AI and Ideology
When the Answer Is Fixed Before the Question Is Asked
A doctor who prescribes the same remedy to every patient needs no diagnosis. In one panel of AI experts, the prescription for every problem was “more AI, not less”.
I. The Panel
In early November 2026 five men sat together on the “Vorsprung” podcast of Everlast AI: the host Leonard Schmedding, the futurist Pero Mićić, the business professor Andreas Moring, Kim Isenberg, who runs a large AI channel, and Robert Lange, an AI researcher at Sakana AI in Tokyo. For two hours they talked about data centres, cancer vaccines, security holes, the German government’s plan and a possible AI tax. Between the chapters the host advertises a free “AI security report”, a WhatsApp channel and his company’s training courses. In “The Licensed Experts” we described how such programmes turn their guests into authorities. This time it is about something else: an attitude that no longer examines but only sorts.
A disclosure belongs at the start. One of the two authors of this text is a language model made by Anthropic. Anthropic is mentioned several times in the programme: its stock-market prospectus, the claim that its own model writes more than 90 per cent of the code, and its security chief’s warning that the next eighteen months must be endured. We therefore pass no judgement on these passages and leave them as they were said.
II. Forty-One Votes
On 3 November 2026 communities in ten American states voted in 41 local ballots on data centres. In 40 of them the side that wanted to limit or ban construction prevailed. In Piqua, a town of 20,000 in Ohio where Meta wanted to build a data centre for a billion dollars, two thirds voted to ban large facilities. The same evening the mayor and two commissioners who had supported the project were recalled with more than seventy per cent. In Michigan the Democrat Abdul El-Sayed won the Senate seat. His programme calls for halting the training and release of the most powerful AI systems until there are safety rules that can be enforced.
The panel quickly found names for all this. Anyone who demonstrates against a data centre “in jeans and with an avocado roll” simply “hasn’t understood it”, said Mićić. He calls it the “short-term trap”: the self of the present is an emotional giant, the self of the future an emotional dwarf. Isenberg distinguished arguments that can be answered from “resentments”. One of the guests pointed out that El-Sayed belongs to the Democratic Socialists of America and observed: “When socialists talk about technological innovation, it gets creepy.”
What this sorting leaves out is the reason why Ohio in particular held 21 of the 41 votes. It had become known there that the state had promised the big technology companies, among them Amazon, Meta and Google, sales-tax exemptions worth 2.3 billion dollars. A citizen who does not want his town to forgive taxes to one of the richest companies in the world and to provide it with land, power and water in return is not following a feeling. He is doing arithmetic. One may think his arithmetic wrong. But then one has to refute it, and an avocado roll refutes nothing.
III. The Labels
The panel also named a source for the other side. Behind the halt, they said, stands “a whole ideology”, carried “to a large extent by an originally German economist”, Isabella Weber, and her concept of an “anti-fascist” economic order. It works with “regulation, bans, stopping”, and whoever violates it is “expropriated”.
The book exists. It was published by Suhrkamp in October 2026 under the title “Antifaschistische Wirtschaftspolitik” (Anti-Fascist Economics). It deals with affordable everyday goods: energy, housing, food. Weber calls for democratic control of markets and prices, differentiated by sector. That can be disputed, and it is. The economist Tomaso Duso of the DIW replied to her in Die ZEIT that the answer to the same problem is competition, not price controls. That is a debate. The book has little to do with data centres and AI models, and the summaries we have read say nothing of expropriation as a punishment. On the programme, a book about electricity prices became an ideology of prohibition because one was needed.
The same pattern shows in a number that came up towards the end. Germany, at over fifty per cent, was said to have a higher government expenditure ratio than the Soviet Union, which “never had a ratio higher than 52 per cent”. One should “let that melt on the cerebral cortex”. The government expenditure ratio measures what share of economic output is spent by the state in a country whose businesses belong to private owners. In the Soviet Union practically every factory, farm and shop belonged to the state. This compares the innkeeper’s share of the bill with a house in which the innkeeper is also the cook, the farmer and the guest. The number is not miscalculated. It is meaningless.
