I was talking to a good friend recently, telling him all about Beeminder and TaskRatchet, and he came back with an idea that I had never thought about. What would your reaction be to a Beeminder or TaskRatchet-like system that, instead of telling you exactly what you would be charged if you derailed, charged you a random amount each time you derailed?
Basically you’d set three variables:
Target average charge
The system would then charge you at random between the minimum and maximum, weighting its random number generator to converge on the target average over time. But the RNG wouldn’t have a memory, so any charge could be the maximum charge or the minimum charge.
If I were going to design experiments around this, I’d definitely have:
actual bounded random
bounded random, but each random selection would set the new minimum. i.e. you’d never get a value chosen that was lower than a previously chosen one
something that scaled “your behavior” to the “punishment”. For me, I’d have it set so that if I had been slowly eating my buffer, I’d have a worse “punishment” than if I were doing just fine for a long time and then had an emergency, or even building buffer and then derailing in a single day.