This book delves into the complexities of otherness and control in the context of artificial general intelligence, offering valuable insights on how these concepts impact our understanding and management of AGI. It explores the ethical, philosophical, and practical dimensions of AGI, making it a crucial read for those interested in the future of technology and society. Cover painting: “The Creation” by Lucas Cranach
"AI risk is substantially... a generalization and intensification of the sort of 'balance of power between agents with different values' problem we already deal with in the purely human world."
This is a collection of essays by Joe Carlsmith in which he philosophizes about our future with AGI. Especially he writes about different ways of relating to AGI (artificial general intelligence), especially through the lens of the way humans have generally related to Otherness, be it animals, other humans sufficiently different or aliens (in works of fiction). The other main aspect of the essays is control, which might in the future, in the age of AGI, not be what we as humans might have anymore - or at least not the most. As many people in the field, Carlsmith is concerned about existential risk from artificial intelligence - sometimes also called just AI risk. He engages with different perspectives but focuses mostly on the work of Eliezer Yudkowsky.
Carlsmith is a rigorous thinker, and a gifted writer. At times he draws inspiration from different movies and literature for analogies, which helps a lot with the understanding of the sometimes quite complex ideas. In general, I would say that the essays are geared towards readers with some background knowledge about the AI safety discourse.
This collection of essays develops a set of ideas around the prospect of AGI and how we as humans should handle it. The essays vary in topic and depth. A lot of it focuses on responding to Yudkowskyian style thoughts around AGI, and so it has some of that flavor of LessWrong content, though in a more elegant and humanisticly educated style.
But a few of the essays are more tangential to AGI, and these, I think, are the better works in the collection by far. In particular, the essay On Attunement is a beautifully written meditation that reaches for, and I think finds, something quite deep about human experience and the universe. As with many of the best writings, On Attunement is itself an instantiation of its subject, and there is no higher praise I can offer.
On Attunement is a masterwork, and its well worth reading the whole collection just to have the full context for that one essay.
4/5 because On Attunement is 5/5 and much of the rest is nearer a 3 (which on my scale is still pretty good)
Articulates a set of insights I've gathered from a rather similar canon to the author. (He's missing Dostoyevsky, I haven't read Angels in America, but we agree on Annie Dillard, Martin Buber, and Le Guin.) Of course the problem is in the actual practice, especially as I am inclined to neither stillness nor empathy. It's difficult to see how this practically translates to alignment, unless the answer is 'try the liberalism again'.
A mind at work, trying to better understand how humans relate to the universe and to our creations. I loved it! It is uneven in places - it's an essay series, in some sense an extended set of notes on investigations, not a well-edited book serving a particular narrative purpose, and so some bits are rather tangential. But I really enjoyed the depth and earnestness of the underlying quest. And certain ideas - that of Deep Atheism, in particular - seem to me to be very important.
Edit: two months after finishing this, and I realize I've been thinking about it very frequently ever since, and it's changing my thinking (and beliefs) in important ways.