Jens Meyer/AP
- Since the 1980s, automation has replaced middle-income workers with robots and software.
- Automation has also made educated labor more valuable and contributed to income inequality.
- Artificial intelligence could cheapen educated labor and reduce inequality.
Automation widened income inequality in the US. The latest developments in AI might just reverse it.
In research released in March, OpenAI, the company that created ChatGPT, found that the newest artificial intelligence could affect white-collar workers more than the rest of the workforce. The technology could dampen the explosive wage growth they've experienced at the expense of their blue-collar counterparts by shifting more tasks to middle-skill workers, which could reduce income inequality. But that rosy future depends on how AI develops and whether we choose to regulate it.
Up to this point, automation has mostly hurt blue-collar workers. Daron Acemoglu, an economist at MIT, found that automation had driven 50% to 70% of the increase in US wage inequality since the 1980s. Machines, robots, and computer software replaced workers without a bachelor's degree in routine tasks, like assembly-line work or administrative office work. Less in demand, these workers saw their real wages fall or stagnate, especially for those previously making middle-class incomes.
This automation has placed a premium on skilled white-collar labor and increased wage inequality. While automation devalued and replaced lower-skill work, the earnings premium between workers with a bachelor's degree — a proxy for high skill — and those with a high-school diploma doubled, David Autor, an economist at MIT, found.
"Not only doesn't it not replace you, it makes your expertise more valuable," Autor told Insider.
Replacing lower-skill human labor with machines raises the demand and wages for abstract tasks that can't be easily automated. As a result, wage inequality has increased considerably over the past half century.
Today, those same skilled white-collar workers who were the beneficiaries of inequality are now in the crosshairs of artificial intelligence.
"The general presumption is that AI — and even before that, office software — would impact middle-skill workers," Acemoglu told Insider.
"We're really in a different era," Autor said.
While previous software relied on scripted instructions to do routine tasks, like a recipe that results in the same meal, AI can be trained to come up with answers for data it's never seen, which allows it to learn more demanding tasks like how to ride a bike, he said.
It's "powerful and opaque" because now "computers know more than us," he added.
OpenAI's March paper found that large language models — the type of artificial intelligence behind ChatGPT, which can process and generate text that sounds remarkably like natural human language — would affect at least 10% of tasks for 80% of the workforce and 50% of tasks for 19% of the workforce, "with higher-income jobs potentially facing greater exposure to LLM capabilities and LLM-powered software." The report also found those holding bachelor's, master's, and professional degrees were more likely to be impacted.
That word "exposure" means AI could either replace high-income workers, which would reduce inequality, or make them even more productive, which would increase it.
If the technology is as capable as technologists hope it will be, AI "could reduce inequality" and "be an equalizing force," as it would replace high-skill workers and reduce the demand for paying them so well, Acemoglu said.
Even if the technology isn't advanced enough to replace these high-skill workers, it could increase the productivity and quality of middle-skill work, thereby raising middle incomes.
"You can imagine devolving some of the most highly skilled tasks to less skilled people," Autor said. For instance, AI could allow nurse practitioners to "do what a doctor did traditionally because they're augmented by this tool," he added.
AI decreasing inequality isn't "a crazy scenario, but there's a lot of uncertainty," Acemoglu said.
"We still don't know the full capabilities of large language models," he added.
"It may turn out that this is an inequality-expanding technology, rather than adversely affecting the more skilled and highly paid workers," he said. He gave two reasons.
First, "we tend to underestimate how complex human tasks are," he said, so it's possible AI will complement the work of journalists, managers, and radiologists, rather than replace them. That could then increase their productivity and wages.
Additionally, "the more skilled workers may be able to shield themselves," he said. If their job is automated, they could take the next layer of high-wage jobs.
"They won't be as badly affected as when blue-collar line workers were when robots were introduced," he added.
"It is a new game, and we don't know what that new game is going to be," Acemoglu said.
Autor said he thought AI "could contribute plausibly to a reduction in inequality" for those pushed out of their middle-income jobs but that would require changes to how workers are trained, employers hire, and the government guides innovation "to make sure firms don't just automate things away."
As of now, "the incentives are not well aligned with social objectives," he said.
"This is all being pursued in an arms race by a bunch of for-profit companies," he added. "There's no reason to assume what's good for Amazon and Google is what's good for humanity."
Nonetheless, he said, "the future is a choice" affected by the way we deploy our tools.