Against using stock prices to forecast AI timelines

Also seems important to note: EA investing should have not only different expectations about the world but also different goals about the point of finance. We are relatively less interested in moving money from worlds where we're rich to worlds where we're poor (the point of risk aversion) and more interested in moving it from worlds where it's less useful to where it's more useful.

Concretely: if it turns out we're wrong about AI, this is a huge negative update on the usefulness of money to buy impact, since influencing this century's development generally becomes much less important. So even if we think the stocks are accurately priced, we probably want to bet a lot of money on the market being surprised in the direction of AI being huge and timelines being short, because if those are true, money would be extremely useful, and it's potentially worth giving up a lot of money in worlds where it's false.

(Edited to add: even more concretely, this means that stocks correlated to AI progress are underpriced for us even if they're not underpriced for a typical investor, and in particular riskier assets correlated to AI progress like options on AI/semiconductor/datacenter stocks are likely especially underpriced for our preferences. Also n.b. by "our preferences" I mean as donors -- I'm talking about the overall longtermist portfolio, not making investment advice for any individuals.)

I weakly agree with practically all the arguments in this post, but I think the conclusion is overstated. I suspect looking at stock prices is still quite valuable in light of these points -- especially semiconductor stocks, as I suggested in the tweet that you linked in this post.

Fourth, and quite importantly, it is not obvious whether expectations of transformative AI would raise or lower stock prices.

It really only makes sense to talk about prices in a relative, rather than absolute sense. Normally people don't care about this distinction because they're talking about nominal prices, and the price of money changes slowly relative to a broad basket of goods and services. But in the context of this post, and especially my linked tweet, I think it's quite important to make the distinction.

When you say it's "not obvious whether expectations of transformative AI would raise or lower stock prices", did you mean in nominal terms (such as after an inflation adjustment)? If so, then I'd agree with the claim as written, but only to the extent that we're talking about the price of equities in general. In my tweet, I was talking about semiconductor stocks rather than the entire equities market. I think it's highly plausible that semiconductor companies will be more profitable ex ante than the average investment if investors begin expecting imminent transformative growth. Therefore, it makes sense to talk about the growth of semiconductor stock prices relative to broader equities.

You mention that,

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.

This is a very reasonable point, and I agree it's something we should keep in mind when talking about how to select the right AI investment index. However, my (admittedly shallow) understanding is that semiconductor stocks are broadly public, considered hard to unseat by private competitors, and have largely benefitted whenever demand for computer hardware is expected to rise in the near-term future.

The real interest rate is a nice alternative measure to look at, but I'm not convinced that it's more reliable than something straightforward like semiconductor stocks. In addition to some nitpicks I have about the reliability of the argument in the main post (which I've talked to Trevor Chow about, at the very least), I'm just not convinced that the real interest rate is a particularly strong benchmark.

(And to be clear, both measures can be unreliable or both reliable. I'm just pushing back against the framing that one is reliable and the other isn't.)

In the main post, you allude to a "mouth-watering cumulative return of 162%" that one can get over the long-term if they bet on real interest rates rising. However, since my original tweet was posted less than one year ago, NVIDIA has already gone up approximately 161%. If aligned transformative AI is coming soon, I predict we'll again see a similar rise in semiconductor stocks, albeit with big uncertainty about exactly how much, and to what extent public semiconductor companies can capture that value.

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.

Microsoft is planning to invest $10B in OpenAI and rumors are that their maximum revenue is $105B.^[1] 

Microsoft is currently trading at a revenue multiple of ~10x, so this would increase their valuation by $1T, or about 60%.

I think this is enough to show up in Microsoft's stock price. It will be interesting to see what happens to the stock if/when the OpenAI deal closes. My guess is that there will be a small bump, but nowhere near 60%.

1. ^{^}

   Revenue is capped because of the weird term structure. But note that this doesn't include things like increased revenue Microsoft would get from using GPT for Bing, which could be much higher