Adaptive social learning for slow Markov chains