Superintelligence: Paths, Dangers, Strategies by Nick BostromMy rating: 4 of 5 stars
Superintelligence is clearly a book about philosophy, focusing on a conceptual examination of the development of artificial general intelligence (and, subsequently, artificial superintelligence). Bostrom's book has an almost prophetic ring. Written in 2014, it lays out maps for the possible ways in which AI development might progress, and a decade later it is clear that some of these paths seem well underway. In places, his concerns seem particularly well-founded, especially when discussing human-human interactions in a world where a potentially revolutionary technology is being created.
On the other hand, this age also makes parts of the book feel less relevant to the modern contexts. Obviously certain ideas and decisions that were viable in 2014 seem less viable or important considering where AI-progress has been seen (e.g. the focus on ML and LLMs). Furthermore, the tech-agnostic manner of writing means that Bostrom's examination of concerns is a little bit coarse-grained, for example he does not dig deep into value-loading specifically through linguistic means, which is becoming a more realistic possibility considering advances in the last ten years.
Overall, however, Superintelligence was a great read. Extremely urgent and, of course, terrifying.
View all my reviews
No comments:
Post a Comment