Grading Extropian Predictions
How well do futurists predict the future?
The Extropians are a group of techno-optimist libertarian futurists, most active in the late 90s. Although a relatively small group, they are still influential thanks to their extremely high per capita rate of important members. Eliezer Yudkowsky, Nick Bostrom, Robin Hanson, all three of the likely candidates for Satoshi Nakamoto (Hal Finney, Wei Dai, and Nick Szabo), Eric Drexler, and Marvin Minsky were all Extropians. If you want to learn more about this group and the people in it, check out my website on Extropian Archaeology where I do a deep dive into the archives of primary sources and uncover the lasting influence of the Extropians.
In the 1995 issue of Extropy magazine there is a four page spread of future forecasts from seven illustrious members of the Extropian community: UC Irvine astrophysicist Gregory Benford, president of the cryonics organization Alcor Life Extension Stephen Bridge, nanotechnologist Eric Drexler, transhumanist philosopher FM-2030, Xerox PARC computer scientist Mark Miller, the Extropian founder Max More, and cryptographer Nick Szabo.
Before reading on, register a guess for who you think will have the best prediction record.
Not all of the predictions have been resolved yet, but even though the list of predicted events includes “Reproducing Asteroid Eaters” and indefinite biological lifespans at least one of the respondents predicted a pre-2023 date for almost all of the events. So, how did these professional futurists do. Each prediction is marked with a color. Orange means the predicted date hasn’t come and the event hasn’t happened yet so the prediction is ambiguous. Red is wrong, Green is correct. A two-colored X means something in between. Red-orange means a range that we’re partway through their predicted range without seeing the event happen yet. Red-green means the event has arguably happened but its up for interpretation or that the predicted date is just on the edge between close enough to count as correct.
Starting with the first 8 events it doesn’t look great. Of the predictions whose dates have passed, none are unambiguously correct. The most common mistake was over-optimism for frozen organ transplants which, while seemingly possible, are still not common. Drexler got the most wrong in this section with wild optimism for life extension, cryonics, and intelligence enhancing drugs.
The only event that was partially green is “Fine-tuned mood/motivation transformation drugs.” Whether they are fine-tuned or not is certainly up for debate but I gave predictions around 2020 half marks because of recent advances in impulse control drugs like Semaglutide.
This section looks a bit better but still far from clairvoyance. The first full correct scores go to Bridge, Miller, More, and Szabo for human germ-line gene editing. I’m giving a generous range around 2018 after the first clinical trials for CRISPR-based (though not germ-line) gene editing started and after the germ-line editing tests in China.
I’m not exactly sure what “Completely genetically composed children” means. I’m assuming it means a functional embryo is created by custom-synthesizing all the DNA rather than getting sperm and egg donors, so certainly not something that has happened yet and it doesn’t seem very likely in the near future.
The most common mistake was being too early on human cloning. Dolly the cloned sheep wouldn’t be revealed until two years after this issue in 1997 so a more Extropian-specific source of hype is to blame. Nick Szabo and the cryonics specialist, Stephen Bridge, will probably be pretty close on extinct species reanimation. And an artifical womb by 2050-2100 seems pretty likely too.
This section is rough. Zero unambiguously correct answers and lots of wrong ones. Extropian excitement for cryonics again misleads them. Benford barely escaped with a technically indeterminate prediction of a billion dollars a year in the cryonics industry in 2035 but it’s pretty clear this will not happen. By 2014 only 1,500 people have signed up for cryopreservation in the United States.
The seven Extropians were also overly optimistic for nanotech. The only event I’m not totally sure about is the “Atomically detailed design” for self-reproducing assemblers. Some searching doesn’t produce a clear example and it feels like the sort of thing where if the conditions of this event were satisfied in good faith it would be hard to miss so I’m counting it as not happened yet. Benford’s unresolved predictions in this list seem the most reasonable but only time will tell.
So far, Drexler is doing the worst with over half of his predictions unambiguously wrong.
This is the best section so far. The Extropians were correctly optimistic about internet content getting big fast. This section also has the first predictions which are wrong for being too late: Bridge and More underestimated falls in memory storage costs and almost everybody underestimated progress brain computer interfaces.
