Take a photo of a barcode or cover
whitakk 's review for:
Algorithms to Live by: The Computer Science of Human Decisions
by Tom Griffiths, Brian Christian
I enjoyed this quite a lot -- none of the basic ideas were new to me but I still found the details quite good and generally engaging.
Three things I learned:
1. The "37% rule" for optimal stopping problem relies on a few assumptions -- in particular, it only guarantees optimizing your chance of finding the very best outcome, not necessarily optimizing your average outcome
2. In optimal stopping experiments, people usually stop slightly "too early" compared to the ideal algorithm. But this is perhaps a learned heuristic -- in the real world, there are usually costs to waiting (and/or benefits to deciding early), so the optimal solution is in fact earlier than it is in the laboratory setting.
3. Similarly, in the lab people tend to "over-explore" in explore-exploit problems -- but in the real world such problems may have (approximately) infinite time horizons, whereas in the lab they necessarily have finite horizons, so people's learned behavior may be closer to ideal in the real world.
Three things I learned:
1. The "37% rule" for optimal stopping problem relies on a few assumptions -- in particular, it only guarantees optimizing your chance of finding the very best outcome, not necessarily optimizing your average outcome
2. In optimal stopping experiments, people usually stop slightly "too early" compared to the ideal algorithm. But this is perhaps a learned heuristic -- in the real world, there are usually costs to waiting (and/or benefits to deciding early), so the optimal solution is in fact earlier than it is in the laboratory setting.
3. Similarly, in the lab people tend to "over-explore" in explore-exploit problems -- but in the real world such problems may have (approximately) infinite time horizons, whereas in the lab they necessarily have finite horizons, so people's learned behavior may be closer to ideal in the real world.