Now you see it, now you don’t: Why data can’t capture the AI revolution

Suppose a technology arrives that makes doctors more effective, accelerates scientific discovery, upends labour markets, shifts trillions in wealth, and destabilises tax systems — all while barely registering in official economic statistics.

That is precisely what Michael Spence told me is coming, in a recent interview.

Spence is no starry-eyed tech evangelist. He is a Nobel laureate in economics, former dean of Stanford Business School, and one of the world’s most rigorous thinkers on growth and global markets. He also happens to have a ringside seat: advising major investment firms as well as Google, tracking developments in Silicon Valley and China, and speaking regularly with technologists and scientists across multiple fields.

When Spence says artificial intelligence’s impact will be “multidimensional” and that focusing on any single dimension — jobs, productivity, military applications — will cause you to miss the big picture, it is worth paying attention.

The paradox of invisible abundance

One of Spence’s most striking insights is that AI’s massive economic benefits may not show up in traditional national income statistics for years. This is not a new problem — it is a feature of the digital economy writ large.

“Twenty years ago, we could buy a computer for $1,500 that did next to nothing by today’s standards,” he notes. “What we can buy now for $1,500 is like what a supercomputer was back then.” The hardware revolution transformed the world – but the economic data, which treats all $1,500 computers the same, barely captured it.

AI will do the same. Standard measurements fail to account for “free goods” – search engines, social media, navigation apps – that generate enormous value at negligible incremental cost. AI will produce huge gains in quality across healthcare, education and professional services that official gross domestic product figures will not reflect.

“We’ll probably get better healthcare and doctors may spend more time on the golf course,” Spence says. “We’ll know something important has happened – but we won’t have data that proves it,” he told me on the sidelines of the recent annual conference of the Asian Bureau of Finance and Economic Research.

The jobs question: Don’t stop too soon

Public anxiety about AI and jobs is understandable – but much of it rests on incomplete analysis. The common approach, says Spence, is to identify replaceable tasks, add them up and project job losses. “That’s an exercise worth doing,” he says, “so long as you don’t stop there.”

The missing step is understanding price elasticities: to what extent will gains in efficiency lower prices to expand consumption, what people buy more of as a result, and how those ripple effects play out across the broader economy. A surgeon assisted by AI may see more patients; cheaper legal services may generate more demand for legal work. The net employment effect depends critically on those second-order dynamics – which most job-loss studies ignore.

How disruptive AI turns out to be will also depend on how fast it spreads. In the world of business, Spence thinks it will move more slowly than enthusiasts expect.

“It’s one thing to give ChatGPT to people and let them have fun with it – that didn’t take long. But implementing it in corporate environments, inventing new business models, changing entrenched behaviour, building systems companies are confident in – all of that takes time.”

He invokes the Silicon Valley rule of thumb: When a new technology arrives, people tend to overestimate its impact in the short run and underestimate it in the long run.

“That’s a pretty good guide to what we’re going to experience. It was the same with the internet. There were crazy ideas about how fast it would go, there was a bubble – but out of that moment came Amazon, Google, and the reinvention of Apple. It wasn’t all hype.”

So, is this a bubble?

“A little bit,” says Spence.

It’s driven by incentives. Between the US tech giants and the US-China rivalry, the competitive incentive is to win at all costs.

“If the choice is between overinvesting on one hand and underinvesting and risking coming third on the other, the preference is to overinvest.”

Sceptics see this as a waste of resources that will fail to deliver the promised productivity surge. Optimists believe that even if returns disappoint relative to initial hopes, the investments will still yield positive long-run returns. History suggests the optimists are more often right.

The bubble could deflate for two reasons, says Spence: a major security incident triggering severe regulatory intervention, or disappointingly weak revenue growth among AI developers, prompting a sharp reset in valuations.

On regulation, Spence is cautious about heavy-handed intervention. Some economists advocate nationalising critical AI infrastructure, as nuclear facilities are nationalised.

“Probably a bad idea,” he says, “especially during an innovation cycle, which governments are generally not good at navigating.” But parts of AI infrastructure could become regulated utilities, he concedes.

Even within the industry, he notes, some self-regulation is already happening for the right reasons. Anthropic declined to release its most powerful new model, Claude Mythos, to the public – restricting it instead to a small group of trusted institutions – because its capabilities were considered too dangerous for general availability.

“So, even among the players, the idea of regulation is not crazy,” says Spence. The same logic should apply to powerful AI systems developed in China.

“It’s fine to have a bunch of open-source models, but if one of them is dangerous, they’ll have to close it – or all hell could break loose.”

International cooperation on AI governance will be unavoidable, Spence adds. The key question is whether major players share enough common interests to agree on basic rules – preventing AI-enabled autonomous warfare, for instance, or blocking state and non-state actors from using AI to attack critical digital infrastructure.

The wealth problem that’s hard to solve

Of all the challenges AI poses, Spence is most pointed about the fiscal and distributional consequences.

AI will make the global economy much more capital-intensive. If income continues shifting from labour to capital, the already stark concentration of wealth will worsen further.

In the US, the top 10 per cent hold roughly two-thirds of total net worth; the bottom half hold around 2.5 per cent. With strong returns on financial assets, that gap will widen.

Universal basic income is frequently proposed as the answer. “That may not be a bad idea,” says Spence, “but it’s certainly not the solution. It just puts a floor under incomes — so you get most people sitting on the floor while a select few capture the astronomical wealth generated by capital.”

The real challenge, he argues, is giving people a meaningful stake in capital itself. Governments could claim a share of capital returns and redistribute them.

More ambitiously, individuals could be given direct claims – for example, a capital asset account established at birth, invested in something like an index fund tracking the broad market.

“Whatever route is chosen,” Spence says, “we’re not going to end up where we need to be if people don’t have some equity stake in the economy to a much greater extent than they have now.”

The AI revolution, in Spence’s telling, is neither the techno-utopia its champions promise nor the job-destroying catastrophe its critics fear.

It is something more complex and demanding: a transformation that will generate extraordinary wealth while making it harder to measure, reshape labour markets in ways that simple task-counting cannot predict, and stress-test political and fiscal institutions that were not designed for a world where the benefits flow overwhelmingly to capital owners.

The technology, in other words, is not the hard part. The hard part is everything that comes after.

Vikram Khanna is a former associate editor of The Straits Times who writes on economic affairs.

Source: The Straits Times © SPH Media Limited. Permission required for reproduction.

PMA retailers face teething issues while adapting to new mobility scooter regulations from June 1