I was generous with predictions on the first two questions since most of their answers could be correct depending on the metric you use. Benford and Bridge are almost spot on for when internet news overtakes TV in viewership. FM-2030 is technically correct but 1990+ is a lazy prediction so he only gets half points. Szabo gets points on this event too because most publications had some electronic distribution within his range even if it didn’t comprise the majority of content views yet.
Similarly for intellectual publications, the ambiguity between the scalar quantity of the number of viewers on the web vs the binary measure of whether each publication has a web version gives the respondents some wiggle room. Still, most of them were close enough to either metric to merit full points.
On brain computer interfaces the Extropians were surprisingly pessimistic. By 1995 there had been lots of progress with non-invasive EEG interfaces and some invasive BCIs in animals so I’m not sure why their horizons were so long. The first invasive BCI was implanted into a human in 1998, so only Mark Miller was even close to the correct range.
On super-human computers and AIs, More’s unresolved predictions strike me as the most reasonable; Drexler adds to his terrible record.
Almost everything in this section is predicted to happen beyond 2023 so it’s hard to evaluate. Except for Drexler’s predictions which continue to be way off. Bridge, Miller, More, and Szabo will probably be pretty close on the Mars landing, but I think they are far too optimistic about asteroid mining coming just a couple decades later.
The most interesting part of the final section is the two questions on cryptocurrency. I think that if you had told the Extropians that cryptocurrencies had hundreds of millions of users and trillions of dollars of value but the world doesn’t look much different because of it they would be very confused.
The ultra-libertarian Extropians are understandably over confident about privatization of law and education, but they are surprisingly pessimistic on betting markets. I give Szabo half points for his prediction of big policy influence because betting markets are a big source of polling data in American elections, but whether that constitutes a “big” policy influence is up for debate.
Final Scores
Counting only predictions that have been resolved, the order from highest to lowest proportion correct is Miller (6/12), Szabo (5.5/12), Bedford (5.5/15), Bridge (3.5/14), More (3.5/15), FM-2030 (2/19), and Drexler (0.5/22). Mark Miller comes in first, although he got two half-points for his predictions in the last section which have infinite range so he should probably be downgraded to (5/12) putting Nick Szabo on top.
My subjective ordering based on best-guesses for how the later predictions will resolve would be Szabo, Bedford, Miller, Bridge, More, FM-2030, and Drexler.
I am not sure how the success of these predictions would compare to the same questions asked to different groups. My sense is that the average person would answer never to almost all of these questions, though they might still come out ahead of Drexler. Measuring by what’s resolved in 2023, I think that a normal, centrist computer science professor would have outperformed most of the Extropians here. They would be just as good on Moore’s law and the growth of online content but they wouldn’t be so pollyanna-ish about cryonics and free-market politics.
The Extropians have an incredible record of cultural and intellectual influence, but their predictions seem biased by their unbounded optimism. Their records might improve as we move further into the 21st century and more of their predictions are resolved, but so far the Extropians have underestimated the barriers to technological progress.
I was going to do this on my own blog, having recently looked through issues of Extropy. Oh well, now I'll have to write a commentary post.
"Almost everybody underestimated progress brain computer interfaces.' I disagree. We didn't mean simple deep brain stimulation or "non-invasive EEG interfaces". My forecast of 2020-2050 is still in the running.
For "Information storage $0.01 per Megabyte": If 2010 is marked correct, but my 2015 is marked wrong, it seems that unless you get the exact year you're counted as wrong. If, as in most cases, I'd given a range centered on that date, I would have been counted as correct. This suggests something about grading forecasts and suggests to forecasters never to give a point date. The same applies to "Most publications are electronic." It seems that I get graded as wrong but Nick as right although his five-year range starts only one year after mine. Looking at 1999 on the chart you provide, it's looks like 1999 is not wrong because the proposition was NOT "most intellectual publications are ONLY on the web."
What year do you have for "1 million+ people using anonymous electronic cash"? If Benford's 2010 is correct but my 1999-2006 is not, it must be 2007-2010. In that case, I was close and a slightly wider range would have got it.
Why is Ocean Colonization 2010-2050 marked incorrect when there is still 27 years left?
We should have been right on cryonics-related predictions, but non-technical factors have been extremely disappointing.
Not long after these forecasts, I concluded that it's pointless to make date forecasts. You can maybe foresee general trends, but even those are hard -- unless they are VERY general.
There were and still are women there in this extropic network of visionaries.