AI comes down to Earth

Open this photo in gallery:

John Rapley is an author and academic who divides his time among London, Johannesburg and Ottawa. His books include Why Empires Fall (Yale University Press, 2023) and Twilight of the Money Gods (Simon and Schuster, 2017).

Whisper it, but we may already have reached peak AI.

Until recently, the share prices of the Magnificent Seven (Nvidia Corp., Amazon.com Inc., Apple Inc., Alphabet Inc., Meta Platforms Inc., Tesla Inc. and Microsoft Corp.), the companies expected to most benefit from artificial intelligence, were on fire. Led by Nvidia, the chipmaker whose share price has risen 600 per cent since the late 2022 launch of ChatGPT, they reached the point that they together became worth more than every stock market in the world outside the United States.

But while the debut of ChatGPT was originally seen as the start of a revolution that would give us driverless cars, superhuman robots and cures for cancer, it seems recently to have petered out. Following its stratospheric rise, Nvidia’s share price recently settled into a range which resembles that of an old-fashioned company. Reality, it appears, has dawned.

As always happens with new technology, investors initially had to rely largely on the predictions of the industry by which to judge its growth prospects. Now that we’ve all had a bit of time to experiment with AI technology, we’re starting to form our own picture of its future potential. As we discover bugs, from legal briefs citing invented cases, academic papers that include conversations with chatbots and the bizarre images produced by Google’s Gemini tool, we’re starting to gain a more realistic, nuanced sense of what AI might actually do.

Although its potential will likely still be revolutionary, its impact on the economy may not be quite as transformative as initially supposed. For instance, there have been some successful experiments in using AI to increase conversion of inquiries into sales or to provide simultaneous translations of videoconferences, and there appears little doubt AI will have profound impacts on medicine. However, skeptics point to the way driverless cars, despite huge investment, have so far failed to materialize as a cautionary tale. That may be because most applications of AI to date have been low-risk but low-value, like chatbots, whereas the big returns would involve risks which, like driverless cars, may turn out to be greater than we’ll tolerate as a society.

Meanwhile, AI is proving more expensive than initially thought. In its initial development, it benefited from what amounted to free subsidies, enabling the technology’s owners to reap the dividends of growth at contained cost. The large language models through which most of us encounter AI were able to refine themselves by combing the vast data trove on the internet. But some of that material is copyrighted, and now that copyright owners are starting to demand compensation, AI firms are facing new expenses.

Equally, the huge environmental externalities of AI, largely ignored at first, are getting factored into future projections. The vast data centres powering AI consume huge amounts of energy and water, and now account for about 4 per cent of global greenhouse gas emissions. Regulators are beginning to look more closely at how prices should be attached to those externalities. So that part of the free AI ride will also end.

Meanwhile, the sea of money that fuelled the bubbles of the last decade is slowly drying up. Since the 2008 financial crisis, growth in the U.S. money supply exceeded that of the economy while the fiscal largesse of governments of all stripes – Donald Trump cut taxes, Joe Biden boosted spending, but both ran large fiscal deficits – provided a pool of money looking for investment opportunities. That cheap credit fuelled bubbles in assets from crypto to housing and provided new firms with inexpensive capital. But as we slowly exit the era of cheap money, investors will have to rely more on performance than promise when judging where to place their money.

Accordingly, they’re starting to assess the models of AI companies more probingly, factoring in both their expected costs and likely revenues and revising their forecasts in a less bullish manner. It’s not that the revolution in artificial intelligence has ended. But it may end up resembling the advent of home computers or the internet, both of which deeply altered society but failed to fulfill hopes of productivity revolutions.

It’s therefore not surprising that the most radical visions of AI are being reconsidered. The spectre of superhuman killing robots taking over the planet and enslaving humans is one that both AI’s architects and critics have warned against. However the jury is still out on whether artificial general intelligence – the ability to replicate the full range of human mental activity rather than specified tasks – will ever come into existence, while some neuroscientists argue that the brain depends on the body to acquire and communicate knowledge, making humans ultimately irreplaceable by machines.

All in all, we may be living less through a revolution than an evolution, one which will profoundly and possibly even spectacularly transform some subsectors, but may not alter the overall economy as much as supposed. Plus ça change.