Viewpoint: we need to invest in ‘pro-worker’ artificial intelligence research, not automation

Funding agencies, governments and universities should launch new grants and competitions to steer the development of artificial intelligence so that it benefits ordinary workers, rather than seeking to automate their jobs, says a leading economist.

Simon Johnson, former chief economist at the International Monetary Fund, thinks research bodies could come up with contests to design AI systems that help plumbers or nurses do their jobs more productively, for example, rather than letting tech giants steer AI in a way that risks mass unemployment.

“AI is still in a relatively formative stage,” he told Science|Business. “And AI could create really good opportunities for less educated people.”

This call to rethink AI research is part of a much bigger argument that has relevance for scientists, innovators and policymakers everywhere. Johnson, and his fellow Massachusetts Institute of Technology economist Daron Acemoglu, are on a mission to challenge the deeply entrenched ‘techno-optimist’ worldview that technological progress is on the whole a good thing, even if there are bumps in the road and growing pains.

Instead, they worry that without more public efforts to redirect technological progress in a beneficial way, current waves of innovation, particularly AI, could actually impoverish ordinary people and shatter democracy. 

Last year the pair released Power and Progress, a 1,000-year sweeping history of technological change which seeks to overturn the dominant idea that “everyone everywhere should innovate as much as they can, figure out what works, and iron out the rough edges later,” as they put it.

This sunny view, that technological progress ultimately does workers more good than bad, is endemic in policymaking, propagated by billionaires such as Elon Musk who want to be free of constraints on their businesses, but also parroted by journalists, politicians, and complicit academics who have lucrative side gigs with tech companies, they write.

While Johnson and Acemoglu are largely writing about the US, techno-optimism is arguably dominant in Brussels too – witness the recent birth of the European Innovation Council, a €10 billion fund to help scale up EU companies, for example.

Rewriting history

The problem with techno optimism is that history tells another story, Johnson and Acemoglu argue.

Contrary to stereotypes of stagnation, there was plenty of innovation in medieval Europe, including the invention of windmills, crop rotation, and better uses of horses in agriculture.

Yet ordinary peasants largely did not benefit from these productivity gains, because feudal oppression allowed elites to get the benefit themselves. The church, a huge landowner, extracted these new surpluses to build the towering cathedrals that dot the European landscape to this day.

Or take the industrial revolution, often seen as a necessary if painful historical step towards modern plenty. On the contrary, Acemoglu and Johnson argue there was little to no improvement in real wages or diet among British workers during the first century of industrialisation, as machines were used to replace workers in the textile industry, not help them better do their jobs.

What changed was the arrival of new technologies that actually created new work – like the railways, telegraph, and telephone networks – rather than automated it. Unionisation and the widening of the franchise also gave workers more power to demand higher wages.

Prizes for pro-worker AI

The relevance today of this history lesson is that there is absolutely no guarantee the technological progress of the 21st century will benefit ordinary people, particularly in AI.

But, in the US, “when it comes to policymakers, they might have reservations or concerns about some direction [of technology], but they basically go along with what the big companies want,” says Johnson.

In Europe, there is more concern about AI, and policymakers have laid down regulations to hem in its potential dangers, the AI Act being the most prominent.

But even here, there is very little discussion of how to publicly create new AI research areas that might benefit the average worker, said Johnson. Instead, the initiative is left to big tech firms.

Indeed, the prevailing hope in Brussels is that private AI firms from the EU will catch up with their US and Chinese counterparts, creating their own rival generative AI models to the likes of ChatGPT, rather than taking a fundamentally different approach.

Instead, “why not push AI to take forms that enhance the skills and productivity of workers who don't have college education, for example?” suggested Johnson. “We think that’s something that’s eminently doable.”

Establishing new forms of “pro-worker” AI, as Johnson puts it, might seem a daunting task.

But whole industries can spring from small grants or competitions, he says. In 2004, the US Defense Advanced Research Projects Agency (Darpa) offered a $1 million prize to anyone who could build a self-driving car able to cross 300 miles of Mojave Desert.

All the cars crashed out, but this ‘Grand Challenge’ competition, and its even more lucrative successors, are credited with spawning the now enormous autonomous vehicle industry.

Research organisations could offer similar prizes to innovators who can invent AI that helps nurses, electricians or plumbers work more productively, suggested Johnson. “I don't think it's that hard to come up with some things if you talk to experts. You just need it to be a fairly well defined problem,” he said.

Government intervention has already steered the direction of AI progress, Johnson and Acemoglu argue – but down a dark path. Demand for facial recognition technology by local government in China has spurred Chinese AI startups to focus on this kind of surveillance technology, in which China is now a world leader.

Redirecting technology

AI that helps a plumber easily find the source of low water pressure, for example, would be a lot more useful than many uses of the technology deployed today, Johnson says.

In Power and Progress, Johnson and Acemoglu critique what they call “so so automation”: uses of digital technology that yield little productivity benefit, but allow companies to cut costs by firing workers.

One example is automated checkouts in supermarkets, which move the burden of scanning items onto the customer. Another is customer service chatbots, which are cheap, but not nearly as helpful or emotionally responsive as humans.

The march of this kind of cost-cutting, automating AI is not inevitable, they argue: instead it is a choice about where to invest research efforts, led by tech firms and executives who believe replacing workers is the priority.

“What we're proposing […] is to find ways to redirect the attention of the inventors and change their priorities, so they're working really hard to try and find ways that machines augment the capabilities of less educated workers,” he said.

“But thinking through the social consequences is not something that's often at the forefront of the technologist’s mind,” Johnson added.

Less automation, and tools to enhance worker productivity or create new tasks, might even spur more economic growth than the current path, Johnson and Acemoglu suggest.  “The idea that the capitalists always do what's best, even for them, seems rather exaggerated,” said Johnson. “Automation is a good example”.

In other sectors, the idea of government-steered R&D efforts to incentivise new products is completely run of the mill, Johnson and Acemoglu point out.

Germany, and then China, mobilised vast public subsidies to spur the development of solar panels, which are now cost-competitive with fossil fuels.

And in the 1990s, HIV activists managed to force an initially reluctant US government into researching drugs and vaccines – with the result that the disease is now largely treatable.

The same approach – that technology can be redirected for the public good – should now be applied to AI and other digital technologies, they argue.