On the adaptive control of finite state Markov processes