AI and the future of work: where I stand

Will AI leave us with a lasting shortage of paid work? This page sets out my reasoning and the reading behind it, so you can check my arguments against the sources and disagree with them properly.

First, what I am and am not saying. I am not saying every job disappears, and I am not saying the next five years will look like mass unemployment. My claim is narrower: new jobs and new demand may not be enough to replace the demand for human work that AI removes. The strongest sources I have read call this possible, not inevitable. I think it is likely enough to plan for.

My three arguments

1. The exits are closing

The usual answer to automation is that people move into the work machines cannot do. That answer assumed the remaining work was a stable place to go. Daniel Susskind calls the pattern “task encroachment”: machines keep entering areas once reserved for people, first manual work, then cognitive work, and increasingly emotional and social work too. If AI can also do the new tasks we would move to, the old escape route narrows. Retraining only helps if there is enough work to retrain into. Susskind separates frictional unemployment (there is work, but people struggle to reach it) from structural unemployment (there is not enough demand for human work). Education and training answer the first problem. They do not answer the second.

I do not expect this to happen everywhere at once. Regulation will slow it in fields such as healthcare and law, and some tasks will lag for technical reasons. But slower is not stopped: encroachment does not need to be simultaneous to be structural. A delay changes the timeline, not the direction.

2. Working with AI is a phase, not a destination

“AI won’t take your job, someone using AI will” is comforting, and it is true for now. But look at chess. After Deep Blue beat Kasparov, human-plus-computer “centaur” teams beat computers alone for a while. Then systems like AlphaZero trained themselves and the human contribution became hard to identify. Susskind uses this example for exactly this reason: complementarity holds only while people add something machines cannot supply. Working alongside AI can be a stage on the way, not a permanent place to stand.

3. A richer economy does not automatically employ more people

The optimistic case rests on productivity: AI makes us richer, and richer economies create work. But output and employment have come apart before. Agriculture and industry both became far more productive while employing a shrinking share of the workforce, and what absorbed the workers was new kinds of work. Whether new demand does that again depends on whether people can still do that work better than machines. A paper by Ping Wang and Tsz-Nga Wong models one scenario where large productivity gains coexist with persistent employment losses. It is one model outcome under stated assumptions, and the same paper also finds scenarios with employment gains. I use it to challenge “higher productivity automatically means enough jobs”, not as a forecast.

Reading and listening

Descriptions say what each source argues and where it is limited.

Start here

Books

Papers

Podcasts and talks

If you have recommendations, please email me at hello@inessilva.me.