HMM models enable to find latent or hidden variables in a timely and efficient manner.

Decrypt hidden messages

Hidden Markov Models allow representing efficiently hidden assumtions. It assumes that the Markov Model underlying the data is hidden or unknown to you. You only know observational data and no information about the states. A true Hidden Markov Model arises when you are not allowed to know what squares a player took to win the game... however, you are allowed to know the sequence of letters that were emitted as they won the game. Yet, the path that the player took to get those letters can't be seen.

We can infer the trajectories and get a feeling of the possible plays. This is essentially what Hidden Markov models do. They assume certain distribution about data and back out the possible scenarios to reconstitue the play assuming certain statistics and infering the required relationship between each play. This is very powerful as it does not provide one single trajectory that obviously would not work as we grow the game and have less memory about the game but the only possible scenarios.