Robin Hanson’s The Age of Em: Work, Love, and Life when Robots Rule the Earth gives us an impressively realistic and unconventional picture of how the future might be different in the domains we most care about. I think there should be more books like it.
Many people have been struck by the realism of Age of Em, but what strikes me about it is the particular kind of realistic it is. It does not seem plausible in the way that most realistic fiction seems plausible. A common method used by both liars and fantasy authors to make stories seem more plausible is including as many details as possible. Adding extra corroborative details is a way to exploit the conjunction fallacy in the minds of your readers, and thereby make a story seem more probable than it should.
Age of Em refuses to use this trick. Thanks to that refusal, it paints a uniquely original vision of the future unlike most anything else in the futurist canon. It is also thanks to that refusal that the book actually seems totally implausible in a way—implausible in the way that a thorough ethnography of a distant culture might seem totally implausible to those who have only ever interacted with cultures much like their own. The way in which Age of Em manages to be realistic is by ensuring that every feature of the future it paints is independently justified by its own piece of reasoning; every bit of absurdity gets its own bit of evidence.
But this isn’t a review of Age of Em (there are already many excellent reviews). Rather, it’s a push for more books (or articles or blog posts or what have you) to be written using Age of Em‘s methodology. In one of his talks on Age of Em Hanson gives us a formula for writing a book like his.
Paraphrasing Hanson: There are a few ways that you can do futurism. One method is looking at current trends and extrapolating them forward. For instance, you could look at the change over time in attitudes towards sex, the price of compute, the amount of time spent on leisure, or the rate of paradigm shifts in physics, etc and draw curves through those trends to predict the future. The other method is to imagine disruptive technologies, and then try to imagine the details of those technologies. Some common technologies that get imagined are space colonization, blockchain, self-driving cars, human-level artificial intelligence, etc.
Hanson’s proposed third way of doing futurism is to imagine a disruptive tech, then rather than focusing on the details of the technology itself, or trying to figure out what effects it might have on current trends, use orthodox social science to answer interesting questions about what society would be like if that disruptive technology were suddenly introduced. The key here, as I understand it, is using established, orthodox social science to answer the interesting questions.
Age of EM picks whole brain emulation as its disruptive tech. Just as an example of Hanson using orthodox social science to answer interesting questions, Hanson asks what sorts of people/brains institutions will be interested in uploading. The costs to upload a brain would likely be large at first. He concludes that they will only want to upload the hardest working, most intelligent, most creative, etc. There will essentially be a frontier of combinations of properties that are correlated with productive labor, and only people who are on or beyond that frontier will be uploaded by institutions. There is not much point in paying to upload a slightly less productive person, since it would be cheaper to just make a copy of the upload we already have. It would only be worth it to upload someone new if they had characteristics that were sufficiently different from those of all the other uploads so as to provide some advantage that could not be had by making another copy of an existing upload.
We can infer from this that the world of ems (this is what Hanson calls the uploaded humans) will be populated by mostly identical copies of only a few people, possibly only a few hundred original brains. Ems who are nearly identical to each other by virtue of being emulations of the same original brain will then likely organize themselves into clans based on that similarity. An Alice–em could check with the other Alice-ems to see if they liked a movie before she decides whether to watch it. She could also check with other Alice-ems before she decides whether to date a Bill-em. This would give her a lot of information about how relationships between Alice-ems and Bill-ems normally work out, what pitfalls to look out for. There might be many things she will decide to do differently with the benefit of a giant dataset on past relationships that were almost identical to hers.
We could further expect that the institutions of this society would treat clans as a natural social category, and organize themselves so as to reflect that fact. For instance, voting might happen on a clan by clan basis rather than on an em by em basis, since ems of the same clan likely share many of the same values and interests. Similarly, contracts might be drawn up that take whole clans as the relevant legally bounded parties rather than individual ems.
Now this is just one of a hundred strange consequences of whole brain emulation predicted in Age of Em. Society being organized into a few hundred clans consisting of nearly identical ems who are very different from members of other clans does not seem like an obvious next thought to have after asking what would happen if whole brain emulation became feasible, but I do think it follows from orthodox social science and just a few other premises. (If you’re not convinced, please read the book. It makes a much better case than I do paraphrasing it here.) Since it does follow, why haven’t we all heard this before? Hanson proposes an explanation.
Some of the other premises needed are facts about the physical sciences. How much compute does a brain use? How much compute will we have? How much energy will be available? Will energy or cooling be the bottleneck? How fast are brains and how much faster could they be? These are the sorts of questions that engineers would be better at answering than social scientists. This is the thrust of Hanson’s explanation as to why this method of futurism is neglected. People who know a lot about STEM are likely to imagine disruptive techs, but either do not know much about the social sciences, or fail to take them seriously. People who know a lot about social science often either do not know much STEM, or else are skeptical of the disruptive techs that STEM oriented futurists propose.
This is part of the reason I am interested in this kind of work. I am lucky enough to not know much about STEM or social science!
Anyway, I hope I have convinced at least some readers that this is a valuable and neglected way of doing futurism. Amazingly, it’s even somewhat formulaic. We have an algorithm for producing interesting, original, and valuable work, and the algorithm only has two steps!
- Pick a disruptive tech.
- Figure out the social consequences of that tech being introduced.
The breadth of knowledge required to carry out that algorithm is likely to be a sticking point, but if Hanson’s explanation of why this sort of futurism has been neglected is right, then this seems like a natural candidate for interdisciplinary collaboration. I would certainly be interested in collaborating on a project like this with a computer scientist, physicist, historian, economist, etc. And since what makes this kind of project valuable is using established, orthodox theory to answer unorthodox questions, rather than showcasing a mastery of cutting edge theory, I think smart and committed graduate students, or even amateurs, could make a lot of useful progress on a project like this.
Writing a book is a bit above my current pay grade, but you may see a series of posts on this blog about some interesting ways we might expect society to look if particular disruptive techs suddenly became viable.
I will leave you with a list in no particular order of disruptive technologies that I would be excited to see plugged into the formula. Some of these might be flops, but I’m sure at least one of them could lead to something awesome. I encourage you to suggest other disruptive techs in the comments, or propose questions about particular ways that society might be different if one of them became technically feasible. If you’re really feeling courageous, you might even try to answer the question you propose.
- Improved surveillance technologies
- Extremely accurate lie detection
- Better than human general sequence prediction
- Automated computer programming
- Much better RL
- Something like Paul Christiano’s HCH, or IDA
- Better than human preference inference
- GPT bajillion
- Solved protein folding
- Far more, far cheaper compute than we were expecting