without once needing to pause for a strategy meeting about messaging.
At first this comparison sounds like provocation. It has the tone of something said to wind people up rather than something intended to be taken seriously. Yet the more time you spend looking at the actual mechanics of government, the harder it becomes to dismiss entirely. Because so much of modern governance is already procedural, data driven and rule bound. It is forms, reports, consultations, committees and responses. It is less about inspiration and more about processing reality at scale.
And then you think about the most familiar interaction most people ever have with politics, which is not Prime Minister’s Questions or some grand constitutional moment, but the simple act of trying to get an answer from their elected representative. A citizen writes an email which may be sometimes calm, sometimes frustrated, but yet sent because people still carry the instinct that clarity might improve the chances of being understood. I am sure that I am not the only one who feels that our emails are sent into the system and disappears into a process that feels vaguely like filing and slightly like waiting for a train that may or may not be running. Eventually a response arrives; in my case 7 months after the action was needed (vote on a debate related on online harm), the reply generic and clearly obselete… By this time, water under the bridge.
Now place artificial intelligence into that exact exchange – I don’t think that I would have waiting longer than maybe 7 hours… I don’t think that it would be either a placeholder acknowledging receipt or a polite deferral. I think it would be an answer: relevant and specific. If there were any uncertainties, they would be clearly stated; if there are competing interpretations, they would be laid out. If there are contradictions in existing policy, they would be flagged without embarrassment or defensive framing. That shift alone would alter the public relationship with government more than most constitutional reforms ever manage. Not because it introduces something radically new, but because it removes a layer of delay that has quietly become normalised to the point where people no longer question it.
Then there is money, which in politics is never just money but a kind of extended philosophical argument about what society chooses to tolerate. Maintaining 650 Members of Parliament is not simply a matter of salaries. It is staff, offices, accommodation, travel, pensions, administration and the entire invisible architecture that surrounds decision-making at national level. The figures are large enough that they tend to lose meaning when presented directly, which is usually a sign that they are large enough to warrant attention rather than avoidance.
The appeal of replacing that structure with artificial intelligence is not that it becomes free because nothing at this scale is ever free, but the ongoing cost structure changes shape entirely. Instead of hundreds of individual political offices each replicating similar administrative needs, you move towards a centralised technological infrastructure with distributed access. There is still expense, still complexity, still maintenance, but the duplication begins to shrink and I am sure that is where a surprising amount of public spending quietly lives.
Of course, the more interesting effect is not the headline saving but the removal of the small, familiar inefficiencies that accumulate over time. The expense claims that make headlines. The administrative overheads that exist because systems were never designed to talk to each other. The constant translation of information from one department’s language into another’s. The quiet friction that exists everywhere in human organisations simply because humans are involved in all of them.
Artificial intelligence does not eliminate complexity, but it does not add unnecessary friction for social reasons either. It does not need to protect its position in a hierarchy or maintain ambiguity for political safety. It simply processes what is in front of it. Which brings us, inevitably, to behaviour.
One of the more quietly radical aspects of the idea is not that AI would be more intelligent than politicians, but that it would be less socially compromised. Human political systems are shaped not only by policy objectives but by incentives that have very little to do with policy. Reputation management, party loyalty, electoral pressure, media cycles, internal positioning and long-term career survival all sit alongside the actual job of governing. These forces are not aberrations, they are part of the system itself.
Artificial intelligence has no equivalent internal theatre. It does not need to win favour. It does not need to maintain a public image. It does not need to anticipate the next election. It does not get promoted or demoted based on visibility. It does not fear embarrassment. It does not experience ambition in the human sense of the word. That absence removes a great deal of noise from decision-making, even if it introduces entirely new questions about control and accountability.
Lobbying, for example, becomes a different kind of problem. In the human version of government, influence often operates through proximity, access and persuasion. Conversations matter because conversations can be persuasive. Relationships matter because relationships create pathways that data alone does not capture. In an AI-mediated system, every input can be logged, traced and weighted transparently; influence does not disappear, but it becomes harder to hide inside informal exchanges – it becomes data rather than atmosphere.
That sounds clean on paper, although anyone who has ever worked in or around politics will know that cleaning up influence is never just a technical exercise. It is also a cultural one. People adapt to systems and find new routes through them. The question is not whether influence would vanish, but whether it would become more visible, and visibility is often the first step towards control.
Scandal, meanwhile, would change shape rather than disappear. Human history suggests that whenever one category of misconduct becomes harder, another emerges to replace it. Expense scandals and personal indiscretions might decline, but algorithmic bias, data governance failures and system design decisions would become the new battlegrounds. The theatre would shift from human behaviour to technical accountability, which would not necessarily be calmer, just differently understood.
And yet, despite all of that, the most persistent objection remains unchanged. Politics is not only about outcomes, but about representation. It is about the idea that people should have a voice in the decisions that shape their lives, even when those decisions are inconvenient, inefficient or emotionally complicated. Artificial intelligence can simulate preference, model trade-offs and optimise outcomes, but it does not belong to anyone in the way democratic representation is supposed to belong.
That is the tension sitting underneath the entire idea. The more capable the system becomes at producing rational outcomes, the more visible the question becomes of whether rationality alone is enough. A perfectly optimised government that nobody feels connected to would still be a failure in democratic terms, even if it worked flawlessly in operational terms.
So the argument eventually circles back on itself. Artificial intelligence exposes weaknesses in political systems that are already widely recognised. Delay, inefficiency, inconsistency, and the persistent gap between public expectation and institutional response. At the same time, those weaknesses exist inside a framework that is not purely technical. It is historical, cultural and emotional; built on the assumption that governance must be human even when it is imperfectly human.
Perhaps the most realistic outcome is not replacement but exposure. A system that constantly compares human decision-making against machine-level consistency would force a level of scrutiny that modern politics is not fully prepared for. Promises would become trackable in real time, statements would carry immediate cross-referencing, decisions would be evaluated against evidence at a speed that removes the usual gap between claim and correction, less tearful resignation speeches outside Number 10…
In that world, the question is no longer whether artificial intelligence should replace Members of Parliament. It is whether Members of Parliament could continue operating in a system where every weakness is immediately visible and permanently recorded and that, perhaps, is the real point. The discomfort does not come from imagining machines in Westminster. It comes from imagining Westminster under conditions where forgetting things is no longer part of how politics works.
What do you think?
