Science Corner: Artificial Intelligence Deploys Eeny, Meeny, Miny, Moe

My son finds decision-making challenging at times (OK, who doesn’t?) and eeny, meeny, miny, moe helps him break his internal deadlock. Actually, it did until he realized it was deterministic and he could decide the outcome by choosing where to start. So now he waves his hand around while he sings a song for a while, then opts for whatever he’s pointing at when the song ends. I’m not sure it’s actually a random process, but it seems like it must be closer than the original version. Plus he finds it satisfying which is valuable by itself. And based on a recent study of the value of randomness in group tasks, my son just might be onto something.

The study involved humans playing a cooperative game to color nodes in a graph such that no two adjacent nodes were the same color. If that sounds abstract, the objective is comparable to coloring a map of the United States such that no two adjacent states have the same color, which is how maps are typically colored. Each player was in charge of a “state”, but they could only see its color and the color of neighboring states. The game is always winnable, but the fact that no one can see the whole picture makes it challenging. Sometimes, two neighbors might wind up with the same color and believe the other one should change because they can’t see all of the other state’s neighbors’ colors.

What the groups didn’t realize is that some of the states were controlled by artificial intelligence. That AI would usually pick a color intended to contribute to the solution, but sometimes it would just pick a random color. A little randomness turned out to be a boon to the group; those groups found solutions faster than groups with all humans or groups where the AI didn’t make random choices. It was possible for the AI to make too many random choices also and slow things down again; 10% of choices made at random was the best arrangement. In some cases, the random choices helped to break deadlock directly; in other cases, seeing another player change color might influence one or more human players to change more frequently.

The concept of randomness inspires strong feelings. Some are uncomfortable with quantum indeterminacy or the role of random mutations in evolutionary natural history; they seem to suggest a lack of purpose. Others embrace those random elements as an escape from a purely deterministic reality. And that’s before we get into questions of whether any phenomena are random in the sense of being genuinely undetermined or whether they are random simply in the sense that we don’t observe all of the causes which determine the outcome. While this isn’t the first result to demonstrate a positive purpose for randomness, it’s always interesting to see where it will prove useful and how. The human psychological element of this study is also intriguing for revealing how we are influenced by each other.

I’d be curious to see if this effect can be further isolated to identify when a little randomness might be most useful. Presumably there are times when the correct move can be discerned from available information and so a random choice would be counterproductive. And perhaps when the options all appear equivalent is when a random choice is most helpful. Knowing which is which might make these results more readily applicable to scenarios beyond games.

Does randomness factor into your work or your field? How do you feel about the potential role of randomness in the way the world works? Do you ever find yourself ‘flipping a coin’ to break an impasse in decision-making, either internally or with colleagues?