Deliberative Cognitive Algorithms as Scaffolding — LessWrong