🎇📈 A Quick Q&A on predicting super AI through bond markets with … Trevor Chow
A Throwback Thursday conversation
For centuries, interest rates have served as the economy's tea leaves, offering glimpses of storms and sunshine ahead, such as inflation expectations, central bank policy, and investor sentiment.
Now, an intriguing study suggests they may forecast something far more profound: the emergence of transformative artificial intelligence.
Trevor Chow and his collaborators, in their paper "Transformative AI, existential risk, and asset pricing," have crafted a framework linking the arrival of advanced AI to movements in financial markets. Their conclusion is striking: Whether artificial intelligence proves to be humanity's greatest triumph (my bet) or its ultimate undoing, real interest rates are poised to rise.
The logic is elegant. As the prospect of transformative AI draws closer, rational actors become increasingly reluctant to defer consumption. After all, why save for a future that might be radically different — or might not exist at all? This behavioral shift puts upward pressure on real rates.
From the paper’s abstract:
We study how financial market prices can be used to forecast the likelihood of transformative artificial intelligence. Transformative AI is a double-edged sword: while advanced AI could lead to rapid economic growth, some researchers argue that superintelligence misaligned with human values could pose an existential risk to humanity. Theoretically, we show that either possibility would predict a large increase in long-term real interest rates, due to consumption smoothing. We then use rich cross-country data on real rates and growth expectations to show that, contrary to other recent findings, higher long-term growth expectations do indeed cause higher long-term real interest rates.
Drawing on data from inflation-linked bonds and international surveys, the researchers demonstrate that heightened growth expectations consistently correlate with elevated long-term real interest rates. This relationship suggests that bond markets might serve as an early warning system for the approach of artificial general intelligence.
Chow is a former researcher at the Stanford Artificial Intelligence Laboratory and is the co-founder of Moonglow, a machine learning startup helping researchers connect their Jupyter notebooks to powerful remote computing resources.
(The following exchange was first posted on July 29, 2024.)
1/ What is your basic methodology here?
At a high level, the approach we take in the paper is to apply the standard microeconomic logic of consumption smoothing towards the question of forecasting the arrival of transformative artificial intelligence. The idea of consumption smoothing is that with diminishing returns to consuming more stuff, you’ll want to smooth out how much you consume so that the amount is even over time. If something makes the returns to consumption smaller in the future, this logic implies that you should pull forward your consumption, such as by borrowing from your future income in order to spend it now. If there is a general increase in the demand for borrowing, we’d expect to see this reflected in the real interest rate, which is simply the price of credit.
2/ How did you determine that the anticipation of transformative AI would lead to an increase in real interest rates, regardless of whether AI turns out to be aligned or unaligned?
To determine how transformative AI affects real interest rates, we break it down into two mutually exclusive cases: one where transformative AI is designed to behave aligned with humanity’s goals, as well as one where it isn’t and it leads to a catastrophic outcome. In the former case, we would expect a surge in economic growth, and this future abundance would encourage people to borrow more now, increasing interest rates. In the latter case where the unaligned AI is expected to cause an existential catastrophe, the fact that there is no need to save for the future would again cause an increase in interest rates.
3/ Could you elaborate on the mechanisms through which financial markets are expected to forecast the development of transformative AI using long-term real interest rates?
The case for financial markets, and in particular bond markets, to reflect the likelihood of transformative AI depends on the assumption that they are efficient information aggregators. That isn’t to say that markets are always and everywhere accurate, but rather that there is rarely free money on the table. If it were clearly the case that transformative AI would arrive soon and interest rates were still low, there would be an enormous profit incentive to keep borrowing at those low rates, and either pay them back when there is explosive economic growth from AI, or not need to pay it back at all in the unaligned case.
4/ Why do most economists appear less convinced than AI researchers about the imminent development of transformative AI, its potential to significantly accelerate economic growth, and its existential risk to humanity?
There are a couple of reasons why this might be the case. One is that AI researchers who are closer to the research frontier may simply be more optimistic about the likelihood of continued advancements in machine learning. Another is that economists have seen how these predictions can often be quite optimistic; for example, many were predicting a similar economic boom from the internet, and this has by and large failed to materialise in GDP growth. Finally, economists may be more cognisant of all the barriers between better technology and economic growth, such as the slow pace of technological diffusion and the many regulatory barriers which stifle the creative destruction caused by transformative technologies.
5/ Is it harder to forecast how development of advanced AI might affect the prices of stocks, land, and other assets?
There are two sources of difficulty for predicting how advanced AI will affect other asset prices. The first is that an unaligned AI would have the opposite effect on equities or commodities than aligned AI, since they would become worthless if an existential catastrophe were to occur. The second is that even in the aligned case, it is not clear what the effect on equities would be: While transformative AI might raise the profits of companies, it would also raise the interest rate. Since equities are valued based on the value of future profits discounted by the interest rate, it isn’t clear which effect would dominate: higher profits or higher rates.
On sale everywhere ⏩ The Conservative Futurist: How To Create the Sci-Fi World We Were Promised
Micro Reads
▶ Economics
▶ Business
Zuckerberg Lobbies US Senators on Artificial Intelligence - Bberg Opinion
It’s Not What Jamie Dimon Said About WFH. It’s How He Said It - Bberg Opinion
▶ Policy/Politics
Nikola’s EV Truck Dream Goes Bust - WSJ Opinion
A Defense of Weird Research - Asterisk
Frequently Asked Questions About US Government Funding for R&D - What’s New Under the Sun
▶ AI/Digital
Was DeepSeek Less Moby, More Wannabe? - Bberg Opinion
▶ Biotech/Health
▶ Clean Energy/Climate
How to Streamline Nuclear Power Plant Construction - Bipartisan Policy Center
▶ Space/Transportation
What Would Actual Scientific Study of UAPs Look Like? - Universe Today
▶ Substacks/Newsletters
How Should AI Liability Work? (Part I) - Hyperdimentional
A Quantum Breakthrough is Coming in AI - AI Supremacy
Europe’s Internal Trade Barriers: A Long Way From a Single Market - Conversable Economist
Baidu's Apollo Go is the Waymo of China - next BIG future