Funny story about perverse incentives (or "unaligned natural intelligence" for the AI policy geeks).
1) Dog rescues toddler who had fallen into local river, gets rewarded with a nice steak from grateful father.
2) Soon, dog rescues an unusual number of toddlers regularly from similar near-drownings on nearly daily basis -- each time rewarded with a steak.
3) Eventually local residents figure out that the dog had been pushing the kids into the water, so as to be able to rescue them afterwards (and get the reward).
4) FWIW, I don't fault the dog.
(Source: NYT, Feb 2, 1908, "Dog A Fake Hero", via Zvi Mowshowitz.)
