Tyler Cowen: Why we will learn to love AI

I got to know Tyler Cowen — perhaps America’s most original economist — nearly fifteen years ago when he published his viral monograph, The Great Stagnation. In a 15,000 word essay (published ironically as an e-book), he laid out why we had hit a technological plateau in which all the low-hanging fruits of easy productivity growth had been exhausted. The age of rapid innovation was over. This was classic Cowen — a bold thesis, cogently argued, which stimulated a public debate. Now Cowen is saying precisely the opposite. “I began to change my mind around 2020 with the new mRNA vaccines and computational biology,” Cowen tells me. “Then AI came along. We are now in the early stages of a revolution that will change everything about our world.”

Swampians will forgive me but I must drop two clichés: “A foolish consistency is the hobgoblin of little minds” (Ralph Waldo Emerson); and “When the facts change, I change my mind. What do you do?” (John Maynard Keynes). Both apply to Cowen in the age of AI. As recently as 2017, Cowen doubled down on his great stagnation thesis with The Complacent Class, in which he set out America’s creeping culture of conformity, risk aversion and safety-first. This was characterised by the gentrified “Banana” syndrome — “build absolutely nothing anywhere near anything”.

He was in good company. Robert Gordon’s The Rise and Fall of American Growth, which argued that the age of disruptive technology was over, came out around the same time. For those who argued that our age’s innovations — its noise-cancelling headphones, driverless cars and WiFi saturation — matched what happened in the 100 years of explosive change that ended around 1970, Gordon asked which you would first give up; Your iPhone or the flush toilet? Laptop or antibiotics? If you had trouble answering those, ponder life without electricity.

I do not know whether Gordon has had a chance to reappraise his technological pessimism. But Cowen’s U-turn is striking. He was an early adopter, paying the $20 a month to subscribe to each of OpenAI’s GPT-4 and Anthropic’s Claude 3, as well as using open-source language models such as Perplexity AI. His students at George Mason University are sometimes told to use AI instead of receiving long reading lists. “Everyone will be using it and everything has to be fact-checked anyway,” he says. “But we will have to move quite quickly to oral exams to test what they really know.” Fields such as academia are likely to be stickier for others — not because they are unaffected by the awesome power of generative AI but because of regulatory barriers.

Others, such as my own profession, are more vulnerable. The “wordcels” — people who are good with words — will be hit the most. GPT-5, which could come out this summer, is likely to be ten times more powerful than GPT-4. By the end of next year, according to Elon Musk, they will have outstripped human intelligence. Even if Musk is exaggerating, which is one of his fortes (his forecasts of landing a rocket on Mars and Tesla’s self-driving rollout are behind schedule), he is in large and growing company. The real division is between those who see the AI future through a glass darkly, and those, such as Cowen, who embrace its optimistic potential. He dismisses speculation about “artificial general intelligence” reaching so-called singularity — where computers reach self-consciousness and escape our control — as unknowable. In the meantime, we are entering a world of massive productivity growth that “is no longer science fiction — we are already in it,” he says.

The pace of change is almost impossible to keep up with for anyone. In the week before we spoke, Cowen listed at least eight new rollouts, which includes an upgraded GPT, Meta’s next AI iteration, Google DeepMind’s Gemini 1.5, which can read and digest enormous documents in seconds, and several others. The chances that public regulators will be able to keep up are virtually zero. “We don’t have much control over what will happen, and are never likely to,” says Cowen. So we should embrace it.

One example that most arrested me was a recent medical paper which showed that AI had a better bedside manner than most doctors. Most doctors are overworked. But the idea that AI’s emotional intelligence is advancing in leaps and bounds is quite unnerving. For someone like me, who is a classic wordcel, it is particularly unnerving. “It’s worse for younger versions of you who do not yet have a name to protect them,” says Cowen. “If you’re over 50, you’re probably OK.” Beneficiaries will include “intelligent” blue-collar workers, such as gardeners, plumbers, carpenters and others who can use AI to manage projects. Children, meanwhile, will grow up with “AI Teddys” that read to them at night, chat to them about their problems and teach them Chinese. Parents, like doctors, suffer from frayed nerves. AI never gets tired.

Perhaps there is some poetic justice to all these potential changes. Or perhaps Cowen is simply wrong. He could either be overstating AI’s reach, or seeing it in far too benign a light. John, this is also your world; you are far better qualified than me to respond to Cowen’s prognosis. Is he overstating things? How much are you now using AI in your work, and life?

Recommended reading

• My column this week looks at the under-appreciated thaw in US-China relations — Joe Biden’s polite version of Trumpism on China. As US Treasury secretary Janet Yellen points out, it is possible to disagree with China without being disagreeable. It is precisely because the underlying differences are so great — and possibly insoluble — that such constant gardening is vital.

• Do also read the great John Plender’s essay on the overlooked threats to the global financial system. As the annual World Bank-IMF joint spring meetings come to a close in Washington, everyone should be aware of the risks Plender sets out.

• Finally, I can’t resist citing my FT Alphaville colleagues’ takedown of Liz Truss, Britain’s shortest-serving prime minister, whose new book argues that only ten years remains to save the west. That amounts to just 74.4 Truss prime ministerships. This is how my colleagues characterised her tenure: “Imagine a bunch of people are stuck in a warm, stuffy room together. Everyone wants the windows to be opened, but they are fastened shut . . . While people try to work out how the locks operate, one of them, Liz, attempts to throw a chair through the window. The chair bounces off and hits her in the face.”

Join Nancy Pelosi, Jake Sullivan, Anne Applebaum, Bradley Whitford, Gucci Westman and many more on May 4 in Washington, DC for our third annual FT Weekend Festival: US edition. As a newsletter subscriber, claim 15% off your pass today using promo code NewslettersxFestival.

John Thornhill responds

Like you, Ed, I’m a big fan of Tyler’s writing. Given his previous stance, it’s fascinating to see how he’s suddenly got AI religion. But my response would be: not so fast. I love experimenting with generative AI models but, like many sceptical companies, I have yet to discover a killer app that I can reliably depend upon every day.

Technologists like to invent “laws”. The most famous is, of course, Moore’s Law, which helps explain the exponential increase in computing power that underlies the latest advances in AI. Then, there’s Amara’s Law: “We tend to overestimate the effect of a technology in the short run and underestimate it in the long run.” That seems particularly apt when thinking about AI today.

But my favourite is the “Budding Effect”, as described in the book What to Do When Machines Do Everything. The three authors — Malcolm Frank, Paul Roehrig and Ben Pring — noted that when the English engineer Edwin Budding first invented the lawnmower in 1827, he imagined his only customers would be Regent’s Park Zoo and a few Oxbridge colleges. In fact, the ability to cut grass neatly gave rise a few decades later to Britain’s most successful cultural export: field sports, including football, tennis, rugby and cricket. 

I have no doubt that AI will similarly create massive and, as yet, unimagined industries in future. It’s just hard to grasp what they are today.