In 2013, researchers at the University of Oxford released some startling numbers about the future of work. They estimate that 47 percent of all jobs in the United States are “at risk” of automation “for an unspecified number of years, perhaps 10 or 20 years.”
But ten years later, the country’s unemployment rate is at a record low. At the time, the tsunami of dire headlines like “Half the world’s jobs are about to disappear because of the rich and robots” seemed utterly irrelevant.
But the study’s authors say they didn’t mean to imply that the end was really near. Instead, they were trying to explain what technology could do.
Think tanks, corporate research groups, and economists will publish paper after paper pinpointing exactly how much work is “affected” or “exposed” to technology. It was the first attempt at what became a long-running thought experiment.
In other words, if the cost of tools is not a factor and the only goal is to automate as much human labor as possible, how much work can technology replace?
When Oxford University researchers Carl Benedict Frey and Michael A. Osborne were doing their research, IBM Watson, a question-answering system using artificial intelligence, won Jeopardy! and shocked the world. I just gave it. A test version of a self-driving car has circumnavigated the road for the first time.Today, a new wave of research is surging with the rise of tools that use generative AI
Goldman Sachs estimated in March that the technology behind popular AI tools like DALL-E and ChatGPT could automate the equivalent of 300 million full-time jobs. Researchers at Open AI and the University of Pennsylvania, who developed these tools, found that 80% of the U.S. workforce could impact at least 10% of jobs.
“There’s a lot of uncertainty,” says David Autor, an economics professor at the Massachusetts Institute of Technology who has studied technological change and labor markets for more than two decades. “And people want to provide those answers.”
But what exactly does it mean that, say, 300 million full-time jobs could be affected by AI?
It depends, Autor said. “To be affected means to get better, get worse, disappear, multiply.”
One complicating factor is that technology tends to automate tasks rather than entire operations. For example, in 2016, artificial intelligence pioneer Jeffrey Hinton considered a new “deep learning” technology that could read medical images.he concluded “If you work as a radiologist, you’re like a coyote who’s already over the edge of a cliff but hasn’t looked down yet.”
He thought it would take five, maybe ten years for algorithms to “work better” than humans. What he probably overlooked is that reading images is only one of many tasks he does (30 of them, according to the US government) that’s what radiologists do. We also offer interviews with medical professionals and counseling services. Today, some people on the ground worry about the imminent danger. Shortage of radiologists. And Hinton has since become a vocal and public critic of the same technology he helped develop.
Frey and Osborne came up with the 47% figure by asking technology experts to rate the likelihood that entire occupations such as “telemarketers” and “accountants” would be automated. . However, three years after the publication of the paper, a group of researchers at the ZEW Center for European Economic Research based in Mannheim, Germany, published a similar study evaluating tasks such as ‘describe a product or service’ and found that 21 National occupations could be automated.
“People love numbers,” says Melanie Arntz, lead author of the ZEW paper. “People always think that numbers must be certain in some way because they are numbers. But numbers can actually be very misleading.”
In some scenarios, AI is essentially a tool building, not a complete job replacement. You are now a miner who can use an excavator instead of a shovel. Or nurses who have access to better information to diagnose their patients. You may need to pay a higher hourly rate because you can get more work done.
In other scenarios, technology complements the workforce rather than replaces it. Alternatively, you can change from a job that requires special skills to a job that does not require special skills. It’s unlikely to work for you.
In either case, Autor says that throughout history, technological developments have tended to affect the distribution of wages and wealth primarily, not the number of available jobs. “With this kind of exercise, you risk focusing on a single tree that stands out too much and missing the forest,” he said of research investigating how much human work could be replaced by AI.
What he sees as another key focus: how artificial intelligence will change the value of skills is harder to predict. Because the answer depends in part on how new tools are designed, regulated and used.
Think customer service. Many companies leave the task of answering the phone to an automated decision tree and deploy human operators only for troubleshooting. But one of his Fortune 500 enterprise software companies took a different approach to the problem. It created a generative AI tool that suggested what to say to agents, keeping them informed of their ability to read human and social cues. Researchers at Stanford University and the Massachusetts Institute of Technology compared the performance of a group that was given a tool and a group that was not given a tool. found This tool has significantly improved performance for less skilled agents.
Even if jobs were fully automated, what would happen to the lives of unemployed workers would depend on how companies harness technology for new kinds of work, especially those we can’t yet imagine. said Daron Acemoglu, a professor at the Massachusetts Institute of Technology and author of Power. And progress: our millennial struggle for technology and prosperity. ” These choices include whether to fully automate the work or use technology to augment human expertise.
He said the seemingly frightening numbers predicting how many jobs will be cut by AI are “alarm bells,” even if it’s not clear how.
He said he believed people could “steer in a better direction” but was not optimistic. He doesn’t think we’re on the “pro-human” path.
All estimates of how much work AI can take over rely heavily on humans. In other words, researchers make assumptions about what technology can do. Frey and Osborne invited experts to a workshop to score the likelihood of occupations being automated. More recent research has relied on information such as the database that tracks AI capabilities created by the Electronic Frontier Foundation, a non-profit digital rights organization. Alternatively, it relies on workers using platforms like Cloudflower where people complete small tasks to earn money. Workers score tasks based on factors that are amenable to automation. For example, if your tolerance for error is high, a technology like ChatGPT is a good candidate to automate.
Many researchers involved in this kind of analysis say the exact numbers don’t matter.
“I would say that our methodology is arguably exactly wrong, but we are headed in the right direction,” said Michael Chui, an AI expert at McKinsey who co-authored the 2017 white paper. , suggesting that about half of jobs and 5% of occupations can be wrong. Automated.
What the data describes is, in some ways, more mundane than commonly assumed. That said, big changes are coming, and they’re worth keeping an eye on.
