The AI boom is shaking up the tech industry and moving markets. But is it all a mirage?

And for what? Startups and tech titans alike hope generative AI will be able to perform a huge range of tasks such as providing customer service, discovering miraculous new drugs, and writing complex computer code.

Enthusiasm about the technology’s startling capabilities has also led to urgent calls to regulate the industry amid concerns about potential threats to workers and society.

In 10 years, annual revenue from generative AI could exceed $1 trillion, according to a widely cited report from Bloomberg Intelligence. Consulting firm McKinsey was less precise but even more optimistic, putting the annual economic impact at $2.6 trillion to $4.4 trillion. And analysts at Morgan Stanley were willing to go even higher, foreseeing a $6 trillion gain.

But the glowing forecasts and huge investments are starting to draw scrutiny and trigger warnings from those who have witnessed past tech bubbles, from the dot-com bust to the recent cryptocurrency crash to prior mini-bubbles fueled by AI.

“We should not take these claims of multiple-trillion-dollar industries seriously, because these are exactly the kinds of claims you see made about new technologies all the time, and they almost never come true,” said Lee Vinsel, a professor at Virginia Tech who studies the history of technology and tech bubbles. “It’s like we just have collective amnesia.”

It was just two to three years ago that bitcoin and cryptocurrency tech seemed poised to take over the world economy. After the value of outstanding bitcoin exceeded $1 trillion, boosters predicted the market would hit $100 trillion. Instead, bitcoin’s market cap now is less than $700 billion, and high-profile crypto failures like FTX have dominated headlines.

The internet bubble of the early 2000s vaporized even more money — $5 trillion by some estimates — though the industry eventually came into its own.

And the Boston area has had its own experience with AI fueling hyperspeculation.

In the 1980s, researchers at MIT working with a programming language called LISP started writing AI applications. Dozens of startups emerged around Kendall Square, then dubbed “Intelligence Alley.” But within a decade, the AI programs had failed to catch on and leading companies such as Thinking Machines and Lisp Machines went bankrupt.

Generative AI supporters argue that this time is different. The underlying “large language models” have improved as computers have become faster and capable of incorporating more data. And the apps’ widespread availability means a lot more potential customers are aware of the technology.

But former MIT professor Rodney Brooks, who has been working in AI since the 1970s and used to run the university’s AI Laboratory, isn’t so sure this is all going anywhere. As so little is understood about how large language models actually work, he has compared the recent achievements of AI to the late Renaissance scientific theory of why things burned. The theory, later discredited by the discovery of oxygen, posited the existence of a flammable solid element called “phlogiston.”

The idea that AI models will be perfected with more computer power “is the phlogiston theory of intelligence,” Brooks said at an MIT symposium this summer. “So be careful, it could take longer to do all these things than we think.”

Another well-known challenge that could stymie generative AI is its tendency to make things up. ChatGPT falsely told an instructor at Texas A&M University that it had authored essays by his students. Lawyers have inserted made-up citations supplied by ChatGPT into court briefs. And the chatbot’s efforts to write computer programming code contain errors more than half the time, researchers at Purdue University discovered.

Such mistakes may not matter much for things like ad-copy writing or creating artistic works, AI expert and NYU professor emeritus Gary Marcus said. But those applications won’t be nearly enough to support a trillion-dollar market, according to Marcus, who coauthored the book “Rebooting AI: Building Artificial Intelligence We Can Trust.”

“I can easily get my mind to a $10-billion-a-year industry across the seven or eight companies that are doing this — I might even get to a $20 billion industry without too much trouble,” Marcus said. “But at some point, things have to work, and they have to work better.”

Boosters claim they will be able to eliminate the mistakes by training AI models with even more data or adding fact-checking software. But some AI experts doubt the models can be fixed. A study at Stanford found ChatGPT had actually gotten worse at some tasks over the past year.

Another problem with the industry is the high costs involved. Generative AI apps require lots of computing power, racking up huge electric bills and harming the environment. Microsoft was losing $20 per month on average for every customer subscribing to its generative AI computer coding assistant, the Wall Street Journal reported.

Looking more closely at the trillion-dollar economic projections also reveals a nuanced picture. Almost half of the revenue gain in Bloomberg’s forecast, for example, is simply from additional sales of Nvidia hardware and Amazon, Google, and Microsoft cloud services.

Other growth areas in the report appear more speculative, such as the $280 billion in increased spending on software, including for smarter AI assistants, researching new drugs, and improving cybersecurity. That would be equal to almost one-third of all software spending in 2023, according to research firm Gartner.

“These are numbers we came up with at this point in time based on our understanding of the market,” said analyst Mandeep Singh, one of the report’s authors and global head of the technology sector at Bloomberg Intelligence. “But as we go along next year, there could be revisions.”

Even if the technology is over-hyped and most generative AI startups fail, the economy could benefit eventually, as happened after the internet bubble popped (see Google, Amazon, Meta).

A hundred years ago, investor enthusiasm about the nascent automobile industry crashed and hundreds of companies went out of business, historian Vinsel noted. “But obviously, the car over the long term has had huge repercussions for how we’ve used it and built society around it,” he said.

“The reality is somewhere in the middle,” Graham Brooks, a tech venture capitalist at Boston-based .406 Ventures, said. “Whether you’re a bull or a bear on the economic impact of generative AI, we know . . . this will create a fertile environment for new company formation and growth.”

But AI expert Marcus remains pessimistic, comparing generative AI to simplistic fads such as the Pokemon Go craze of 2016.

“A lot of these markets, people look at some early thing that’s exciting and they just add zeroes,” he said. “It’s a little hard to see how you’re going to get to those with the technologies we know about now.”

Aaron Pressman can be reached at aaron.pressman@globe.com. Follow him @ampressman.