Economic Pathways

It seems I’m not the only one trying to predict this, although I have more immediate near term interests on what is the exact path that AI will take for the economy.

GATE

The first interesting study, if you can call it that, is from Epoch AI, which has put out a GATE model of the total Gross World Product that we could see. They disclose their assumptions and parameters, and you can play around with the knobs to see the curves go wheee.

The gist of the thesis is that increasing AI increases automation, creating virtual workers, and if you calculate output based on virtual workers, then add another term for total factor productivity increase due to automation, and then calculate the increase in capital over time, you can predict global growth. Its a fun exercise, with 100% economic growth within our lifetimes.

METR

The second and perhaps more solid study is an exploration of a tweet from Richard Ngo of OpenAI from a couple of years ago.

The good folks at METR have expanded on this idea, measuring the difficulty and time that a human takes to perform a software task, say a web search, to a complete refactor of a codebase, and how long a software agent would be able to do this task given a minimum 50% success rate.

From this, they estimate that AI task length is doubling every 7 months, with typically 1-4 doublings per year. They extrapolate this to AIs being able to take on months long tasks by 2030.

Reaction

The X audience was generally receptive to both studies, but I think people can’t really grasp how high economic growth looks like.

I was in Asia around the period of the Beijing Olympics and it felt like the world was taking off. Pig farmers in China became billionaires. Skyscrapers were built in a year.

That was just 10% growth. (Although to be honest we foreigners only saw Beijing and Shanghai, which were growing at 20%, while the rest of the country was at 5%).

You almost could not lose money, because every deal would have at least an option on land, and if you waited 2-3 years you outskirt industrial land would be purchased for residential towers as the ring roads expanded around the megalopolises.

I am currently unable to imagine the physical environment in the West changing that much. Growth of 30% per year sounds like if must come from off planet, either space or the metaverse. I could imagine for example that the zoomer generation, brought up on Roblox, values virtual goods more than physical, which creates a Metaverse economy unconstrained by physical laws.

We will see.