by Trevor ChowBasil Halperin, and J. Zachary Mazlish

[Note: This is an appendix to "AGI and the EMH: markets are not expecting aligned or unaligned AI in the next 30 years"]

One naive objection would be the claim that real interest rates sound like an odd, arbitrary asset price to consider. Certainly, real interest rates are not frequently featured in newspaper headlines – if any interest rates are quoted, it is typically nominal interest rates – and stock prices receive by far the most popular attention.

The importance of real rates. However, even if real interest rates are not often discussed, real interest rates affect every asset price. This is because asset prices always reflect some discounted value of future cash flows: for example, the price of Alphabet stock reflects the present discounted value of future Alphabet dividend payments. These future dividend payments are discounted using a discount rate which is determined by the prevailing real interest rate. Thus the claim that real interest rates affect every asset price. 

As a result, if real interest rates are ‘wrong’, every asset price is wrong. If real interest rates are wrong, a lot of money is on the table.

Stocks are hard to interpret. It may nonetheless be tempting to look at stock prices to attempt to interpret how the market is thinking about AI timelines (e.g. Ajeya CotraMatthew Barnett/r/ssc). It may be tempting to consider the high market capitalization of Alphabet as reflecting market expectations for large profits generated by DeepMind’s advancing capabilities, or TSMC’s market cap as reflecting market expectations for the chipmaker to profit from AI progress.

However, extracting AI-related expectations from stock prices is a very challenging exercise – to the point that we believe it is simply futile – for four reasons.

  1. First, and most importantly, these companies will only have the possibility of high profits if transformative AI is aligned; under unaligned AI, the value of stocks along with everything else is converted to zero. 
  2. Second, it is not obvious that even in the aligned case that these companies will earn high profits. For instance, OpenAI has committed to a capped profit model, and others may sign on to a similar ‘Windfall Clause’. Beyond corporate altruism, it seems extremely plausible that if a private company develops truly transformative AI technology then the state will (attempt to) nationalize and expropriate it to distribute the benefits more broadly, preventing profits. 
  3. Third, stock valuations are extremely idiosyncratic: which stock should we be looking at? And critically, even if we take a basket of tech companies and average over them, then this only includes public companies. If the market expects transformative AI in 12 months, but only because it will be developed by OpenAI – a company which is not traded publicly – then this will not show up in any equity index.
  4. Fourth, and quite importantly, it is not obvious whether expectations of transformative AI would raise or lower stock prices. This is because, as described in the previous subsection, stock prices reflect the present-discounted value of future profits; and advanced AI may raise those future profits, but – as the central thesis of this piece argues – advanced AI would also raise the interest rate used to discount those profits. The net effect on stock prices is not immediately obvious. 
    1. (In math, briefly: if the price P is the value of future profits D discounted at rate r, i.e. P=D/r, then transformative AI may raise future profits D but it could raise the discount rate r by even more.)
    2. (Higher growth causes lower average stock prices if the intertemporal elasticity of substitution is greater than one, rather than less than one. This parameter is subject to significant debate; see the linked slides for useful discussion. John Cochrane offers additional intuition here and argues that the empirically-relevant case is the one where higher growth causes lower equity prices: expectations for transformative AI would lower equity prices.)

If you want to use market prices to predict AI timelines, using equities is not a great way to do it.

In contrast, real interest rates do not suffer from these problems. 

New to LessWrong?

New Comment
2 comments, sorted by Click to highlight new comments since: Today at 12:43 AM

Epistemic status: I don’t know much about the subject. The following is stated much stronger than I believe it.

Counterargument: The market consists mostly (I assume? Please correct me if I’m wrong) of quants and traders trading for other people’s money. Either rich people or pension funds. If the quants can make good trades by buying stock that will increase and sell stocks that won’t, they will be rewarded. Hence, such trades are incentivised and we should expect the EMH to hold between stock prices. However, no one is really incentivised to tell the owners “look, I think the world is going to end soon, you should just spend your money instead”. If the world was obviously going to end soon, some financial advisors would say so. But no quant will be paid to try to find out if the probability of the world ending is 1% or 10% to find out if the client should spend their money now. So we should not expect EMH to hold for interest rates.

I think we can theoretically get around 4 by comparing the value of AI stocks to non-AI stocks.

I think an additional problem is that we don't have a no-AGI-baseline to compare prices to-- we can see that Nvidia is worth $400B but we can't directly tell whether that includes lots of expected value from an AI boom or not.