This post helped nudge me toward Feedbackloop Rationality, but the stated solution feels kinda cheating – it wouldn’t have occurred to me that doing a complete run-to-the-end-of-the-ramp was allowed (it seems like the rules explicitly imply they can’t?).
I haven’t actually done this exercise, nor used it in my workshops. I think it’s probably a mistake that I haven’t actually tried it at all. As I sit and think about it now, it doesn’t seem that hard to patch the exercise so it doesn’t feel like cheating, and a fixed version seems worth trying at Cognitive Bootcamp.
(I once asked John if he thought the exercise was possible without doing a complete “roll the ball to the end of the ramp”, and he said “I dunno that does seem way harder.”)
It seems like the current implementation of the exercise requires you to figure out you can do a single complete run, and then taking advantage of that. If I were try and fix the instructions, I’d want to make that feel “allowed” without “directly drawing their attention to it.
Some options:
State “some approaches of rolling the ball down are cheating, some are not. You are encouraged to think of options, and ask if they are cheating.”
State “you only get one run from top-to-bottom-of-ramp, but as many partial runs as you want.” I think this will make it too easy for some people, but many still probably won’t think of it.”
State “you get more points the smaller-percentage-of-the-track you use for your tests, and you get 0 points if you don’t get it into the cup after 1 hour.” This changes a vague feeling of dishonor into a rule to measure and evaluate.
Actually figure out if there’s a tractable way of solving it without doing a full-ramp-test. idk.
I would generally either be clear cut about the rules, or clear cut at the meta-level about how to ask what counts as cheating (and be clear internally about what counts).
This post helped nudge me toward Feedbackloop Rationality, but the stated solution feels kinda cheating – it wouldn’t have occurred to me that doing a complete run-to-the-end-of-the-ramp was allowed (it seems like the rules explicitly imply they can’t?).
I haven’t actually done this exercise, nor used it in my workshops. I think it’s probably a mistake that I haven’t actually tried it at all. As I sit and think about it now, it doesn’t seem that hard to patch the exercise so it doesn’t feel like cheating, and a fixed version seems worth trying at Cognitive Bootcamp.
(I once asked John if he thought the exercise was possible without doing a complete “roll the ball to the end of the ramp”, and he said “I dunno that does seem way harder.”)
It seems like the current implementation of the exercise requires you to figure out you can do a single complete run, and then taking advantage of that. If I were try and fix the instructions, I’d want to make that feel “allowed” without “directly drawing their attention to it.
Some options:
State “some approaches of rolling the ball down are cheating, some are not. You are encouraged to think of options, and ask if they are cheating.”
State “you only get one run from top-to-bottom-of-ramp, but as many partial runs as you want.” I think this will make it too easy for some people, but many still probably won’t think of it.”
State “you get more points the smaller-percentage-of-the-track you use for your tests, and you get 0 points if you don’t get it into the cup after 1 hour.” This changes a vague feeling of dishonor into a rule to measure and evaluate.
Actually figure out if there’s a tractable way of solving it without doing a full-ramp-test. idk.
I would generally either be clear cut about the rules, or clear cut at the meta-level about how to ask what counts as cheating (and be clear internally about what counts).