Every time Milo and I teach about how organizations can make sense of their environments, we are confronted with the difficulty of explaining why uncertainty is so different from risk and why understanding that difference matters to entrepreneurs and managers. In this article, we address those questions and discuss the practical implications that flow from them.
Let’s start with the basics. Frank Knight, wrote a seminal book on the issue in 1921: Risk, uncertainty and profit. The book is about much more than just distinguishing uncertainty from risk; it is actually a profound discussion of what profit is. For today, however, we’ll stick to the uncertainty-risk issue. According to Knight, we can make three assumptions about our environment and the distribution of events. First, we can assume that we know the distribution of events theoretically, and hence that we can make predictions based on this knowledge. This is typically the case when we roll dice.
Second, we can assume that the type of event we consider is repeated, and that we can know the distribution of events empirically, i.e. discover it by studying the occurrences over time; this is risk. Risk is found in the insurance industry, which for instance knows the probability for a certain model of car to be stolen in a certain geographic area. Insurance companies can know this quite accurately for the simple reason that it is (unfortunately) an event that occurs frequently. As a result, insurers have built historical tables of the event on which they calculate probabilities, and these tables are fairly stable over time. Hence, risk in this instance is concerned with the occurrence of events that belong to a known category.
Third, we can assume that the type of event we are considering is unique, and therefore that the corresponding distribution is not only unknown, but objectively unknowable, not even in theory. This is uncertainty. Uncertainty corresponds to situations where events are new and therefore no information on them exists, and therefore no history is available to assess probabilities. The transistor can only be discovered and commercially introduced once, and there is only one Vietnam War. Analogies are always possible, but in the absence of objective information, calculation is impossible. We must distinguish unique events, where uncertainty is high, from rare events where it is usually reduced. An event can be rare and perfectly certain: consider Olympic games or solar eclipses. Note that uncertainty comes from an objective lack of information – the information does not exist- not from a subjective source, whereby the problem would be the inability to access certain information. Hence, uncertainty is concerned with events that do not belong to a known category, even if they can be close to analogous situations. For instance, in many ways the emergence of the Internet is similar to the emergence of electric power; but there are some profound differences between these two phenomena, and that is where uncertainty originates. Note too that though newness is our focus here, it is not the sole source of uncertainty.
Where do black swans fit in the picture? Nassim Taleb defines a Black Swan as an event of low probability but high impact. In other words, a black swan does not create a new category of events, but is simply the occurrence of a known category, the probability of which was underestimated. Black swans therefore belong to the field of risk. They occur not because their probability is inherently incalculable, but because the model used to calculate them is wrong, or because though the model was correct, the possibility of occurrence was dismissed in practice (psychological bias). This is what happened in the world of finance: the theoretical possibility of a crash was well known since at least Kindleberger, but both its probability and its impact were greatly underestimated. Again, a black swan results from a miscalculated risk.
Where this rather basic discussion gets interesting is when one considers empirical probabilities more deeply. Take the example of the car theft mentioned above. We said that because the phenomenon is not new and occurs quite often, insurers are able to estimate the probability of your car -say, a Toyota Yaris- being stolen fairly accurately. This is thanks to the historical tables they use to calculate their rates. But this calculation rests on the fundamental but often unstated assumption that nothing will fundamentally change the number of stolen Yaris. Now imagine a special event: a famous rap artist records a hit with Yaris in the title. The car, which until now was relatively banal, gets instant worldwide visibility and its appeal among youths skyrockets. Suddenly, Yaris thefts rise rapidly, too. The distribution law empirically observed for many years by the insurers is no longer valid, and their models become a source of losses to the insurer. This illustrates the more general problem of induction that Taleb describes using the famous example of the Christmas turkey: the turkey is fed for many days and grows increasingly more confident that the farmer cares for her until, just before Christmas, it is killed by him. Months of empirically derived observations by the turkey about the benign nature of humans prove useless on that day. Put otherwise, just because something has been true for a certain period of time doesn’t mean it will remain true forever. This is so because what has been observed is only empirically observed; the inherent distribution law, if there is any, is not known and mostly cannot be known. Hence, any model based on empirically established statistics is only true until it becomes false; it is a victim of a black swan event in waiting, so to speak. Just ask a US homeowner.
The Black Swan is an essential concept for understanding how we make mistakes in estimating the probabilities of different events belonging to a known universe. These errors are mainly due to bias, either psychological or methodological. However the Black Swan tells us nothing about what to do in uncertainty which characterizes an emerging environment, in which events are too new to be categorized, and therefore enumerated. For example, will the car of the future be electric? Will humans accept genetic modification to learn foreign languages more easily? Will China experience a financial collapse? These events, or clusters of events, are so complex and contain so many variables that are indeterminate at the point of decision. They are uncertain, and, uncertainty is a phenomenon of its own. They are not about risk, just a bit more complicated. They are fundamentally different, and must be approached as such. But that won’t stop risk “experts” from assigning a single numerical score to them – humans have always needed fortune tellers!
The question that remains, once the theoretical distinctions among prediction, risk and uncertainty is established, is knowing whether these ideas matter in practice. To answer that question we need to consider the proportion of the environment characterized respectively by prediction, risk and uncertainty. Clearly, predictable human environments are extremely rare. Statistics manuals might be filled with examples of dices rolling and cards being drawn, but reality is far more complex. Risk is more common, but the example of car theft suggests that much of what is seen as risk might be sensitive to black swans. In fact, behind seemingly benign environments often hide uncertainty. It is especially here that the danger lies, because of years of accumulation of information gives us a false sense of understanding of the environment, just like the turkey.
In the end, there seems to be really very few human situations of calculable risk in theory or practice. Hence, while in theory, risk and uncertainty are fundamentally different, in practice real risk is actually rather rare, and it is uncertainty that is abundant. Risk is a special case, uncertainty is the norm. Uncertainty that should therefore be the main focus of managers and entrepreneur’s attention…
More on Taleb’s work here: Welcome to Extremistan!