Economic Responses to AI

Technologies like ChatGPT are potentially quite disruptive to a lot of economic activity. This is really true for a lot of other AI technology today. Whenever you have a new capability where you can through technical means produce something that was hard to do before that was surprising. Obviously, ChatGPT has been surprising to many people, not just lay people but experts as well as to how good it is. Whenever you have such a new capability, it can adjust the balance of many equations that are out there that determine how people do what they do. In particular, for language generation technologies like ChatGPT, generating coherent on point prose is historically has always, for most of human history been very costly and hard thing to do. If it suddenly becomes much cheaper and easier to do by automated means through AI, it might first enable all kinds of new applications that were not cost-effective to try before. That's really exciting. It can also enable a lot of AI decision-making that was possible before but limited in its applicability because of the need to communicate results by language. I think that has been something that has been a barrier to many AI applications up to now. If all of a sudden AI is able to communicate fairly credibly, fairly fluidly with people, it's going to enable many applications that we didn't see before. Now, how will that affect current economic actors and the economy? Well, whenever you have new opportunities, new changes, there's really two effects you need to think about. One is substitution. If there's some other way that something is being done that way maybe pushed out by the new, cheaper, more effective way perhaps. That could include some possible substitutions of things that people do now that will be done more with AI. The other possibility is compliments. When a technology can amplify the effectiveness or reduce the costs of a person doing something, they can do it better, more efficiently by the addition of AI that may lead to more opportunities for those people to be more productive and to produce better products. I think typically both of those factors are in play. Technology like language generation by AI have both substitute effects and complement effects and the relative sway of those may vary depending on the context and it may change over time. One example, I think to illustrate how AI can be complimentary and substitutable is through our experience with the game of chess. Now chess is not an economically significant or important activity but I think it's instructive to think about that example. Twenty years ago, actually, maybe 25 years ago by now, an AI program defeated the world chess champion. So no person was as good as a computer could be a chess. Now, for a while, the best chess players were actually teams of humans working with chess programs. The combination can beat any person could be any chess program. For about 20 years that remain to be the case but the chess programs got better and better and now with a human and a chess program, the human is not really bringing much to the table. We saw this pattern of supremacy by the AI. The complement that the teamwork between a human and an AI being complimentary and their skills. But then eventually as the technology changed, it was no longer complimentary. We may see such patterns like that and a whole range of other AI technologies over time.