A note on historical economic growth
A couple of times in the Most Important Century series (particularly in The Duplicator), I say that economic growth over the last few thousand years is a reasonable fit with the pattern (described here) of accelerating growth, driven by a feedback loop: "more ideas → more output → more people → more ideas → ..."
This point is the subject of an ongoing debate (see this EA Forum post by Ben Garfinkel, and the extensive back-and-forth in comments).
My best guess is that the past data is, in fact, a reasonable (though ambiguous) fit with the pattern of accelerating growth. However, I'm far from confident of this, and I want to address how it would affect my arguments if better, future data turned out to decisively undermine this fit.
Extrapolating future economic growth based on (a long view of) past economic growth
I've cited the projection, made in Modeling the Human Trajectory, that the economy is "on track" to hit infinite size this century if the pattern seen in the past continues. If it turned out that past data is inconsistent with accelerating growth, this would undermine Modeling the Human Trajectory, and a new extrapolation would be needed. However, my best guess is that a good replacement extrapolation would still show a good chance of explosive (even "infinite") growth this century. Reasoning for this guess follows.
When discussing the pattern of past growth, the main alternative I've seen to accelerating growth (including in the EA Forum post linked above and comments) is a series of fundamentally different 'growth modes,' each with its own growth dynamic and/or growth rate. For example, perhaps - rather than thinking of economic history as a gradual acceleration - one could think of it as divided into distinct phases:
- A pre-agriculture phase (starting some millions of years ago), in which growth was likely extremely slow and perhaps pretty chaotic.
- A phase after the development of agriculture (starting ~10,000 years ago), during which growth was probably faster than before, but still quite slow by today's standards, and perhaps pretty chaotic as well.
- The modern, post-Industrial-Revolution phase (starting ~200 years ago), with by far the fastest growth.
It seems undisputed to me that the third phase is both much shorter (in calendar time) and has dramatically faster growth, compared to the first two. This could be the result of continuous acceleration, or it could be because a fundamentally new growth mode emerged. The latter would then raise the question of whether a transition to another, still faster "growth mode" might be possible.
Robin Hanson's 2000 paper, Long-Term Growth As A Sequence of Exponential Modes, is the main attempt I know of to explore that question. It attempts to model long-run economic history using a couple of different approaches, both of which are designed around the idea of "growth modes," and (on pages 14-17) to extrapolate patterns observed to date into the future. It states:
In summary, if one takes seriously the model of economic growth as a series of exponential growth modes, and if relative change parameters of a new transition are likely to be similar to such parameters describing old transitions, then it seems hard to escape the conclusion that the world economy could see a very dramatic change within the next century, to a new economic growth mode with a doubling time of roughly two weeks or less ...
If the next mode had a “slow” doubling time of two years, and if it lasted through twenty doubling times, longer than any mode seen so far, it would still last only forty years. After that, it is not clear how many more even faster growth modes are possible before hitting fundamental limits. But it is hard to see how such fundamental limits would not be reached within a few decades at most.
This is qualitatively pretty similar to the projection I've given in the blog posts: both imply a dramatic economic acceleration in the 21st century, and both imply "infinite growth" or "hitting fundamental limits" not too long after (although the potential delay is longer for Hanson's approach, and could go modestly into the 22nd century depending on which Hanson projection one uses).
This extrapolation is less straightforward than the Modeling the Human Trajectory extrapolation. There aren't very strong reasons to think that the series of growth modes will follow any particular pattern, in terms of how they're timed and what kind of growth they bring. Hanson's extrapolation is merely a best guess at what to expect if they do follow a relatively regular pattern. Still, it does look reasonable to me as a best guess.
Other implications if it were to turn out that past economic data does not fit an "accelerating growth" pattern
- Throughout the series, I argue that various technologies (The Duplicator, digital people, "PASTA"1) could lead to an "accelerating" pattern leading to explosive growth. This is an implication of most mainstream theoretical models in growth economics, as discussed in Report on Whether AI Could Drive Explosive Economic Growth.2 I cite the fact that past data seems to fit this dynamic as further support that such a thing is plausible. If past data did not fit the dynamic, it would not affect this theoretical case for expecting explosive growth, but it would make the overall solidness of the case some amount weaker.
- I also will argue against the idea that "If transformative AI were to be developed this century, it would break the pattern we've seen of constant economic growth; therefore, we should have a very high burden of proof for predictions of transformative AI this century." For this purpose, either the "accelerating growth" or "series of different growth modes" dynamic seems sufficient for my case that we should consider a future growth explosion plausible, although I do think the case would be a bit weaker if it had to rely on the latter as opposed to the former.
Overall, if it became clear that economic history contains very little acceleration (and is instead best thought of as a series of distinct "growth modes,") I think my remaining claims and conclusions would still look about right, though the arguments would be some amount weaker.
It's also possible that if we had perfect information about long-run economic history, we would see a mix: some instances/periods of the "accelerating growth" dynamic described here, some periods that look more like "distinct growth modes."
Process for Automating Scientific and Technological Development - to be discussed in a future piece. ↩
More precisely, most models imply that full automation of both R&D and goods production would lead to explosive growth. What about growth before full automation of both these things? First, if automation proceeds more rapidly than its historical rate before full automation, then growth models typically imply growth will begin to accelerate before we achieve full automation (e.g. see section 18.104.22.168 of the report). Second, if R&D but not goods production is fully automated, I think this would be sufficient for explosive growth (see section 6.1.6 of the report). ↩