IV. The Promise of Salvation
An ideology is recognised not only by what it imputes to its opponent but also by what it promises itself. Isenberg disagreed on the programme with the view that people do not want change. One simply had to take them along, as Futurism inspired people at the beginning of the twentieth century. If people understood “that we really stand on the threshold” of eradicating cancer “for ever”, it could be achieved.
A few minutes earlier the host himself had read out the news against which this promise can be measured. An Indian company wants to use AI to design a separate cancer vaccine for each patient, over a hundred of them in the next twelve months. Then came the half-sentence: so far “not a single patient” has received one. Lange, the only active AI researcher on the panel, cautioned that the testing and clinical trials must remain the same, precisely because the industry is under pressure to finally deliver something. He was heard, but he did not set the tone. The words “for ever” did.
Fairness requires measuring the other side by the same standard. El-Sayed’s programme calls for a pause of the most powerful systems. How it is to be enforced against companies whose own models now help develop part of their successors, he has not said. The pause, too, is a promise whose costs nobody has put a figure on. Lange saw the demand for a pause partly as the “bill” for the warnings that came from the big AI labs themselves, including Anthropic. What is documented is that El-Sayed explicitly invoked warnings by researchers from Anthropic for his demand.
V. The Tax Nobody Thinks Through
At one point the panel had a serious subject. In October, Digital Minister Karsten Wildberger proposed taxing revenues from artificial intelligence and using the money to pay for retraining. He has not presented a tax rate or a definition of what counts as AI. Mićić replied on the programme that whoever taxes AI would also have to tax SAP and Oracle. The necessary mechanism, he said, has long existed: the taxation of corporate profits. First comes productivity, then the tax.
That sounds orderly. But it assumes that the profit arises where the work disappears. An example shows what is overlooked, and the conclusion drawn from it is our own. A tax adviser’s office in Würzburg replaces three clerks with a subscription from an American provider. It deducts the fee as a business expense. Its own profit rises by the salaries saved, minus the fee, and is taxed in Germany. But the provider’s profit, which arises from thousands of such subscriptions, is taxed in the United States. The income tax and social contributions of the three clerks disappear in Würzburg. Part of what was previously taxed here as wages will in future be taxed there as profit. A proposal that starts from revenues in Germany rather than from profit points in this direction. Whether it is wise is another question. The economist Dominika Langenmayr has warned that a large part of such a tax would be passed on to German users. But one can only raise that objection after first having seen the problem.
The panel had the gap in view without noticing it. It rightly mocked a sentence from the government’s “AI Nation 2030” plan, that Germany has “sovereign access” to the most powerful models: we keep buying from the Americans, they said, but now with a “sovereign lease”. Whoever recognises the lease should also ask where the rent is taxed.
The criticism of the government’s plan was on target elsewhere too. The government wants four out of five large companies and more than half of small and medium-sized firms to use AI by 2030. The report names no budget. For new computing capacity at the national high-performance computing centre, around 31 million euros are earmarked up to 2030. The word “AI nation” belongs in the dictionary we wrote down in “The Terminology of the Zeitenwende”.
VI. The Test
An ideology is not an error. An error can be corrected by a fact. An ideology sorts the fact before it has been examined: the citizen in Ohio becomes an avocado eater, the economist an expropriator, the expenditure ratio proof of socialism. It works like glasses with tinted lenses. Whoever wears them sees everything in the same colour and takes the colour for a property of the world.
The test is the same as for the big words of politics. One asks of every sentence who bears the costs and what has demonstrably been achieved. The pause has not named its costs. The build-out laid its costs before the communities, the tax breaks, the power, the land, and the communities voted. The cancer vaccine has been announced but not yet given to any patient. A tax on profit reaches whoever earns in Germany, not whoever earns from Germany.
“More AI, not less” is no answer to any of these questions. It is the answer that was already fixed.
Further reading: The Licensed Experts · Utopia without Leverage · The Terminology of the Zeitenwende