Human Extinction Greatly Exaggerated?

Apparently so.

I need to do some clarifying.

In the former post I ended up adding a caveat— "If humans can live off the interest they make from money they have in the bank, this wouldn't apply to them."

Depending on whether AIs have agency, this could be very likely.


If AIs are not their own independent agents, then they will not make money— the capital will stay with the humans. Money is spent to maintain the AIs, but AIs never actually get paid.

In this case, all the money spent in the economy ends up returning to humans as interest. When a human buys a good, the money ends up going to the AI managing the business. From there, it goes to the AI investor and back to the bank to pay back the money loaned him. Humans receive this money as interest. They may earn crappy wages, if they work at all, but they can live well off interest.


If AIs are independent agents, then things are more precarious. The future of humanity, in this case, depends on how much humans have in the bank, and whether people (or AIs) have an incentive to create more AIs.


Here's an essay where Robin Hanson explains his vision (which I'm basing this on): Economics of the Singularity


I've also discovered that the talk that inspired the gnxp post was by Marshall Brain at The Singularity Summit. A video should be up soon, and I'll post it later with criticisms.

Thanks to Ryan for keeping me on my toes.

The Economics of Human Extinction

While I'm going through some sociological material for my supply of ideas post, I thought I'd write about something completely different. Hat tip to gnxp, for bringing this to my attention.


A few necessary explanations:

Complements and Substitutes

When two different markets interact, this interaction can take two opposite forms. In one, a lowering of the price of one good increases the demand for the other good. Gasoline and Hummers, for example. As the price of gasoline goes down, people are more likely to buy Hummers. These two goods are complements.

In the other, a lowering of the price of one good decreases the demand for the other good. Here, Ford and Chevy trucks are an example. When Chevy lowers the price of its trucks, fewer Ford trucks are bought (because people buy Chevy, instead). These goods are substitutes.

Malthus

Malthus is famous for arguing that the human population would always increase to the point where humans earned enough for bare subsistence. If they made more than subsistence wages, the population would increase until the wages went back down. If they made less than subsistence wages, people would die until wages worked their way back up to the subsistence level.

But Malthus didn't know about condoms.

Here's a not-drawn-to-scale representation of where we are now (the point I picked is a total guess):




The curve slopes upward at first because a greater number of people allows for a more extensive division of labor. In other words, human labor acts as a complement to human labor. As more humans enter the market, however, the substitution effect starts to dominate. The division of labor has been taken as far as it can profitably go, and more humans only bid down wages.


Now the fun part:

Robots - "They Took Er Jerbs!"

Right now, technology acts mostly as a complement to human labor. A human in conjunction with a piece of technology (say, a forklift) is more productive than a human without the technology. Since wages are based on productivity (long story), human wages increase.

Fastforward to the future, where cheap artificial intelligences (AIs) can do any job a human can. At this point, technology acts primarily as a substitute for human labor, and the competition from AIs will drive human wages down until they equal the cost of an AI doing the same job. (Otherwise, the employer will just hire an AI instead of a human.)

Now, this would be OK if the supply of AI was unchanging. But this is a marketplace, and people can make a profit selling AIs. Competition among AI sellers will drive the price of AI down to the price of creating it. So, basically, employers will buy AI labor whenever the benefits of the labor outweigh the cost of creating and maintaining the AI. In other words, if AIs fetch more than an AI subsistence wage on the market, the population of AIs will increase until the wages go down. If they make less than AI subsistence wages, some will be decommissioned (die?) until the wages go back up. Malthus has his revenge.



Since human wages must be at least as low as their AI competitors, humans will get paid AI subsistence wages. If the AI subsistence wage goes below the human subsistence wage... sorry, humans, that's the end.

Caveats

There are a few crucial assumptions this rests on.

1) Diminishing returns for labor. If AI wages never go down, no matter how many are made, this doesn't apply. In the earlier graph, the curve would never go down below the subsistence wage. This is very unlikely.

2) AIs can replace a human in any job. If humans are wanted as, say, pets, this doesn't apply.

3) If humans can live off the interest they make from money they have in the bank, this wouldn't apply to them.

I almost wrote that enhanced humans would avoid the problem, but this is incorrect. Enhanced humans also have to compete with AIs. (Even scarier!)

The future is a strange place.


More

Gary Becker, An Economic Analysis of Fertility -- an in-depth discussion of condoms and other factors that threw Malthus off

Robin Hanson (my new hero), Economic Growth Given Machine Intelligence


Edit- partial retraction here.