A Brief Letter to EAs and Economists
Does AI Eat the World?
Today marked a novelty in interspecies relations: we had Effective Altruists and economists talk to each other. Being friends with many of both, I’d like to further integrate EA and econ styles of thinking.
If I were to choose one fundamental insight from economics, it would be this:
Equilibria respond to imbalance
If I were to similarly oversimplify computer science, it would be:
Simple, self-improving software dominates its industry
or, as Marc Andreesen puts it
Software eats the world
The fundamental insight of economics is one of inevitable sublimation. AI, or whatever innovation, is bound to equilibrium under a system hopelessly more complex than itself. The fundamental insight of computer science is one of monopoly (or at least near monopoly). It is a story in which the hungry protagonist fills all available space, devouring all competitors not based in tech.
In many cases, these insights resonate. Many digital technologies reached equilibrium by completely displacing their analog counterparts. Even ones that haven't, such as Uber, have left a permanent mark on their industry. But what if the industry in question is human labor itself? That is the question asked by AI safety.
The goal of this letter isn't to present the full argument of the economists or the EAs. Hopefully, both groups feel fairly represented, albeit somewhat shortchanged on time. The goal is for both parties to begin applying the fundamental logic of the other side to their own arguments.Â
Good luck, friends.Â
I could’ve read a much longer article on this.
Is the claim about software eating the world really as profound as it seems?
I mean, compare to the claim that written language eats the world. Since the invention of written language we've turned almost every task into one that involves written language in some way. And while that was a profound shift I'm not sure there is any insight to apply there.
Or, to put the point differently, is there anything more here than the fact that computation is cheap and useful so why wouldn't we involve it in many aspects of endeavor. Not sure it implies anything beyond that.