The Beginning of Infinity
David Deutsch, 2011. Status: reading, ch. 1. Started 2026.09.04. Updated 2026.09.07.
Why I’m reading this
I’ve had this book (among many, many others) on my bookshelf for many years, as a book that I would have liked to have read, but never having even opened. A few days ago, I was chatting with an AI researcher at a frontier lab who recommended reading the first 60 pages of this book, reassuring us that it would be challenging but worth it. It also coincided with my resolve to have more clarity of what my bet as an AI for Science researcher is, and to my great delight this book seems like a great sparring partner for the next several weeks (or months). In particular, some of the questions I have in mind approaching this book are: “How is scientific understanding different from prediction?” “What is a theory?” “Can scientific understanding be compressed into the substrate of computing?”
Questions I’m holding
1. The Reach of Explanations.
- What does he mean when he says that the reach is intrinsic*
to an explanation?*
- It feels like a subtle point, that the extrapolation is not a detatchable property of an explanation, but is core to the explanation, but it also seems like a vacuous point?
2. Closer to Reality
Surprises
1. The Reach of Explanations.
- Occam’s razor is not what makes an explanation good, it is the lack
of variability.
- The distinction is that Occam’s razor focuses on the (narrative) simplicity of an explanation, whereas variability corresponds to the range of predictions that are constrained by the explanation.
- A complicated function whose outputs have a wide variance is an excellent explanation if the reach of such predictions is wide.
- The reach of a good explanation is exactly the same point as the transferability of theories (versus predictive black-box models).
2. Closer to Reality
Claims ledger
- A good explanation is one that doesn’t vary:
- A good explanation is one with a wider reach:
Chapter reconstructions
1. The Reach of Explanations.
A lot of our knowledge consists of explanations for things that we have not observed. How is that possible? The scientific revolution triggered a search for good explanations. An an explanation is good if it is (1) testable; and, more importantly, (2) less variable. By virtue of a good explanation being less variable, a good explanation has reach, and excellent ones have infinite reach.
2. Closer to Reality