AI and GDP: The Caplan-Karnofsky Bet
Caplan takes a risk
Holden wins $5000 if, by the end of 2044, there has been at least one year for which it is plausible — according to Bryan Caplan — that at least one of the following was true:
Real Gross World Product was (conceptually, as opposed to officially) at least 4x the real GWP of 2024.
Or: real GWP was (conceptually, as opposed to officially) at most 1/2 the real GWP of 2024.
Otherwise, Bryan wins $1000.
My remarks:
This is an unusually risky bet for Bryan to make. Usually, he goes for what seem like sure things. But there is no sure thing when it comes to AI.
Especially on a 20 year time horizon! Yes, I know that there are adoption lags, but with 21st-century communication, I think we get wide adoption much earlier than 20 years.
If Holden wins on the low side of the bet (GDP falls by half), it won’t be because of AI. War or natural disaster are the only plausible scenarios as I see it.
If Holden wins on the high side of the bet, then at least one of three things must happen:
Technology finally achieves great things in education. We have been promised dramatic improvements in education ever since the appearance of television. So far, spending on education continues on a rising curve, and results continue to disappoint.
Technology achieves great things in health care. This could come from improved health care delivery, as machines powered by AI deliver better results more cheaply. But the big potential is in research, with dramatic ways to prevent and cure disease, extend healthy lifespan, and enhance the human body.
Robots become highly skilled and inexpensive. They tend to crops, obtain minerals, take care of energy production, handle transportation, and provide all sorts of goods and services.
I think that measurement issues could matter for this bet. I believe that over decades, estimates of GDP vastly understate the improvement in living standards. One problem is that the better we get at something, the less it matters in GDP. For example, we are tremendously better at food production than we were 150 years ago. As a result, agriculture has shriveled as a share of employment and GDP. Instead of using GDP as the conceptual indicator for this bet, I would have recommended some specific targets related to education, health care, and robotics.
I would take Holden’s side of the bet. There may be a 50 percent chance that we don’t observe the high side of the bet (and I give ~0 percent chance for the low side), but Bryan is giving 5-1 odds.
“This is an unusually risky bet for Bryan to make.”
I understand what you’re saying, in particular in terms of harm to his reputation should he lose.
But *financially* it’s not particularly risky, if one agrees with you that he ain’t gonna lose the bet on the downside.
Because in that utopian-adjacent scenario Bryan has almost certainly become so rich that the $5,000 will be a pittance.
I have a few thoughts.
1: I don't think that really large educational gains will really be seen in 20 years. That would require a huge swing from the current status quo to extreme adoption of (very good) AI in a short period of time. AI is only kind of ok for education now, the education system is very hidebound and slow to adapt, and even if those two were to change within 5 years that would only leave 15 years for students to finish school under the new system (5-10 years) and work their entry level jobs for a bit. It might be lucky to see 5-10 years of much better educated people get infused into the economy, a small percentage of the total work force and one with very little experience and rank. Without having an entirely new avenue for changing the economy like ecommerce in the late 90's and early 2000's, I don't see how much better educated young people shake things up so quickly.
2: There might be a possibility that AI replaces a lot of middle management compliance roles for very large cost savings. Very large businesses probably won't care a huge amount, nor very small, but there is currently a large gap whereby there are hardly any businesses in the 50-200 people range (roughly, might vary by state). The issue is usually attributed to regulatory burdens, as most rules have cut outs for small businesses, and larger businesses have more issues with legal, HR, accounting, etc. and so it is a big step change in how a firm runs. It seems that most firms either stay small, get acquired, or try to grow very large quickly or die trying. If instead of having to suddenly get a dozen mid range experts on various compliance and legal fields one could hire a few people to manage those tasks predominantly using AI that might allow a lot of growth in the mid size field of the economy. That might be quite a boost, through more efficient business size and increased competition.
3: Business planning, both long term strategic and short to mid term planning operational planning, could really improve a lot. Most ERP/MRP systems are clunky and awkward, saving on actual planners while requiring technical teams to keep running, and are based on technical systems many decades old. An AI that could do decent finite scheduling, what if analysis and other decision making tools without massive amounts of specialized customization and maintenance could be really huge. Oracle and SAP are both working on such stuff, but I don't think they will be setting the world on fire there; I expect there will be smaller companies that use AI pulling from those data sources to plan and then feed things back, and at vastly lower costs than the big ERP companies. That could be huge for the small to mid size firms.
I think I would still take Holden's side of the bet if I were forced to, but if the odds were 2:1 it would be really hard to pick the better side. 20 years is a pretty long time, but given institutional inertia and the "pretty good, but not as trustworthy as a person yet" nature of AI, I don't see the inertia getting overcome by the power of new tools for quite a while.