The idea this week: AI may be progressing fast — but that doesn’t mean it will rapidly transform the economy in the near term.

Large language models like ChatGPT are becoming more and more powerful and capable. We’ve already started to see a few examples where human workers are plausibly being replaced by AI.

As these models get even more skilled, will they substantially replace human workers? What will happen to the labour market?

To try to answer these questions, I spoke to labour economist Michael Webb, for the latest episode of The 80,000 Hours Podcast. He’s worked for Google DeepMind, in the UK government, and at Stanford University.

Michael argues that new technologies typically take many decades to fully replace specific jobs.

For example, if you look at two of the biggest general-purpose technologies of the last 150 years — robots and software — it took 30 years from the invention of each technology to get to 50% adoption.

It took 90 years for the last manual telephone operator to lose their job from automation.

a switchboard

So if we look to the past as a guide, it may suggest that if AI systems do replace human workers, it will take many decades. But why does it take so long for very obviously useful technologies to be widely adopted? Here are three reasons:

  • Adopting new innovative technologies can take lots of money and time. Companies have to decide it’s worth the risk to redesign their systems, rethink a set of processes designed around humans, and implement those changes — which can take years.
  • There are often many regulations that delay major social impacts. Labour unions often push against new technology that threatens workers, and regulating bodies need to assess new technologies before they’re widely used. These hurdles may not stop the adoption of new technologies forever — but they can slam on the brakes.
  • Humans are often just averse to both risk and change. Many will choose to wait for others to try out the technology first. The net effect of this is a big delay between when a technology is developed and when it’s widely adopted.

This process isn’t always perfectly smooth, but roughly that’s most of what’s happened historically when we’ve developed new technologies with the potential to replace human labour.

That said, AI might be different.

A lot of the new AI systems already show the capacity to be general purpose and very user friendly. They can already write, code, and make art right out of the box. And once trained, language models like ChatGPT are cheap to use.

What’s more, unlike transitioning from steam power to electricity, many AI systems can likely be integrated into our work lives without massive overhauls to physical infrastructure.

What does all this mean? It’s part of a debate about how fast AI systems will develop and how they will change the world. If AI systems are rapidly adopted to replace human workers, that will affect a lot of people’s lives, and we should be preparing ourselves.

Rapid integration of AI into the workforce would also arguably increase the greatest risks from AI because it would likely result in even greater investment and rapid technological development of these systems — which may prove very difficult to control.

If you want to understand these issues better, I encourage you to listen to the full episode of the podcast with Mike. You can also listen to another interview I recorded with Tom Davidson, who thinks AI will likely change the world more rapidly.

This blog post was first released to our newsletter subscribers.

Join over 400,000 newsletter subscribers who get content like this in their inboxes weekly — and we’ll also mail you a free book!

To learn more about how other people are thinking about the effects of AI on jobs, we recommend: