Good AI sci-fi
Mini-reviews/descriptions of sci-fi I like that includes AI as a key part of the story. I try to avoid explicit spoilers in these mini-reviews but where necessary to make my point, I will include them.
Over time, I expect some of these mini-reviews to evolve into full-fledged review posts, but for now, they are all short blurbs.
Finally, if you do end up reading one of these recs and have thoughts (positive or negative), feel free to comment here or reach out to me directly!
Books
Crystal Society (⭐️) by Max Harms: This book is about and written from the perspective of an AI sub-agent that’s part of a recently created human level AI. The sub-agent we follow, Face, is the one created to understand and interact with humans, so it’s an interesting take on what a superhumanly manipulative (and understanding) AI’s path to influence and power could be.
Having been written prior to the deep learning era, Crystal Society’s take on how AI would develop and learn feels a bit dated in that the AIs seem to basically learn and think like people. Despite that, I still think the dynamics of reactions to the AI hold up relatively well.
Crystal Society is the first in the Crystal Trilogy. I liked the first and second (Crystal Mentality) better than the third as the third switches perspectives a lot and just got a little beyond my comprehension. That said, I suspect it’s the sort of book that rewards careful rereads, which I have not done.
Short Stories
Tool use by the Humans of Danzhai County from Flight From the Ages by Derek Kunksen (⭐️): An optimistic but not utopian look at what AI automation could do for people and cities set in a county in Southeast China. Unlike the Crystal Trilogy, this story feels pretty informed by the deep learning era in that there’s a heavy focus on acquiring domain-relevant training data to teach AIs to do the various pro-social tasks towards which they’re being aimed.
Autoimmune AI apocalypse by 1a3orn: One of the most persuasive takes on how AI safety could backfire in the form of a short story. 1a3orn knows and follows modern deep learning closely and so, in addition to the persuasive element, has some interesting takes on how models can simultaneously be superhuman but focused.
Tinker by Richard Ngo: A story of an AI designing the compute substrate for its own successor. One of the most detailed, realistic looks at what AI automation of scientific research could look like with lots of fun bio ideas to boot. My maybe controversial take on this is that a huge part of the magic here is the powerful automated bio lab tools the AI uses combined with fast feedback loops. If humans had all that, we could do crazy stuff like this too!
Green and Golden by Richard Ngo: A short but evocative peek inside the mind of an AI whose goal is to empower and help humanity flourish. Has a novel perspective on why AIs having learned so much from imitation vs. Darwinian selection might bias them towards more measured “emotions” and patience.
The Gentle Romance by Richard Ngo: Clearly I really like Richard’s fiction. Narrates through a potential AI trajectory where AI becomes more and more integrated into human life in the form of, first, an assistant, eventually helping humanity evolve into a hive mind of sorts. As is the mark of a thought provoking story, I waffle between thinking the world describe here is a dystopia and a utopia. Richard also manages to weave ideas from internal family systems, which is a psychological/therapeutic theory, that many Twitter folk, myself included, are currently interested in.
Malak by Peter Watts: What if a software bug led military AI to develop a conscience? Peter Watts has a knack for exploring nonhuman cognition, and this is but another example of that.
The Upload by Swyx: Short story about trying to upload a human’s consciousness and how the upload might feel about the whole thing.