We are immersed in the greatest technological experiment ever and that has created the biggest bubble ever. Last November, a former chief economist for the IMF estimated the size of the bubble to be $35 trillion. There have been some dips since then, but the bubble has persisted despite a war in the Middle East and studies by MIT, Atlassian software and others showing that businesses aren’t benefiting that much from AI.

More recently, interviews reported by The Guardian report:

AI researchers, economists and tech workers said that essentially, we’re all living through an experiment. Over the next few years, tech companies’ experimentation with AI will probably lead to several critical outcomes: more job cuts across industries, unforeseen consequences from overreliance on AI and a fundamentally different model of work.

Even one of the creators of this hype, Ethan Mollick, an associate professor at the Wharton School of the University of Pennsylvania, agrees. “The maximum hype you have right now, which is that AI is replacing people, is not true. But it’s also not true that AI will never threaten jobs. It’s going to be complicated.” Well, experiments are often complicated.

A new global survey of 3,750 executives and employees across 14 countries that was funded by SAP, “finds that more than 54% of workers bypassed their company’s AI tools in the past 30 days and completed the work manually.” Another 33% “haven’t used AI at all. Combined, roughly eight in 10 enterprise workers are either avoiding or actively rejecting the technology their employers are spending record sums to deploy.”

Behind these numbers is a misunderstanding by executives. They are “blind to how employees really feel. What the early enthusiasm covered up is now visible in the numbers. Only 9% of workers trust AI for complex, business-critical decisions, compared to 61% of executives, which is a 52-point trust chasm.” Some 88% of executives say their employees have adequate tools, but only 21% of workers agree, which is a 67-point gap. In other words, “executives and their employees are describing fundamentally different companies,” which previous studies have also found.

Steve Hanke, a Johns Hopkins economist, says: “AI didn’t deliver. Welcome to the real world. Forget the AI bubble. You know, it didn’t deliver. You look at all the surveys and yeah, everybody’s using it a little bit, but you dig into it and it hasn’t done much.”

We also know that most AI consumer services aren’t working that well. A recent study funded by the New York Times found that AI-generated summaries are accurate around 91% of the time, meaning that you had better check the answers that AI gives you. Unfortunately, most people don’t.

Another study found that only 8% of users actually double-checked an AI’s answer. Why? Because large language models adopt an authoritative tone and can confidently present fabricated information as a fact when in reality it can’t immediately give straight answer. “Add the convenience that Google’s AI Overviews offer, and it’s easy to imagine untold numbers of users taking its summaries at their word.”

One result of users believing what AI tells them is health anxiety. Many believe they have a health issue, either from a doctor or a friend, and they check it out using AI, and keep checking it out until they find the most distressing ailments. Some estimate that health anxiety affects more than 12% of the population. Many more people struggle with other forms of anxiety that can be exacerbated by AI chatbots. Even Sam Altman admitted AI causes these types of problems.

A perhaps bigger problem is the impact of AI on the critical thinking of young people. The younger the AI users are, the smaller the chance they will ever develop critical thinking skills. And we can see those students now in colleges. They have spent the last few years depending on ChatGPT to complete homework and write essays and they can’t think critically.

For instance, class discussions are becoming increasingly homogenized with students giving answers right out of ChatGPT and others searching frantically on their laptops for those answers. “Everyone now kind of sounds the same,” one Yale student said. “I feel like during my freshman year in college, I would sit in seminars where everyone had something different to contribute. Although people would piggyback off each other, they approached from different angles and offered different commentary.”

Numerous studies have come to similar conclusions about AI’s impact on human expression. A paper published in a journal called “Trends in Cognitive Sciences” argues that “LLMs dull the ways their users approach issues, deploy language, and reason through problems. When we use AI chatbots to think, the authors posit, we’re silently exchanging our own human thoughts for LLM output: a homogenized aggregate of our chosen model’s training data.”

What about coding you might ask? Isn’t AI-assisted coding providing a huge productivity boost, so much so that young coders are being laid off or not being hired? Listen to what AMD’s Stella Laurenzo says about her role in this great AI experiment: “Claude cannot be trusted to perform complex engineering tasks.” Her team reached that conclusion by “referring to months of logs from the very consistent, high complexity work environment in which they use Claude Code.” “Every senior engineer on my team has reported similar experiences/anecdotes,” since Anthropic revised Claude code a month ago.

The experiments continue, giving us a very mixed view of AI tools. AI isn’t terrible; AI-assisted coding will likely succeed, but AI isn’t worth $35 trillion either, or even a few trillion. And investors might not notice for a very long time.

Heck, they haven’t even noticed that the Iran war continues. Just like a little hype from Elon Musk and the other tech bros can keep the AI bubble inflated, daily positive comments from President Trump can prevent people, including investors, from noticing that very little oil is passing through the Strait of Hormuz.

Jeffrey Funk is currently completing a book with Gary Smith entitled “AI Slop, Snake Oil, and the Bubble of All Bubbles”