Assumptions:

• People don’t get more intelligent over time (generation over generation)

If we are to believe the AI marketing, AI is about to be super smart. Like take all the jobs smart. We are so close! We’ve been 6 months away from complete workforce displacement for only about 4 years!

So obviously AI isn’t about to take everybody’s job. But it may take some. Here’s why it could be extremely bad.

Two Futures

If AI continues to get smarter

If sites like this one are to be believed, then AI is already well into the 90th percentile of humans. As we interact with these models however, we can feel that that’s not true. Therefor IQ isn’t the best metric to judge intelligence, but the idea of IQ is what matters here. It works as a stand-in for some sort of quantitative value of intelligence.

Every job has a lower bound on how smart the worker has to be. With a normal innovation, say the cotton gin, that only displaces one job type. The cotton gin didn’t change how tobacco was processed. The workers displaced by the cotton gin can flow into other jobs of similar mental caliber (perhaps tobacco farming). There is a possibility with AI that all jobs of low mental compute requirements could all get displaced (the cotton, tobacco, etc, laborer). Purportedly that is the goal of AI. Then what do we do with all these displaced persons?

Hopefully many of these workers have mental headroom, meaning that to obtain a higher intelligence job and they don’t need to change how smart they are. This would be ideal, considering you can’t change how smart you are. You can learn things, but an increase in head-knowledge doesn’t increase IQ. This is an important distinction because AI can also “learn” things, but it, like you, can’t get smarter.

Since humans have a range of approx: 80-120 IQ, and AI’s IQ will continue to rise, fewer and fewer humans will be necessary / useful. Every generation of both humans and AI (each model being considered a “generation”) somewhat “roll the dice” as to how smart they will be. It is somewhat inheritable, but not without fluctuations. The difference however, is that AI’s “generations” are about 3-6 months, and since they are just fancy math machines, there is no problem disposing of non-useful members. With humans, obviously our generations are about 25 years and we don’t get rid of non-useful members.

As the intelligence of AI rises, this will bring the lower bound of useful humans’ IQ, up with it. We will have a growing “useless eater” population. This could lead to things like more useless wars to kill them off, or things like UBI to keep them alive.

If AI has plateaued

Then we are safe, and all this investment into AI companies is mostly meaningless and the basket with all our eggs has an overstated value. Open-source can take it from here, building agents and tooling. We’ve already seen things like OpenClaw being “hugely disruptive” in the AI world. But this wasn’t because models got better, it’s because a cool tool was overlaid on top of the existing technology.

If you are concerned with being replaced I dug through some of FRED’s data to product the projected most valuable skills in 2024-2034. It could be interesting for you.