State of the AI bubble. Economists weigh in

It’s an awkward thing. On one hand, like many other computer programmers, I root for the success of Artificial Intelligence, by which I mean the specialty within Computer Science. But in most discussions today, “Artificial Intelligence” is used as a synonym of ChatGPT, Claude, and other tools and companies in the current generative AI race.

So, on the other hand I find myself shaking my head when I hear people saying that these tools are “like having a PhD on any subject you need”, are ready to take over coding from humans, or are on the verge of “AGI” or “the singularity”. An idiot whose name I don’t wish to recall suggests that we have already entered “the gentle singularity” and “crossed the event horizon.”

I’m seeing more and more analysis from economists and economics journalists that mention the word “bubble”. I hope this starts to bring some balance and accountability.

I wish we could have clear eyed debates about what these generative AI tools are good for, what they’re not good for, and what their limitations are. But who has time for that when there’s so much money changing hands?

The US economy and AI. Sell shovels

The investment on AI infrastructure is so large that it has been a de facto stimulus for the US economy. Some commentators are starting to worry about the potential consequences for the economy when the bubble bursts.

This article by the economist Kyla Scanlon gives a good overall picture: How AI, Healthcare, and Labubu Became the American Economy.

The Financial Times has a great article (paywalled): America’s top companies keep talking about AI — but can’t explain the upsides.

The article shows that the “magnificent 7” (the tech giants) are driving the stock market, and that the US stock market, discounting them, looks underwhelming. Most of the companies that have adopted AI don’t seem to have benefited from it. Outside the magnificent 7, the companies that have benefitted most from AI are the ones that help build the massive data centers.

Filings do reveal that the companies able to give clear AI upsides include those that serve the rising AI-driven data centre boom. Energy companies First Solar and Entergy cited AI as a demand driver.

Equipment manufacturer Caterpillar reported that its energy business was benefiting from supporting “data centre growth related to cloud computing and generative artificial intelligence”.

As they say, “During a gold rush, sell shovels.”

This little comment in the FT article made me laugh with sadness:

One of the most common benefits cited by companies is that AI will help differentiate their products.

Yeah, that’s it, we’re all going to differentiate by … all deploying one of a handful of generative AI products that work by … homogenizing all the internet into a predictive text model.
A fool-proof plan.

If you’re into podcasts, this talk is interesting: This Is How the AI Bubble Could Burst.

It explains in some depth the level of penetration of AI into the American economy, and into securitization schemes.

Strained optimism

Then there’s this … thing from the Wall Street Journal. Which. Huh: Stop Worrying About AI’s Return on Investment.

Nearly three years after the start of the artificial intelligence boom, business technology leaders are starting to change their thinking on return on investment. The new wisdom? Don’t worry so much about AI’s ROI.

Only after an AI project scales, or expands across an entire organization, will most corporate technology leaders be able to determine the technology’s true ROI, some experts say.

The article makes it clear it’s quoting from industry “leaders” who were at an event organized by the WSJ. I guess it would have been bad form from the journal to call out the magical thinking in their own event.

Sobering up

Newsweek conducted a series of interviews with AI luminaries in February 2025 to assess the impact and future of AI. They’re worth listening to.

In the interview with Rodney Brooks, he talks, among other things, about the current state of humanoid robots:

Robots do not have good general-purpose picking. 50% of the robots that are doing picking have suction cups, so that’s not dexterous. The others have parallel jaw grippers. There has not been much progress building robot hands.

For most of the things you need a human to do the picking.

And in the interview with Yann LeCun, who is famous as one of the fathers of machine learning, he brings some perspective by comparing to animals:

You can think of intelligence as a couple of things … an ability to acquire new skills quickly, possibly without learning.

As of today we’re really at that low level of direct connection of perception to action. We haven’t cracked planning

Where is a robot that is as good as a cat at understanding the physical world?