8 July 2022
Atomizing the Sequences
by Rachel Shu
Status: phasers set to stun, totally not finished
Rationality: From AI to Zombies wasn’t written to be canonized, but there hasn’t been a serious attempt to just completely revamp it in the past 7 years. In the meanwhile, it’s been edited for clarity, published as hardcover books, received multiple summaries, website refurbishings, and highlights. All of these are appreciated, but risk canonizing the original work, especially those which do more to strip it of its context (especially the comments, in which some of the concepts are highly contested.)
In fact, a lot of the core rationality publications are awfully pat.
- Rationality: From AI to Zombies, with exactly 26 chapters, one for each letter A-Z. Things being tidy should raise suspicion.
- The Twelve Virtues of Rationality, a very nice number. Why not 4, 17, or 122?
Arbital was the last serious try to replace the sequences; and it failed for a number of technical reasons. New tooling for this is currently on the horizon; one of the most useful, Roam, was written by someone who’d found the sequences influential in their life. Other key developments exist in the tools for thought space, notably contributions by Andy Matuschak.
There are some (non exhaustive, non definitive) simple principles that a revamped rationality project ought to follow.
- Atomic. One idea per webpage, with explanation, key evidence, and supporting examples, clearly differentiated.
- Unlike the current sequences, ideas which are corollaries should get subpages. Right now the blog posts don’t have much informational hierarchy
- Books are generally bad for information storage and retrieval and could be much better.
- Lively. Fun and engaging to read, with colorful examples easily understood and relevant to daily life.
- The sequences are, in some sections, some of the most inspiring writing I’ve ever read. A successor should be just as nice.
- Contestable. Claims should be assigned compound probabilities. Claims which are incorrect should be revisable by public comment.
- Many claims about cognitive bias are drawn from parts of
- One way of operationalizing this would be to place replication markets or other types of evidence-based betting on
- Impartial. Take the sequences as a starting point, but be prepared to toss unsupportable or badly formed concepts.
- ~30% of sequence concepts seem to be in this category.