Rewarding Criticism Put Nicely Produced Long-Lasting Change

Eliezer Yudkowsky, I’m told, used to be a not-nice critic. The problem was his delivery: “blunt, harsh, not sufficiently tempered by praise for praiseworthy things” (Alicorn Finley). However, this changed about a year ago, when Anna Salamon and Alicorn Finley decided to try to train him to be nicer. Alicorn describes it like this:

Me, Eliezer, Anna, and Michael Blume were all sitting in my and Michael’s room (where we lived two houses ago) working on, I think it was, a rationality kata [= way of doing things], and we were producing examples and critiquing each other.  Eliezer sometimes critiqued in a motivation-draining way, so we started offering him M&Ms when he put things more nicely.  (We also claimed M&Ms when we accomplished small increments of what we were working on.)

Eliezer added:

Some updates on that story. M&M’s didn’t work when I tried to reward myself with them later, and I suspect several key points:

1)  The smiles/approval from the (highly respected) friends feeding me the M&Ms probably counted for more than the taste sensation.

2)  Being overweight, M&Ms on their own would be associated with shame/guilt/horror/wishing I never had to eat again etc.

3)  Others have also reported food rewards not working.  One person says that food rewards worked for them after they ensured that they were hungry and could only eat via food rewards.

4)  I suspect that the basic reinforcement pattern will only work for me if I reward above-average performance or improvement in performance (positive slope) rather than trying to reward constant performance, because only this makes me feel that the reward is really ‘deserved’.


  • Andrew Critch advises that ‘step zero’ in this process is to make sure that you have good internal agreement on wanting the change before rewarding movements in the direction of the change
  • The Center for Applied Rationality (CFAR) has some experience learning to teach this.
  • CFAR has excellent workshops but not much published/online material.  A good mainstream book is Don’t Shoot the Dog by Karen Pryor.

I like this example because the change was long-lasting and important.

2 Replies to “Rewarding Criticism Put Nicely Produced Long-Lasting Change”

  1. See also:

    What Shamu Taught Me About a Happy Marriage

    The author uses animal-training techniques to address her husband’s annoying behaviors.

    The central lesson I learned from exotic animal trainers is that I should reward behavior I like and ignore behavior I don’t. After all, you don’t get a sea lion to balance a ball on the end of its nose by nagging. The same goes for the American husband.

    Back in Maine, I began thanking Scott if he threw one dirty shirt into the hamper. If he threw in two, I’d kiss him. Meanwhile, I would step over any soiled clothes on the floor without one sharp word, though I did sometimes kick them under the bed. But as he basked in my appreciation, the piles became smaller.

    I was using what trainers call “approximations,” rewarding the small steps toward learning a whole new behavior. You can’t expect a baboon to learn to flip on command in one session, just as you can’t expect an American husband to begin regularly picking up his dirty socks by praising him once for picking up a single sock. With the baboon you first reward a hop, then a bigger hop, then an even bigger hop. With Scott the husband, I began to praise every small act every time: if he drove just a mile an hour slower, tossed one pair of shorts into the hamper, or was on time for anything.

  2. Alex,

    I like that example. Presumably this would be less effective if the person being trained had no affection for the trainer. But even more I wonder how well it would have worked if the trainer (wife) did not love the trainee (husband).


    I came across your blog about a month ago and somehow ended up reading through most of it. I believe I learned quite a bit. To me, most interesting topics I’ve read about here are: the Shangri-La “Diet” (ahead of its time), development of skill in context of human evolution, Jane Jacobs as a thinker on all sorts of things, self-experimentation as part of a new literacy, most of your thoughts on statistics (e.g., against uncritical “Correlation is not causation”). There’s more, I’m sure. Thanks for sharing.

Comments are closed.