Often researchers focus on the capabilities of AI models, for instance what kinds of problems they might solve. Or how they might boost productivity growth rates. But a different question is to ask how they might lower cost of living for ordinary Americans. And while I am optimistic about the future prospects and powers of AI models, on that particular question I think progress will be slow, mostly though through no fault of the AIs.
If you consider a typical household budget, some of the major categories might be:
A. Rent and home purchase
B. Food
C. Health care
D. Education
Let us consider each in turn. Do note that in the longer run AI will do a lot to accelerate and advance science. But in the next five years, most of those advances may not be so visible or available. And so I will focus on some budgetary items in the short run:
A. When it comes to rent, a lot of the constraints are on the supply side. So even very powerful AI will not alleviate those problems. In fact strong AI could make it more profitable to live near other talented people, which could raise a lot of rents. Real wages for the talented would go up too, still I would not expect rents to fall per se.
Strong AI might make it easier to live say in Maine, which would involve a de facto lowering of rents, even if no single rental price falls. Again, maybe.
B. When it comes to food, in some long run AI will genetically engineer better and stronger crops, which in time will be cheaper. We will develop better methods of irrigation, better systems for trading land, better systems for predicting the weather and protecting against storms, and so on. Still, I observe that agricultural improvements (whether AI-rooted or not) can spread very slowly. A lot of rural Mexico still does not use tractors, for instance.
So I can see AI lowering the price of food in twenty years, but in the meantime a lot of real world, institutional, legal, and supply side constraints and bottlenecks will bind. In the short run, greater energy demands could well make food more expensive.
C. When it comes to health care, I expect all sorts of fabulous new discoveries. I am not sure how rapidly they will arrive, but at some point most Americans will die of old age, if they survive accidents, and of course driverless vehicles will limit some of those too. Imagine most people living to the age of 97, or something like that.
In terms of human welfare, that is a wonderful outcome. Still, there will be a lot more treatments, maybe some of them customized for you, as is the case with some of the new cancer treatments. Living to 97, your overall health care expenses probably will go up. It will be worth it, by far, but I cannot say this will alleviate cost of living concerns. It might even make them worse. Your total expenditures on health care are likely to rise.
D. When it comes to education, the highly motivated and curious already learn a lot more from AI and are more productive. (Much of those gains, though, translate into more leisure time at work, at least until institutions adjust more systematically.). I am not sure when AI will truly help to motivate the less motivated learners. But I expect not right away, and maybe not for a long time. That said, a good deal of education is much cheaper right now, and also more effective. But the kinds of learning associated with the lower school grades are not cheaper at all, and for the higher levels you still will have to pay for credentialing for the foreseeable future.
In sum, I think it will take a good while before AI significantly lowers the cost of living, at least for most people. We have a lot of other constraints in the system. So perhaps AI will not be that popular. So the AIs could be just tremendous in terms of their intrinsic quality (as I expect and indeed already is true), and yet living costs would not fall all that much, and could even go up.
The post A household expenditure approach to measuring AI progress appeared first on Marginal REVOLUTION.
