In Praise of the Coin Flip

The Importance of Being Random

It’s commonplace to tell those facing an agonizing decision to flip a coin, the argument being that the moment the coin lands on one outcome or the other, they will know where their heart truly lies. I believe that is under-appreciating the power and value of the mighty coin flip.

In a world where much of our decisions are guided in some fashion not by necessity but by choice, a greedy decision policy — one that always picks the best choice at every decision point — often leads to suboptimal long-term outcomes, simply because they are highly correlated with each other. This seemingly paradoxical fact is at the heart of the discipline called reinforcement learning, where stochasticity — meaning randomness, is a necessary ingredient for any meaningful learning to take place. With that in mind, it may be excellent practice to build an exploration policy into one’s daily life choices. I make a point once in a while to come out of a bookstore with a book that is definitely not my type, which the move to digital media has made both more difficult — recommendation engines are often much too biased against exploration, and easier — by removing the stigma of having to present to the cashier a book cover that … I’d rather keep to myself. Another practice I apply consistently at restaurants is to always order the weirdest thing on the menu, which has turned out to be, by far, the most successful of these little comfort-zone-expanding nudges I’ve ever managed to apply to my life.

Tossing a coin, once in a while, is one of the more radical exploration strategies. The hardest part is actually deciding whether to toss the coin, and entrust an outcome to fate. It always feels wrong, no matter how benign the decision, to remove one’s conscious choice from the equation. But it’s important to keep in mind the decision to flip that coin is still ours: one main benefit of moving the decision point ahead of the fork in the road is that your responsibility, and guilt, is no longer engaged in it, irrespective of outcome.

When faced with a binary decision, I try to consciously surface the third alternative that’s always implicitly present: that of not deciding at all. It strikes me how that hidden option is often by far the best, or worst, option.

Not deciding, or deferring to decide, is often quite naturally the worst choice to implicitly make. In corporate life, I see much more angst being rooted in indecision, than in teammates disagreeing with any given call. I try to foster a culture of ‘disagree and commit’ as a way to lower the fear of consequences that comes with every decision. As professional gamers know, often the best correlate for success is simply the rate at which one makes decisions. I will come back to that reassurance when I reflect back on a bad call: it could have been worse, I could have not decided at all.

Paradoxically, not deciding is sometimes as important, because who makes the call is often more important than what the call is. And in those cases, that person should not be you. Decisions have side-effects beyond the choice being made: they change the decision-maker, and that’s a more important, empowering outcome than the decision itself.

Randomness gets a bad rap when it becomes part of our collective decisions. It strikes me that perhaps the central aspect of our democracy, more important even than its representative nature, is its inherent randomness. Random outcomes are what keep elected officials both engaged and in check. What would a leader’s incentives be if they knew for certain a long tenure was guaranteed? Conversely, what outcomes would you be chasing if you knew with certainty that your opposition would be taking over soon? I know what my political leanings are, but I would be terrified of what the outcomes may be if my representatives of choice were securely bolted to their seats for the decades ahead.

When I have established that not deciding is by far the worst possible outcome, my willingness to randomize a decision goes up very quickly. It takes a lot of explaining, and on occasion really befuddles and scares the other parties involved in the randomization experiment. But recognizing that not all decisions are equal and willing to relinquish control once in a while is not only liberating, but can be great bait for catching more of these moments of serendipity that could otherwise pass you by.

I am a Principal Scientist at Google, working on Machine Learning and Robotics.