Future Timeline, in Econ Growth Units

Polls on the future often ask by what date one expects to see some event X. That approach, however, is sensitive to expectations on overall rates of progress. If you expect progress to speed-up a lot, but aren’t quite sure when that will start, your answers for quite different post-speed-up events should all cluster around the date at which you expect the speed-up to start.

To avoid this problem, I just did 20 related Twitter polls on the distant future, all using econ growth factors as the timeline unit: “By how much more will world economy grow between now and the 1st time when X”.

POLLS ON FUTURE (please retweet)

World economy (& tech ability) increased by ~10x between each: 3700BC, 800BC, 1700, 1895, 1966, 2018. In each poll, assume more growth, & give best (median) guess of how much more grow by then.

POLLS ON FUTURE (please retweet)

World economy (& tech ability) increased by ~10x between each: 3700BC, 800BC, 1700, 1895, 1966, 2018. In each poll, assume more growth, & give best (median) guess of how much more grow by then.

— Robin Hanson (@robinhanson) July 7, 2020

Note that I’ve required a key assumption: growth continues indefinitely.

The four possible growth factor answers for each poll were <100, 100-10K, 10K-1M, and “>10M or never”. If the average growth rate from 1966 to 2018 continues into the future at a constant growth rate, then these factor milestones of 100, 10K, 1M will be reached in the years 2122, 2226, and 2330. That is, the world economy has been growing lately by roughly a factor of 100 every 104 years.

I’ve found that lognormals often fit well to poll response distributions over positive numbers that vary by many orders of magnitude. So I’ve fit these poll responses to a lognormal distribution, plus a chance that the event never happens. Here are the poll % answers, % chance it never happens, and median dates (if it happens) assuming constant growth. (Polls had 95 to 175 responses each.)

Many of these estimates seem reasonable, or at least not crazy. On the whole I’d put this up against any other future timeline I know of. But I do have some complaints. For example, 21 years seems way too short for when <10% of human protein comes from animals. And 35 years until <20% of energy comes from fossil fuels seems more possible, but still rather ambitious.

I also find it implausible that median estimates for these four events cluster so closely: ems to appear, frozen humans to be revived, and AI to earn 9x humans, and AI to earn 9x humans+ems. They are all in the same ~2x growth factor range (factors 670-1350), and thus all appear in the same constant-growth 16 year period 2165-2181. As if these are very similar problems, or even the same problem, and as if they reject what seems obvious to me: it is much harder for AI to compete cost-effectively with ems than with humans. (Note also that these are far later dates than often touted in AI forecasts.)

My main complaint, however, is of overly high chances that things never happen. Such high chances make sense if you think something might actually be completely impossible. For example, a 46% chance of never finding aliens makes sense if aliens just aren’t there to be found. A 25% chance that human lifespan never goes over 1000 might result if that is biologically impossible, and a 11% chance of no colony to another star could fit with such travel being physically impossible.

A 31% chance nukes never give >50% or energy could result from them being fundamentally less efficient than collecting sunlight. And a 6% chance that AI never beats humans, an 12% chance that we never get ems, and an 19% chance that AI never beat ems could all make sense if you think AI or ems are just impossible. (Though I’m not sure these numbers are consistent with each other.) Most of these impossibility chances seem too high to me, but not crazy.

But high estimates of “never” make a lot less sense for things we know to be possible. If there is a small chance of an event happening each time period (or each growth doubling period), then unless that chance is falling exponentially toward zero, the event will almost surely happen eventually, at least if the underlying system persists indefinitely.

So I can’t believe a 50% chance that the human population never falls to <50% of its prior peak. Some predict that will result from the current fertility decline, and it could also happen when ems becomes possible and many humans then choose to convert to becoming ems. Both of these scenarios could fit with the estimated median growth factor 152, date 2132. But a great many other events could also cause such a population decline later, and forever is a long time.

The situation is even worse for an event where we have theoretical arguments that it must happen eventually. For example, continued exponential economic growth seems incompatible with our physical universe, where there’s a speed of light limit and finite entropy and atoms per unit volume. So it seems crazy to have a 22% chance that growth never slows down. Oddly, the median estimate is that if that does happen it will happen within a century.

The 13% chance that off Earth never gets larger than on-Earth economy seems similarly problematic, as we can be quite sure that the universe outside of Earth has more resources to support a larger economy.

For many of these other estimates, we don’t have as strong a theoretical reason to think they must happen eventually, but they still seem like things that each generation or era can choose for itself. So it just takes one era to choose it for it to happen. This casts doubt on the 39% chance that the biosphere never falls to <10% of current level, the 28% chance that ten nukes are never used in war, the 24% chance that authorities never monitor >90% of spoken & written words, and the 22% chance we never have whole-Earth government.

The 28% chance that we never see >1/2 of world economy destroyed in less than a doubling time is more believable given that we’ve never seen that happen in our history. But in light of that, the median of 70 years till it happens seems too short.

Perhaps these high estimates of “never” would be suppressed if respondents had to directly pick “never”, or if polls explicitly offered more larger growth factor options, such as 1M-1B, 1B-1T, 1T-1Q, etc. It might also help if respondents could express both their chances that such high levels might ever be reached, separately from their expectations for when events would happen given that such high levels are reached. These would require more than Twitter polls can support, but seem reasonably cheap should anyone want to support such efforts.