Friday 2014-05-16

Everything is Obvious (once you know the answer) by Duncan Watts

Another rehash of behavioral fails, wherein one learns about one's mental foibles.

Common sense is great for everyday problem solving, but “problems” to do with government planning, policy, business, and marketing are not like everyday problems. Whereas everyday problems involve making decisions about the immediate here and now, decisions about planning, policy, etc., all involve large numbers of people over extended periods of time. Common sense is simply not designed to solve these sorts of problems, but its limitations are rarely apparent to us. The rest of the book will explain both these claims.

The first problem with common sense is that when trying to explain someone’s behavior, or to anticipate it, we focus on certain conscious motives and incentives that are most obviously relevant. In doing so, however, we ignore a multitude of other possibly relevant factors, many of which operate below the level of consciousness. Thus, while it is true that people respond to incentives—somehow—this insight tells us little more than that “people have reasons for what they do.” It doesn’t tell us either what they will do or what reasons they will have for doing it. Once we have observed their behavior, the explanation for it will seem obvious, but this ex-post obviousness is deeply misleading.

The second problem with common sense is that when we try to explain the behavior of groups—crowds, firms, political parties, markets, etc.— we instinctively invoke the language and logic of individual behavior. Yet this sort of explanation ignores a critical component of collective social behavior—namely that it is driven as much by the networks of interactions between individuals as by the attributes of the individuals themselves. Combined with the observation from the previous chapter— that once we observe an individual’s action, we can always reconcile it with something we think we know about human behavior—our expla­ nations of collective social behavior are not explanations at all, but really just descriptions of what happened. For example, conventional explanations of success—like the popularity of the Mona Lisa, the suc­ cess of Facebook or Harry Potter—are circular, saying little more than “X succeeded because it was more like X than anything else.” True, maybe, but also vacuous. The result is that although we always con­ vince ourselves that we understand why certain products or companies or strategies have succeeded or failed in the past, predicting the next hit product or hot company or successful policy is notoriously difficult even for experienced professionals.

A related problem is that when we do try to understand social net­ works, we are intuitively drawn to the idea that social networks—and, by extension, social trends— are dominated by certain influential people. Yet these “influencers” also turn out to be a product of circular reason­ ing. Explanations that invoke influencers, that is, are effectively claiming “X happened because the influencers made it happen and we know they were the influencers because they made it happen.” Again this explana­ tion is possibly true, but again it is vacuous, and of little predictive value. Thus although marketers and the like are good at identifying influencers after the fact, they are unable to reliably identify them in advance, which of course is what matters.

The difference between what we feel we can explain about the past and what we are able to predict about the future seems like it also ought to be obvious. Yet the way we learn from history prevents us from seeing this difference. After the fact, it seems to us as if our predictions would have been right if only we had known something else at the time— something that now seems obvious. But because we are always explaining events after they have happened, at which point we know what it is that we are trying to account for, our explanations dramatically overstate the prob­ ability of the sequence of events that actually took place, thus failing to account for all the outcomes that might have happened but didn’t. Even worse, a deep result of the philosophy of history shows that knowing what is “relevant” in the historian’s sense may not be possible at the time it is happening, even given infinite knowledge of the present and the past, and infinite information processing capabilities. The reason is that what will be deemed relevant in the future depends on events that haven’t yet happened; thus knowing the true importance of events as they are hap­ pening requires more than prediction—it requires a form of prophecy, meaning the ability to observe the present as if looking back on it from the future. And finally, because the future continues to unfold, the nature of the explanation we give for the past may change continually—what seems like a smart move one day may seem dumb the next, depending on other events that haven’t happened yet. For all these reasons, the past is far less deterministic-—and far less informative— than it seems.

Our tendency to see the past as more deterministic, and hence more pre­ dictable, than it really was in turn distorts our perception of the future. Rather than seeing the future as something that is fundamentally proba­ bilistic, meaning that the best we can hope for is to be able to predict probabilities of particular outcomes, we act instead as if the future has in some sense already been determined and simply hasn’t been revealed to us yet. In this way we persuade ourselves that we ought to be able to predict things that we simply can’t predict, even in principle. It is this difference between what we can predict and what we think we ought to be able to predict that causes all the problems with common sense, and also makes it so hard for us to appreciate these problems.

Just because we can’t predict some things, doesn’t mean we can’t predict anything. When the probability of future events is consistent with the frequency of similar events in the past, statistical models based on his­ torical data, and other methods that exploit the wisdom of crowds, can all generate relatively reliable— albeit still probabilistic— predictions. The main mistake to avoid is to trust any one single opinion, even that of an expert (and especially your own), and the main objective is to keep track of the predictions that you make (or that anyone else makes), thereby learning over time which predictions can be made reliably and which can­ not. Almost surely, however, some outcomes—including outcomes such as the next financial crisis or the next disruptive technology—that we would much like to be able to predict cannot be reliably predicted at all. And in these cases, we need to build uncertainty into our strategic plan­ ning, devising strategies that are robust to different versions of the future.

Another response to uncertainty about the future is to avoid relying on predictions altogether and instead adopt a “measure and react” ap­ proach; that is, to become very good at measuring the state of the world in real time and reacting quickly. A variety of methods, such as crowd sourcing, field experiments, and bootstrapping, can be deployed both in business and in policy setting, in some cases to improve performance, and in other cases to avoid disasters.

Moving from business applications to broader questions of social jus­ tice, our overly deterministic view of cause and effect leads us to over­ weight outcomes when trying to evaluate behavior (hence “all’s well that ends well”). In particular, the combination of the Halo Effect (attribut­ ing success to talent) and the Matthew Effect (the rich get richer) system­ atically distorts our perception of merit, often leading us to attribute to individual performance, both good and bad, what is really a conse­ quence of environmental conditions or simply luck. Compounding this problem, the instinct to attribute collective success to the genius of a great leader leads us to overstate the importance of CEOs, which in turn distorts the market for CEO compensation. Finally, the libertarian view that individuals owe nothing to society for their success is fundamentally at odds with the interdependent nature of social and economic systems.

Evaluating the success of the social sciences, by comparing them to the physical sciences is deeply misleading, in part because social scientists historically have lacked the accurate measurements of the physical sci­ ences, and in part because human and social phenomena are inherently messier than physical phenomena. The technological revolution of the Internet, however, may herald a new era in social science—one that is equal in magnitude to that ushered in by the invention of the telescope. The result will still not look like physics, nor should it, but it will lead to new ways of thinking about social problems, and possibly better solu­ tions.