A PhD physicist friend on Facebook:
The reason I’m all-in on AI investment, and there may not be a top to this “bubble”: Unlike tulips, railroads, or internet connectivity, there is no upper bound on the value of intelligence.
Me:
30-year TIPS yield is 3% real currently. So your perspective isn’t shared by the market. Investors wouldn’t lend money to the federal government at 3% (after inflation) if they thought there was no limit to NVIDIA’s value. If investors overall thought that there was a low-risk way of making super high real returns over the next 30 years they the government would have to pay 8% real, for example, if buying an AI index was expected to yield 12% real for the next 30 years.
I think AGI is already here from the perspective of most users of ChatGPT and similar. A generally intelligent human isn’t great at everything and makes mistakes… just like ChatGPT! People ask ChatGPT all kinds of questions and give the answers at least the same weight that they would give to an answer from a typical human. In that sense, ChatGPT has passed a Turing Test for general intelligence. Maybe Advanced Superintelligence is already here. The typical human is not good at mathematics. A math professor could have been described as “super intelligent” before WWII. LLMs are supposedly doing all kinds of advanced work in mathematics right now, even if they might fail at a plumbing task. The math professor in 1935 who was a failure at plumbing would still have been considered superintelligent, right?
In other words, we can’t expect hockey stick growth for the economy due to AGI/ASI arriving because the current growth is already an example of what an economy does with the gift of AGI and ASI (but maybe not robotics!).
Will the $40 trillion in federal debt be a drag on economic growth? Scott Bessent says “no”:
Let’s use a 20-year time horizon for the U.S. to potentially get out of Argentina territory (150% debt-to-GDP max; we’re at over 125% right now). For US debt to GDP to fall to the level of a high IQ society (e.g., Taiwan, which has lower-than-US tax rates and debt of about 20% of GDP), GDP growth would need to be 10%/year real for 20 years with Congress not borrowing any more money (the latter condition seems unlikely to be met, since Congress now borrows even in the most robust economies, contrary to Keynes). Investors plaintly don’t believe that this will happen because they’re willing to lend to the Feds at 2.75% real (20-year TIPS current price) and they wouldn’t do that if investing money in domestic stocks would generate a roughly 12% real annual return (real GDP growth plus 2% as a return on investment from corporate earnings).
Does this mean that we’re in an AI bubble? Not necessarily. Only that AI by itself apparently doesn’t hugely lift the overall U.S. economy (a huge part of which is government spending/welfare state!). We’ve got about 1.5% per capita real GDP growth right now. Maybe that includes the AI lift? This NBER paper by a Nobelist (sort of) predicts minimal per capita growth, but cites estimates as high as 3.4 percent per year as the boost (nowhere close to the 10% we’d need to get out debt down to Taiwan’s relative level):
(He cites McKinsey, the giant brains behind Enron!)
The only way to make $40 trillion in debt insignificant, therefore, would be to grow the U.S. population to about 1 billion humans at roughly the same level of skill/income as the current U.S. population. Until Donald Trump showed up (again!), our wise politicians were working on this, but they forgot to apply and skills test for immigrants.
(In case this blog post is going to be a source for an NPR or PBS story (example), let’s not forget that both Turing and Bessent were/are members of the 2SLGBTQQIA+ community.)