Back in June, I read Robin Hanson’s book: Age of Em: Work, Love and Life when Robots Rule the Earth. I found it fascinating. Two months later, the book’s ideas still pop into my mind daily. Nothing else I’ve read in the past year has done that. This post is a summary of the core ideas in Age of Em, followed by my observations and critiques.
I expect my analysis to be relevant for a large cloud of different but similar scenarios. In particular, conditional on my key assumptions, I expect at least 30% of future situations to be usefully informed by my analysis. Unconditionally, I expect at least 10%.
– Robin Hanson, Age of Em
Let me make my bias known: I am a fan of Robin Hanson. He possesses a rare combination of intelligence, expertise, creativity, and civility. Moreover, he makes exceptional use of these gifts. Hanson looks at the big picture– the really big picture. For example: When examining the growth of humanity over millions of years, the invention of writing is a rounding error. At such a high-level view, all of human history can be described by three eras of growth1:
- 2,000,000 BC to 5,000 BC: Foraging, in which population doubled every 250,000 years.
- 5,000 BC to 1600 AD: Farming, in which population doubled every 1,000 years.
- 1600 AD to present: Industry, in which GDP doubled every 15 years.
These numbers are approximate, of course. Hanson also notes that each transition to the next era took less than a doubling time of the previous era. E.g. industry replaced farming as the dominant growth mode in less than the doubling time of farming (1,000 years).
If one extrapolates the numerological trend (admittedly, an absurdly weak argument), the next era after ours would have an economic doubling time on the order of months or weeks. This sounds crazy. How could a 200x speedup in economic growth possibly happen? What would such a world look like? Age of Em is Hanson’s attempt to answer these questions.
Today, economic growth is bottlenecked by people. Machines can be mass-produced, but people take decades to mature and learn skills. It’s possible to make more machines per worker, but one quickly runs into diminishing returns. Someone with 10 laptops or 10 bulldozers won’t be 10x as productive.2 To speed-up economic growth by 200x, the bottleneck must be removed. Something must substitute for human intelligence.
When considering substitutes for human intelligence, most people’s thoughts turn to artificial intelligence. But Hanson is pessimistic about the arrival of de-novo AI. He thinks it is more likely that brain emulations will come first.3 (The “em” in Age of Em is short for “emulation”.) Creating such ems requires three key technologies, and none of them exist today:
- Fast, cheap computers.
- High-resolution scans of one or more human brains.
- Accurate models of all relevant cell types in the brain.
Hanson doesn’t give any concrete timeline for when these technologies will exist. He only says “some time in the next century”. Fortunately, none of his analysis depends on when ems become feasible, just that they arrive before AI.
Once ems are made, they’ll differ from us in several ways:
- Ems are immortal…-ish. So long as their hardware is maintained and powered, they can live indefinitely. This isn’t as great as it sounds. Today’s cars and houses are just as “immortal”.
- Ems can run at different speeds. Faster ems cost more to run, similar to how faster cloud servers cost more.
- Ems can “teleport” by transmitting their mind-state across a network. If latency is low enough, they need only to transmit their input and output signals.
- Ems can be copied. Many consequences stem from this one difference.
What makes Age of Em interesting is that it rigorously explores the consequences of ems, and that it does so by applying standard scientific models. There’s no hand-waving or contrarianism. The only unusual thing is the question being asked: “What would things be like in an em world?”
Seen up close and honestly, I expect the future usually to look like most places: mundane, uninspiring, and morally ambiguous, with grand hopes and justifications often masking lives of quiet desperation. Of course, lives of quiet desperation can still be worth living.
– Robin Hanson, Age of Em
Reading Age of Em opened my eyes to how sloppy most futurism is. The field is filled with incomplete analyses and moralizing about the present day. In contrast, Age of Em treats the future like a real place. To use language from construal level theory: It kept me thinking in near-mode. I was pleased by the broad range of topics covered. These included: thermodynamics, software engineering, reversible computing, surveillance, city planning, language, and charity. No matter one’s interests, there’s something to engage with.
