Should AI Automate Jobs — or Augment Them?

Should artificial intelligence be used mainly to automate jobs, replacing people with technology? Or should it instead be used to augment jobs, helping human workers make better decisions and be more productive?

Already, AI systems are putting some humans out of work. In a survey of nearly 1,025 U.S. workers, conducted in January by Resume Now, an AI resume service provider, close to half the respondents (43%) said they know someone who has lost a job to AI. And nearly nine in 10 (89%) said they’re concerned about AI’s potential impact on their own job security.

AI’s impact on work was the agenda item at an in-person event held recently in Washington, D.C. Titled AI at Work: Navigating the Next Evolution of the Labor Market, the event was co-sponsored by the Bipartisan Policy Center(BPC), a not-for profit organization that encourages government policymakers to work across the aisle, and Anthropic, a provider of advanced AI tools.

Consider Humans First

To set the stage, Cheryl Oldham, BPC’s Executive VP of Human Capital, told attendees: “We believe America’s greatest asset is its human capital — the skills, knowledge and experience of our people.”

Rep. Valerie Foushee, a House Democrat from North Carolina and a former member of the Bipartisan House Task Force on AI, told attendees that the advent of AI, machine learning and other advanced technologies has brought us to a historic moment.

“The future of AI is exciting and holds immense potential for progress,” Foushee said. “But we must take proactive steps to ensure the United States remains at the forefront of the development.”

The country, Foushee added, should take decisive action now, laying out clear technology rules to innovate, educate and protect. “Then we can navigate this evolving landscape,” she said, “and ensure that the benefits of AI are realized in a secure, equitable and ethical manner.”

‘Change is the Constant’

With those positive comments made, moderator Heather Long, an opinions columnist and former economics correspondent for the Washington Post, addressed the elephant in the room: Are we all going to lose our jobs to AI?

The panelists’ answers, in short, were no…but don’t expect things to stay the same, either. “AI changes how jobs get done, and most jobs are going to change radically,” said panelist Jack Clark, a co-founder of Anthropic and its Head of Policy.

Panelist Jack Malde, Associate Director of BPC’s Human Capital Program, agreed. “Even if we do see a lot of automation of specific jobs, that doesn’t mean the jobs are going to disappear,” he said. “They could simply transform.”

John Horton, an Associate Professor at MIT’s Sloan School and a research group lead at the MIT Initiative on the Digital Economy (IDE), also held that position. “Work doesn’t disappear with technology,” he said. “Change is the constant.”

But Horton also added a note of caution: “People think everyone else’s jobs are more automatable than their own. They say, ‘Oh, you can automate that, but I have all these special things I have to do.”

That led Malde to offer a warning of his own. As AI transforms work, he explained, people will need to gain new skills, even if they stay in the same jobs. But schools may not be up to the task. “The bad news is that our current educational workforce system is not agile, it’s not nimble,” Malde said. “It doesn’t facilitate the constant retraining and re-skilling that we’ll need.”

Wanted: Better Measurement

One problem encountered when trying to answer Long’s opening query about AI and job loss is that we lack good data. While we do have reasonable figures on how many people are using AI (short answer: a lot), we don’t know much about the specific tasks they’re using the technology for. Nor do we have solid statistics on the levels of productivity and other improvements AI may be delivering.

“We need details on exactly how people are using AI,” Horton said. “That matters because it will let us zero in on which skills are being displaced or augmented.”

To that end, Anthropic has created the Anthropic Economic Index. It’s a set of measurements aimed at understanding in detail AI’s effects on the job market and broader economy over time, in large part by recording how the technology is being incorporated into real-world tasks.

To support this work, Anthropic recent set up an Economic Advisory Council. Founding members include Horton along with fellow professors from the University of Chicago, George Mason University and University of Virginia.

Mapping AI’s Work Path

Given the shortage of in-depth data, the panelists couldn’t provide firm answers about what kinds of jobs are most vulnerable to AI, which jobs could benefit from AI augmentation, or which jobs, if any, are likely to be out-of-bounds for any AI inputs.

But they were able to offer some highly educated guesses, thanks to their own experiences and efforts to accumulate more data, such as the Anthropic Economic Index. The Index shows that the tasks and occupations with by far the largest adoption of AI in its initial dataset are those in the “computer and mathematical” category.

That feeds into Clark’s conclusion that AI is mainly affecting jobs that are highly digitized, such as software coding. That’s likely to increase. “We find that as you get more and more expert with how you use the systems, the amount of automation you do goes way up,” Clark said. That could lead coders and other early adopters of AI to automate their own jobs, transforming the nature of their work.

As of now, while AI use is widespread, occupations for which the technology can automate all tasks appear to be rare. “This indicates that few occupations will disappear, but many will be augmented,” Horton said. “At least for now.”

Which occupations are most likely to be AI-resistant? Panelists pointed to work that involves complex human communications, such as nursing, eldercare and teaching. While these jobs could be augmented by AI, they most likely won’t be automated out of existence.

But how about people working in fields like software development that arehighly subject to AI automation? What can they do?

“Be incredibly curious, and use the technology in an unbounded way,” Clark advised. That could mean viewing AI as both a universal tutor and a tool that can massively speed up your day job.

And, perhaps, take solace in Clark’s assessment: “There’s never been a better time to re-invent yourself.”

Learn more: Watch the full hour-long video of the AI at Work event.

Peter Krass is a contributing writer and editor with the MIT IDE.