AI Won't Steal All Our Jobs: The Surprising Historical Truth About Automation

The AI Job Panic: What's Really Happening?

The latest employment statistics paint a concerning picture of the labor market and the apparent disruption caused by artificial intelligence. Following earlier warnings about unemployment among recent graduates, new data suggests AI's impact is spreading to a wider range of workers. October saw over 150,000 layoffs - the worst October for job cuts in over two decades - with approximately 50,000 attributed directly to AI. Overall, 2025 has witnessed more job reductions than any year since 2020.

However, it's premature to determine how much AI is truly responsible for these losses, even as companies publicly blame the technology. Researchers from the Yale Budget Lab and Brookings argue that the broader labor market isn't experiencing more disruption from AI than it did from previous technological revolutions like the internet or personal computers. Meanwhile, Anthropic CEO Dario Amodei has predicted AI could eliminate half of entry-level white-collar jobs.

An Expert Perspective on AI and Employment

To shed light on this complex issue, we spoke with Neil Thompson, principal research scientist at MIT's Computer Science and Artificial Intelligence Lab (CSAIL), who has extensively studied how automation transforms labor value.

The Dual Phenomenon of AI Job Impact

Thompson suggests we're witnessing two simultaneous phenomena: "One is that AI is becoming more prevalent in the economy. For some cases, like customer service, that's probably pretty legitimate. These systems seem awfully good at those tasks, and so there are going to be some jobs being taken over."

"At the same time," he continues, "it would be surprising if these systems could do as many things as the job loss numbers imply. I suspect there's also a mix of either people deciding to cut jobs and put some blame on AI, or they're cutting jobs in advance with an aim to implement more AI."

The Critical Role of "Last-Mile Costs"

Thompson emphasizes that adoption barriers significantly slow AI implementation. "For most businesses, there are very large last-mile costs involved with actually adopting these systems. Someone using ChatGPT in the interface is very different than 'we now run our business trusting the system will get it right every time.' You often need to bring in specific data, and there are many costs that come with that."

He also distinguishes between a system being "good" versus "good enough to be better than a human" - they're not the same thing.

The Expertise Framework: A New Way to Understand Automation

Thompson recently co-authored a groundbreaking paper with MIT colleague David Autor using expertise as a framework for understanding how automation affects labor value. Contrary to popular doom scenarios, historical evidence reveals a more nuanced picture.

When Automation Actually Increases Value

"When we think of automation, we imagine a doom scenario where jobs disappear and wages decline," Thompson explains. "But looking at the last 40 years of computerization, we found that while routine tasks were automated, wages didn't necessarily go down. Some increased, some decreased - that's the puzzle."

Their research reveals that which tasks get automated matters crucially. Automating high-expertise tasks has one effect, while automating low-expertise tasks produces entirely different outcomes.

Real-World Examples: Uber Drivers vs. Proofreaders

Taxi drivers represent the first scenario: "The most expert thing you did was know all the roads in a city. Then Google Maps came in, and suddenly anyone who can drive can do a pretty good job. Your most expert tasks got automated away, so wages go down - but the number of people in that profession goes up because now many more people can drive for Uber."

Proofreaders illustrate the opposite dynamic: "Spellcheck automated their least expert tasks. The meaningful work they did - reorganizing paragraphs, ensuring proper phrasing - remained. Because they were using their expert skills more frequently, their wages actually increased faster than average, though there are now fewer of them."

Historical Precedents and Future Predictions

Learning from the Industrial Revolution

Thompson points to skilled artisans like wheelwrights and blacksmiths: "These were incredibly expert jobs. Through industrialization, we figured out how to produce wheels on assembly lines where average expertise was lower, but vastly more wheels were produced with many more people involved."

The AI Bubble Question

Regarding concerns about an AI bubble, Thompson distinguishes between utility and valuation: "I don't think the question is whether AI will prove itself. These capabilities are improving fast enough to be incredibly useful. The bubble question is about valuations - are we building out faster than these effects will kick in?"

Why AI Creates Anxiety and What History Teaches Us

Recent surveys show Americans are more concerned than excited about AI technology. Thompson understands this anxiety: "It's understandable that people have anxiety about how AI will change their jobs because it's a very powerful tool. It will change many jobs - yours included, mine included."

The Challenge of Unknown New Tasks

"Historically, when new technologies automated tasks, humans moved to new tasks that didn't exist before. We really don't know what those new tasks will be ahead of time. That lack of visibility is challenging, but historically there's been a remarkable wellspring of new tasks and jobs that emerged."

The Critical Factor: Speed of Transition

Thompson emphasizes that the pace of change determines how difficult adjustment will be: "If transformation happens medium- to long-term, humans are pretty good at adapting. But if displacement happens in a compressed period, that makes economic adjustment much harder."

The Bottom Line: Historical Lessons for AI

Thompson offers cautious optimism: "We can take comfort from historical lessons. The question is whether AI is different from previous technologies in ways that would produce different outcomes. If AGI arrives quickly and can do everything we can do, that's very different. If it rolls out gradually, we're more in a world where we can adjust as we have in the past."