Notes from Algorithms To Live By, Brian Christian and Tom Griffiths.
I enjoyed it a lot, despite my prior familiarity with the content.
My favourite new insight came from the concluding chapter on computational kindness: We can make life easier on ourselves and others by minimizing the computational requirements of our interactions. Being flexible when making group decisions is actually less “nice” than being decisive.
For example, would you like to meet for coffee next week at 2:00 PM on Thursday at the Bridgehead on Bank and Albert? Or would you like to meet for coffee “sometime next week, preferably somewhere downtown – I’m flexible”? Giving people fewer options can make their life easier. If they don’t like your suggestion, they can always push back.
Or maybe this just my new rationalization for being a selfish jerk when making plans with friends…
Computing takes energy and should be avoided – there are many other examples of this theme in Algorithms To Live By. Often when we think of rational decision-making, our mental image is a vast amount of time spent collecting data, testing hypotheses, reflecting on the results. But rationality implies an optimization process for deciding how much time and energy to allocate to each decision.
AlphaZero isn’t the best chess AI because it runs more cycles, or has more lines of code. It’s the best because it makes the best moves and wins games.
Final review: 8.5/10, a great read if you’re already knee-deep in the Rationalist oeuvre, and it if you’re not, it may also introduce some useful new decision-making strategies.