Thinking Straight, Go to a Simple Case

THINKING STRAIGHT, GO TO A SIMPLE CASE

Most American readers are familiar with the acronym KISS (Keep It Simple, Stupid), and some readers might be familiar with Occam's razor.[9] Einstein famously said: "Everything should be made as simple as possible, but no simpler." While many ideas that Richard discusses in his academic work can be complex or subtle, he has a knack for simplicity. He learned this from mentors, whom he credits for teaching him how to think.

He often tells the story of a moment early in his career when he met with Howard Raiffa. Howard had introduced Richard to the principles of Bayesian decision analysis, such as thinking in terms of probabilities and defining your preferences over uncertain outcomes. Richard entered Howard's office and proudly said, "Howard, following your instructions, I simplified. I solved the challenging group-decision problem we discussed assuming that there were just three types of agents." "That is excellent," Howard replied. "You have made good progress. Now, go back and try it with two." Howard Raiffa's passion for the greatest feasible simplicity enabled him to pioneer the fields of game theory and decision analysis, and play a founding role for the field of negotiation analysis.[10]

Richard tells about how two of his other major mentors, Thomas Schelling and Kenneth Arrow (both Nobel Prize winners), sought simplicity. Schelling, for example, regularly employed everyday and easily grasped examples, such as a parent negotiating with a child, as an analogy to a much more complex situation, such as an international negotiation. A Schelling lecture often moved from a simple example, say a stop sign, to a riff on a dozen critical issues that involved a metaphorical stop sign. One element of Arrow's genius was to see, and then distill, simple principles, where others saw only murky complexities.[11]

Richard provides his own students with disarmingly simple advice when they are faced with a decision or dilemma. Ingrid Ellen, one of Richard's former students, and now a professor at New York University, recalls: "'I came to Richard for advice when I was deciding whether to pursue a PhD in economics or public policy. I was agonizing. He responded with a question: he asked me whether I was more excited to answer narrow questions very precisely or to study the social and economic policy questions I cared most about but to answer them with somewhat less precision. If I wanted the former, I should pursue a PhD in economics; if the latter, I should pursue a PhD in public policy. The choice was easy. As always, he took a difficult decision and made it very simple. And I have never had any regrets."

Benjamin Van Roy, a Stanford professor who met Richard when Benjamin was at MIT, uses the simplicity maxim in the field of reinforcement learning. That field is known for impressive feats carried out by complex systems of artificial neural networks that learn through trial and error using massive computational resources. A well-publicized example is a computer program that learns to play just about any game – for example, chess or go or arcade games – better than human world champions.[12] With all the complexity in such games, it can be difficult to understand what works and why, and how to improve things.

He says, "I advocate Richard's maxims on clarifying thought by going to simple and extreme cases. I don't think of the two as distinct. To design an algorithm that demonstrates a sophisticated skill, it is helpful to formulate a task that's simple enough for a human to easily solve while pushing to the extreme the level of skill required of a general algorithmic solution not specialized to the task. Not everyone thinks this way; some find it strange to work on such simple problems when the field is already able to address such complex ones. But I've found this to be a fruitful way to develop ideas that can then be ported to complex domains. With these maxims in mind, my teaching tends to focus on such simple and extreme cases."

He adds, "There is evidence that this spirit has carried on. For example, there is now open-source software for simulating a variety of simple and extreme reinforcement learning tasks. This development was led by Ian Osband, a former student of mine, and serves the research community as a platform for reinforcement learning algorithm design."

John Horton, a faculty member at the MIT Sloan School of Management and a former student and coauthor of Richard's, used this simplicity maxim to tackle a question he was trying to understand in the ride-sharing market (Uber, Lyft, etc.): From a societal perspective, should drivers know where a passenger is going when they decide whether to take a trip? On the one hand, knowing that information allows drivers to accept the "good" trips and not the "bad" ones. If many drivers do this, the wait-time for "bad" trips could be very long or, even worse, passengers needing "bad" trips might never get picked up. On the other hand, if drivers differ in their costs (for example, it is cheaper to pick up a passenger going your way as your day is ending), letting them pick and choose among trips can be economically efficient; the trip will go to a driver for whom the trip is more beneficial. Creating a model that fully captures all such relevant parameters can get very complicated very fast.

John simplified the problem by assuming that there are just two kinds of trips, "good" and "bad," and the two types are equally likely. He also assumed that drivers only have two possible strategies: 1) accept all trips, or 2) accept "good" trips and reject "bad" ones. By simplifying the problem (two kinds of trips, two strategies only), he gained insights he would not be able to get otherwise.[13]

"Once I've got a foothold on the problem, I can go much farther. If I do need to make it more complex or general, I'm more likely to understand the implications of what this added complexity does. This is just an example, but it's entirely representative of how I work. I keep simplifying and interrogating the implications until I've earned some intuition for how the problem "works." It's an extremely useful heuristic for gaining understanding of a world that's often overwhelmingly complicated."

Some readers with mathematical inclinations might interpret this maxim as a way of avoiding hard thinking. But as Chris Avery, Richard's colleague at the Harvard Kennedy School and coauthor of the book The Early Admissions Game, reveals, "Before I met Richard, I thought that it was somehow cowardly – taking the easy way out – to simplify a complicated calculation to gain traction in this way. I learned from Richard that astute choice of technique is always preferable to virtuosic but misguided computation."[14]

MAXIM 3