What if the Federal Chancellor and his ministers discovered AI for themselves?

The first phase would probably be rather German

You can imagine it quite well.

The Federal Chancellor discovers AI. Not just in passing, because someone on a plane showed him how to have a speech summarised. But properly. With a demo. With enthusiasm. With an adviser saying in a presentation that the biggest administrative revolution since the introduction of the fax machine is now at hand, only this time hopefully with a bit less fax.

Then what probably happens first in Germany is what traditionally happens in Germany with new big topics: there is a summit, a task force and shortly afterwards a paper with guidelines for the strategic development of a framework for the responsible use of AI in an interdepartmental context.

So, to begin with, a great many very German sentences.

That is not even unfair. Politics has to be careful with a topic like this. It can hardly decide on a Wednesday morning that from now on a language model will help draft half the machinery of government, and on Thursday we’ll see what the Federal Constitutional Court says.

But behind the predictable stack of paper there is still a real question. Because AI in politics is not just another tool. It shifts who prepares, who filters, who drafts and, in the end, who gains interpretive authority over complex matters.

What AI would actually change in politics

Politics consists of text work much more often than it looks from outside. Draft bills, amendments, memos, briefing notes, position papers, minutes, voting aids, press statements and the daily art of wording a sentence so that three coalition partners consider it agreeable and still nobody knows exactly what it means.

In exactly this world, AI is of course highly attractive.

It can read texts faster than whole units, draft variants, cluster points of dispute, write summaries, sort counterarguments and turn mountains of material in minutes into what used to keep the light on somewhere in the ministry until half past midnight.

That is useful. Very much so.

But it would be a mistake to confuse political work with text production.

An AI can effortlessly draft ten wording variants for a delicate statement. But it doesn’t bear political responsibility if variant eight sounds like technocratic coldness the next morning and variant three is understood across the country as a loss of touch with reality.

Nor can it sense the mood in a country. It doesn’t notice when a formally correct sentence lands politically completely wrong. It doesn’t know when people don’t need a better wording but the feeling that someone has actually understood the situation.

And this is exactly where human intelligence suddenly becomes quite important.

Human intelligence would have to do more in politics, not less

When AI moves into political processes, some abilities become more valuable that were scarce already.

Judgement, for example.

A machine can sort options and justify with impressive calm why measure A works more efficiently than measure B. But it has no political conscience. No sense of what is reasonable to ask of people. No understanding that a solution can be arithmetically clean and still socially explosive.

Then responsibility.

In politics it is not irrelevant who decides in the end. It isn’t enough for a system to produce a plausible proposal. Somebody has to put their name under it, stand up for it and, if in doubt, be criticised for it. Democracy doesn’t live on decisions appearing as smooth as possible. It lives on being able to name those responsible.

Then a sense of reality.

The Chancellery and the ministries already move in a world in which very many problems first appear as a document. If systems now join that eagerly build new texts out of texts, we need people who regularly remind themselves that out there real authorities, real schools, real railway stations and real citizens exist, which unfortunately don’t work as they do on the org chart.

And finally the ability to say no.

Not every technically possible automation belongs in state processes. Not every scoring is a good idea. Not every prediction should prepare administrative action. And not everything that looks more efficient is automatically better in a democracy.

The temptation would of course be enormous. A technology that reads fast, writes fast, sorts fast and never asks whether you really still need a commissioner, a coordinator and two rounds of coordination for the same matter looks like a small miracle in Berlin at first.

That is exactly why caution would be in order. Because democracy is deliberately slower, more contradictory and messier than a well-optimised decision process. That is not a flaw. It is a protective function.