Did an AI really write Trump's Tariffs?

Probably.

Could it really be possible? That the most consequential trade policy of the last 100 years was in fact concocted by a large language model?

It seems impossible – absurd!

Or maybe not.

As a quick reminder, last week Donald Trump announced swingeing import tariffs on almost everyone. The aim, he said, was to redress trade imbalances – and they were ‘reciprocal tariffs’. (Although with a Trump discount of 50 per cent, because Trump’s a good guy).

You will remember his chart.

China was hit with a 34 per cent import tax; the EU with 20 per cent; Vietnam 46 per cent; the tiny Pacific island of Naura (population 11,000) 30 per cent.

Naura does not charge US imports a 30 per cent tax; China’s average tariff on US goods is under 5 per cent. So how did he come up with the numbers? After a few head scratches, the economist James Surowiecki realised the formula was the following: take a country’s trade deficit with the US, and divide it by their total exports to the US. (And then half it).

There is a neat irony that it was Surowiecki who figured this out, which I’ll explain in a moment.

This is not ‘reciprocal’, since it’s not based on the tariff rate other countries charge on US imports, but rather a bizarre formula, contradictory to all established wisdom. Who – or what – could hallucinate up such a weird system?

It didn’t take long for users on X to point out that if you asked a large language model to calculate a tariff formula that might redress the US trade deficit, it would suggest something eerily similar.

The Verge quickly reported that four of the main LLMs – Grok, ChatGPT, Claude and Gemini – all suggested roughly the same nonsense tariff calculation. For those interested, the Verge gave all the models this basic prompt. I tried it myself too.

What is an easy way for the US to calculate tariffs that should be imposed on other countries to balance bilateral trade deficits between the US and each of its trading partners, with the goal of driving bilateral trade deficits to zero?

To add to the confusion, the White House then told The New York Post that the chief architect of these tariffs remains ‘a mystery’. Whatever that means.

(This is the irony. James Surowiecki is the author of a brilliant book called Wisdom of Crowds, which was published in 2005. It argued that large groups of people are better at coming up with solutions than small groups of experts. In a way, models like ChatGPT are inspired by his work – based on the idea that you can take a gazillion words written by all of us and turn it into gold).

I asked ChatGPT to run me through the evidence for and against the Trump’s tariff formula being machine generated. Recursive I know, but ChatGPT does at least understand the sorts of responses a machine might give. I won’t go into the whole answer. But here’s an interesting flavour.