Two specific parts of the book captivated me. They were software-related, of course. One section described the life of a software engineer:
For software engineering tasks where parallel software and tools suffice, and where the software doesn’t need to interact with slower physical systems, em software engineers could be productive even when sped up to the top cheap speed. This often makes it feasible to avoid the costs of coordinating across many engineers, by having a single engineer spend an entire subjective career creating a large software system. For an example, an engineer that spent a subjective century at mega-em speeds would complete this period in less than 1 objective hour. Thus when such a delay is acceptable, parallel software may be written by a single engineer taking a subjective career length.
This scenario may seem absurd, but it’s a decent prediction of what will happen in a world where minds can be copied and run at different speeds. While you may not want to live such a life, there are some people who will choose it. Those minds will dominate the em economy.
Another section briefly discussed software rot. That is: Adapting mature software to new circumstances requires more time and effort than creating new software from scratch. There are tons of examples of this in the open source world. I’ll write another post on software rot and ways to mitigate (though not solve) it.
Hanson claims to be applying standard models and theories to his em scenario, but I don’t know enough social science to tell if that’s true. When the text entered my areas of expertise (computer science, software engineering), I found nothing objectionable. Also, I haven’t heard any domain experts disputing his claim, so I’d bet money that Hanson correctly applied standard theories to this scenario.
I’m not sure if the author intended it, but the text works well as an ebook. There are plenty of references and links to related chapters, so I had no problem jumping around. Many footnotes contained URLs, making it easy to track down more info on conclusions I was more skeptical about. I wish more ebooks took advantage of the medium in these ways.
In sum, even though many critics have reasonable points, I still think the analysis in this book was worth the effort.
– Robin Hanson, Age of Em
While the ideas fascinated me, the writing itself was rather matter-of-fact. It wasn’t as dry as an academic paper, but neither was it particularly compelling. This could be taken in a positive light. It means the book succeeds on its ideas alone, not just due to fancy writing. And it’s not all dry. The penultimate chapter –on policy recommendations– uses more emotionally stirring language.
When it comes to Age of Em’s core ideas, I can’t find any major flaws. To dispute the book requires either rejecting Hanson’s core assumptions or rejecting widely-accepted scientific models. The book contains so much detail that it would be easy to nitpick, but such pedantry is tiresome to read and to write.
In my opinion, the weakest part of Age of Em is the assumption that em minds won’t be very tweakable. To be fair, I think it’s a valid simplifying assumption. The book is a straightforward, first-cut analysis. Complicating the base assumptions would have made for an intractable project. Still, one should not be very confident about how tweakable brain emulations can be. And even if most tweaks are hard, there may exist a few easy tweaks which could create a world we find utterly abhorrent. I’d really like to know what neuroscience has to say about this.
…many of your ancestors would be tempted to disown you, if they were told many things about you. While they’d be pleased and impressed by many of your features, other things about you might horrify them.
– Robin Hanson, Age of Em
When examined closely, the world described in Age of Em doesn’t sound so bad to me. Ems will be smarter, healthier, and harder working than almost everyone today. They’ll live long and fulfilling lives. Compared to our civilization, theirs will be larger, more capable, and more robust.
I think many detractors neglect that last adjective: robust. Existential risk is a crucial consideration. Hanson notes that it would be very hard to eradicate ems. Also, sped-up ems could monitor nanotech or AI experiments and react far more quickly than humans. Even if one thinks an em world is less desirable than ours, the reduction in existential risk may be worth it.
Lastly: As positive as my review is, I must admit that Age of Em will only interest a small subset of people. Fortunately, most people reading this will qualify.
You might ask, “If people are the bottleneck for economic growth, and population doesn’t double every 15 years, then how does our economy do that?” The answer is that much of today’s economic growth comes from making better machines. ↩