Although mining pools smooth out Bitcoin rewards and make it more predictable for an individual miner, they concentrate power to the pool’s operator. There are multiple huge mining pools, and each of them may possess up to 30% of the total computation power of the Bitcoin network (the same applies to some other altcoins). Putting such enormous computation power in the hands of pool operators provides the necessary incentive for them to misuse their power over the network. One way to misuse this power is to launch a block withholding attack against other mining pools. Indeed, this ability starts a block withholding game among the pool operators. Some researchers have analyzed such a game. However, their analyses were limited because they considered simple scenarios, e.g., a single-shot game between only two mining pools. In this paper, we first demonstrate that the block withholding game is a stochastic game with finitely many states and actions. Then, we use a reinforcement learning method to analyze this game. Our simulation results show that in the recent four years, by launching a block withholding attack some pools had the potential to reach to the majority (51%) of the total computation power of the network with much lower initial computation power (even with less than 25% of the total computation power of the network).