Why hasn't AI increased unemployment?

In most cases, it's actually (much) worse. For example, the infamous Ford rehiring.

The guy with his back to the volcano says "what problem?"

simple, the trillions of dollars bieng spent at some ridiculous rate of whatever number of milliins per hour is bubbling the whole worlds economy that and while indivual productivity isn't great, the ever increasing scale of industrial production of everything is reducing real unit costs and providing more profit froth to the foam based economy and lastly is the simple fact that AI does not actualy do anything except burn money and the vast majority of what AI gets credited for still needs human support and monitoring and the real situation is one of REDUCED efficiency , the rest is just a shell game needing people to do the shuffling, speaking of which!, hey! time to write a quote!

This is from Peter McCrory, Head of Economics at Anthropic. Calling that out since it wasn't obvious to me at first.

There are more people in need to operate it or added it to reduce workload. It is more likely automates tasks more often than entire jobs and increases productivity before reducing jobs. However, there will be the day when the ulitmate utopian goal will be to remove the word 'work' out of the dictorary for sure.

The task list only gets bigger

The idea that AI isn't increasing unemployment is an illusion created by statistical modeling. Usually, statisticians are good at convincing others of their modeling choices. For example, even if programmers get laid off and janitorial jobs increase, the overall employment rate might not show any damage.

When stakeholders cite statistics, they can always shape the modeling to fit their narrative. I'm not sure if this is a general trait of statisticians, but they tend to interpret overall indicators in ways that favor their own school or affiliation.

One possible interpretation of why the overall job market looks stable, despite AI eroding employment, is that macroeconomic forces are strong enough to keep unemployment from rising. But that interpretation might have been excluded precisely because the speaker is affiliated with Anthropic.

Predictive statistics are also subjective. Surveys can be manipulated depending on how you select the sample group. This is really a mix of hope and bias, framed as objective prediction. The public tends to think of technological change as a single 'event,' while technologists experience it gradually in real time. But the public only notices it when it suddenly appears. Nokia employees probably felt safe too, right up until Steve Jobs walked out with the iPhone.

Curious about this, I looked into the US BLS survey. I found that working just one hour counts as being employed. So I wonder if they've looked into whether quality jobs are disappearing while bad jobs are increasing.

And the real impact of AI will likely show up not as layoffs, but as a freeze on new hires. That is, existing employees, because of company-specific tacit knowledge, might be kept, but the hiring rate for new workers, especially younger ones, will drop in roles highly exposed to AI. So instead of looking at total unemployment, we should look at youth employment as a leading indicator.

Of course, this could just be the limit of my knowledge. I'm self-taught with no formal degree. I could be wrong. But honestly, it doesn't take much digging to see that what this person is saying doesn't hold up.

One of the skills I've developed most while reading HN is learning how to present statistical truths in a polished way, even when the content isn't actually true.