When the desired behavior is performed, a sound is played. When the test subjects reach deep sleep, that same sound is played repeatedly. Subjects were then more likely to perform the desired behavior.
The article, "Unlearning implicit social biases during sleep" appears in the journal Science; an abstract and full report are available.
(Score: 2) by TGV on Saturday May 30 2015, @06:49AM
I'll repeat what I wrote elsewhere: If the effect (which is, as the author even admits, surprising) will hold, it changes "implicit bias", i.e. what the implicit bias test measures. That is much less surprising, since that test measures priming effects, i.e. how much does condition A speed up recognition of condition B, rather than attitude. So it doesn't change anything fundamental, but rather weakens the speeding up of the priming process, possibly by adding interference.