The application of Markov state probabilities in developing artificially intelligent managerial strategies: A case study based on major league baseball

Baseball is a very large industry that influences the lives and financial assets of millions of people.

Many baseball analysts estimate that even the best Major League player can increase his team's expected number of yearly wins by only 3 or 4 games. In 1993 Bobby Bonilla, a New York Mets National League outfielder, earned $5,200,000 more than the average Major League player earned. The Mets apparently believed that expenditure of this money was worth his expected additional victories. If the strategy of an Artificially Intelligent baseball manager could produce 4 more team victories than the corresponding human baseball manager, would this artificial expert also be worth \$5,200,000 per year?

Baseball, like many other processes, transitions from state to state until it terminates in either a win or a loss state. Each intermediate state has an associated probability of win, PW. For instance, the home team's PW might be 0.350 for the state--losing by two runs during the bottom of the 3rd inning, with one out and a runner on 1st base.

Human managers frequently have no concept of a given state's proper PW. But this paper develops a computer simulation model that estimates the PW for each of the 9072 states that comprise this Major League baseball process. Consequently, it devises representative strategies to artificially steer the process into a winning terminal state. Since the specific path to any given state is irrelevant, the sport of baseball is defined as a Markov process.

This paper is more than a treatise about baseball. It employs the relatively concise and clear baseball process as a guide to help any industry develop successful processes by artificially selecting the next discretionary Markov State.