My bookshelves and Kindle library show that in 2020 I started dozens of books but finished few. Here are some of the latter that I liked:
Human Compatible – by Stuart Russell. This is an actual down-to-earth, non pie-in-the-sky book about Artificial General Intelligences: computers that are as intelligent as a person, in a non-specific way. These aren’t machines that can only, say, recognise objects in photographs. These are machines you can have a chat with, and that can figure stuff out.
Creating such entities seems plausible. Since it’s possible in brains, it…should be possible in software? We just don’t know how to do it yet. But that doesn’t mean we know nothing. In fact, we know quite a lot. Including much about what we don’t yet know. This book goes over the current state of research, the challenges ahead, the potential for such technology, the spectacular dangers that could result, and a framework for reducing such risks from the early stages.
AGI is a world-changing possibility, and you don’t need to go full Kurzweil to get excited. Basic extrapolation takes you to interesting places, even without thinking about machines that can themselves build more-intelligent machines. Two ideas that stood out for me:
–machines with human intelligence, but a far greater working memory. Who knows how much knowledge exists out there that just needs someone to join the dots? Feed an AGI every medical journal in the world and see what they spot.
–AI tutors. We know personal tuition works better than class instruction. Once AGI arrives, we teach them how to teach. Then everyone can get a personal tutor. I am particularly fond of this as I think school is trauma for millions of kids – and would love to see schools replaced with something that doesn’t involve children spending a decade being bullied.
But this kind of potential isn’t actually the focus of the book. It’s more about mapping how we might get there, and why we should take the possibility seriously. If you’d like to know more, Scott Alexander has a proper review – one informed by an actual understanding of the AI landscape.
Schrodinger’s Killer App – by Jonathan P Dowling. I wanted to understand what a quantum computer was in more detail than ‘it can do multiple things at the same time, ta-da’ but less detail than ‘here is the equation for the collapsing wavefunction, thus encryption doesn’t work’. I needed quantum computers for someone who has read enough Sean Carroll not to require the double-slit experiment explaining, but still needs their hand holding for the next bit. This was the book! It’s a weird book! Half explainer, half rambly memoir, it takes you down a road and by the end of it you can see how quantum computers work in practice. Not understand, but at least get the general idea without feeling like it’s still a metaphor. You end up knowing why some problems will be helped by quantum computers, and some won’t. You end up knowing why there’s currently only a limited business model for quantum computers (you build one and sell it to the NSA; the end). You get a glimpse of new vistas in the utter weirdness of quantum mechanics, but not in a Deepak Chopra way. You also end up knowing a lot about the author’s life and his philosophy of being. He seemed like a fun, loud guy. He died unexpectedly this year.
The System – by James Ball. I haven’t finished this yet. But it’s a look at the structure of the Internet, from the physical cabling to how data is regulated in a borderless infrastructure – and how the money flows. I thought I knew roughly what was going on. Spoiler: haha I did not.
Superman: Up in the Sky – by Tom King / Andy Kubert. The best Superman story in a long time. No bloody Krypton. No bloody tantrums. No bloody heart-to-hearts with wise priests. It doesn’t even lead you down a trail of whether Superman will do the right thing: *of course he’ll do the right thing*. But how do you decide the right thing? And what happens when you make that decision? It’s about the nature of heroism and moral duty – and the role of inspiration.
Radicalized – by Cory Doctorow. Four short stories about Cory things. Tech. Power. Surveillance. Control. Rebellion. Identity. Look, I don’t know whether his ideas make sense. They’re the far carriage of a long train of thought. I don’t know how people feel equipped to evaluate critiques of capitalism and modern tech without, like, doing a PhD. But as with AGI I think it’s possible to get the gist of what people are thinking about. Cory is great at that. And he writes like a wizard. He sort of zaps you in the first two paragraphs, and then you’re in whirlpool till you’re done. In his attic there’s a shit book where every word is a slog, the characters are flat, and every page drops in overused references to Oscar Wilde. Be warned, though: Radicalized is a grim read. It will not cheer you up.
Outside of books:
–Joe Wicks kept me at least slightly fit
–Mrs Maisel, The Good Place and The Expanse made me laugh and think and occasionally upset me
–I got an air conditioner and don’t know what I’d have done without it
Happy New Year!