Begin with the End in Mind

“Begin with the end in mind.” This is Stephen R. Covey’s second habit in his wildly successful book, The Seven Habits of Highly Successful People. He begins with a dramatic exercise in imagining your own death — the end of your life that you should keep in mind. In the everchanging and unpredictable world of the 21st century, this may seem like outdated advice. Long project plans are out, and agile is in. Should we still begin with the end in mind?

Even in the computer age, we can benefit from keeping the end in mind. I am learning this more and more as I teach computers to play common board and card games. The computers themselves benefit when they begin with the end in mind.

In multiplayer games, players take their turn and then give up control to another player. This creates uncertainty for the player. The player cannot know which move her opponent will choose. Chess Grand Masters must anticipate every single possible move their opponent will make and think several moves ahead. Computers can do this much better than humans and therefore can easily beat most human players in these types of games.

At first glance, a player may decide to make the move that is most likely to result in a win, assuming an equal chance of the opponent making all available moves. Let us briefly consider this methodology. Consider the decision tree below. Green nodes are your moves and red nodes are that of your opponent. Wins are W and losses are L.

As you can see, moves A and B can both result in either 2 wins or 2 losses. Using the methodology described above, both move A and B are equally strong moves. The wise game-player, would choose move A, however. When the green player makes move A, then she can win, regardless of whether the red player makes move A1 or A2. If the green player makes move B, however, the red player can force a loss by playing B2.

When teaching a machine to play this game, the machine would learn to play the moves that are more likely to result in a win. Initially, the machine would make an approximately equal decision between the two moves. As both the red and green players are learning, the red player would start to make move B1 more often and the green player would start making the final win moves more often as well. This would cause green to learn to make move A rather than move B. The machine will learn slowly though, retaining a bias from the earlier games in which red made move B2.

Can this method be improved? Yes! The machine just has to begin with the end in mind! The learning algorithm creates a larger weight for moves made at the end that result in a win. Since the final green nodes both have a move that guarantees a win, these decisions should create a greater weight towards the algorithm picking A. One could, for example, count a winning move at the end as two wins. If you end up in this position again, you have a guaranteed win.

Real life has many similarities to these games. You make your moves and control what you can, but there are many factors beyond your control. If you weigh your final moves more heavily and position yourself to strike and win at the end, you will increase your chances at success. Even in the information age, we should begin with the end in mind.