Author: Joshua Rothman
Originally published in The New YorkerDecember 12, 2025.
The story of boom and bust is familiar, but disconnected from the possibilities of new technology.
Over the past few months, I've been introducing artificial intelligence into my 7-year-old son Peter's hobby life. He takes a coding class on Saturdays, where he recently created a version of rock-paper-scissors. I'm seriously thinking of making more sophisticated games at home. After letting ChatGPT and Claude know his skill level, they immediately suggested next steps.
…
I attended college from 1998 to 2002 at the height of the first dot-com boom. I paid much of my college tuition by running a small startup with my roommate. I mainly created websites and applications for other startups. Back then, as now, there were countless companies offering products that were out of proportion. (We worked for several of them.) It was easy to predict that many of these businesses would fail, leaving investors of all sizes with significant losses. Still, there is no doubt that the underlying technology, the Internet, is powerful. I can't help but say the same about today's AI.
Still, the story of artificial intelligence is even stranger when compared to the dot-com boom. When the Internet came out, people didn't know how to make money using it. Still, I felt that the technology itself had been completed to some extent. It was clear that connectivity would become faster and more widespread. Additionally, there are already widespread uses for the Internet, including streaming media, e-commerce, collaboration, and cloud storage. (For example, around 2000, our small company was hired to create a workplace collaboration system that had many of the features we now associate with Slack.) Over the next several decades, the engineering effort required to create the modern Internet would be enormous. For example, building a cloud requires extraordinary ingenuity. However, the basic nature of the Internet itself was largely determined from the beginning.
With AI, it's different. From a scientific perspective, the work of building and understanding AI is not yet complete. Experts in the field disagree on important questions, including whether scaling up today's AI systems will significantly improve their intelligence. (Perhaps a new system formed by further breakthroughs will be needed.) They also disagree on conceptual issues, such as what “intelligence” means. They have strongly different views on the crucial question of whether today's AI research will lead to the invention of systems capable of human-level thinking. People working in AI tend to express their opinions clearly and forcefully, but consensus is still elusive. Whoever creates the scenario differs in opinion from most of her colleagues. As researchers try to build better AI and see what works, they will empirically answer many questions about AI. So the AI bubble is not just a bubble, but a collision of scientific uncertainty and evolving business thinking.
There are currently two big unknowns about artificial intelligence. First, we don't know if and how companies will be successful in extracting value from AI. They're trying to figure it out, so they could be wrong. Second, we don't know how smart the AI will be. However, there are some clues about the first unknown. I can tell you from first-hand experience that having AI available can be extremely useful. that it helps learning. That it can make you more competent. It means helping to utilize human capital more effectively and even expand it. We can also say with some confidence that AI cannot do many important things that humans do, and that it is better at enabling humans than replacing them, except in certain narrow circumstances.
Click here to continue reading this article.
