1The main point of Kay and King’s book is that, by neglecting radical uncertainty, economists and policymakers have run into a twofold intellectual failure: i. they misunderstand human rationality and ii. delusively tend to believe that good public policy should be exclusively based on forecasting. Their goal is to show why this is wrong and how this happened. As they indicate in the introduction, their argument is based on three “propositions.” Proposition 1 is that social systems are not governed by non-stationary laws. Proposition 2 is that humans do not optimize, and this is rational. Proposition 3 is that humans are social animals, and communication plays a crucial role in decision-making. The book is compounded of twenty-three chapters that the authors organize in five parts of unequal size.
2The first part (50 pages) introduces the concept of radical uncertainty and shows how pervading it is. Their argument is fairly standard: whether they be individuals, firms or institutions, decision-makers rarely face well-defined problems (“puzzles”) which can, as in a “small world”, be exhaustively understood and optimally addressed. We live in a “large world” full of ill-defined, ambiguous, or vague issues (“mysteries”), for which optimal solutions—when they exist—are far too demanding for our imagination to be found. Radical uncertainty is a situation of that kind. It is not so much about a lack of information, a topic that has spawned tremendous literature in economic theory. Rather, it arises from the impossibility to conceive clearly and exhaustively all the events that could occur. Moreover, the authors see radical uncertainty as a pervading phenomenon because, given their first proposition, the worlds of economics, business, and finance are governed by unstable laws so that decision-makers usually have to cope with problems they have not encountered in the past.
3With an impressive collection of wide-ranging examples, this section successfully demonstrates how essential radical uncertainty is (or should be) to cope with socio-economic phenomena. This makes the contribution vivid and thought-provoking. Using Merton’s idea that social systems are reflexive—social systems are influenced by our beliefs about them—the authors also justify adequately their non-stationarity proposition. However, they borrow most of their arguments from management sciences and completely ignore what has been proposed by economists to deal with radical uncertainty. This is an important drawback of the book as a whole.
4To motivate their view, they mostly rely on the supposed absence of interest among economists—specially decision theorists—regarding this issue. Yet, contrary to what they suggest, many well-established economists have worked on radical uncertainty after Keynes and Knight. Kreps (1979) and Kreps (1992) clearly paved the way for a revealed preference analysis of the effect of unforeseen contingencies (see also Dekel et al., 2001). Following Ellsberg’s paradox, the literature on ambiguity has clearly contributed to analysing the limit of probabilities (Gilboa and Schmeidler, 1989; Klibanoff et al., 2005). A small but growing literature at the interplay between economics, computer sciences, and philosophy focuses on unawareness (see Schipper, 2012 for a bibliography), and awareness of unawareness (Karni and Vierø, 2013); that is, the incapacity for a decision-maker to imagine something (action, consequences, object, or state of the world) and the awareness of the decision maker that she might not be able to consider some states of the world. By omitting these works, the authors miss many points that would have helped clarify their own conceptualization of radical uncertainty.
5The second part (80 pages) goes back to the early development of decision theory and proposes a historical account of the neglect of radical uncertainty. According to the authors, the main historical driver of this neglect is the impressive emergence of probability theory in the early 20th century. At this period, the concept of probability stopped being a simple tool to deal with well-defined problems—based on a frequentist interpretation of probabilities—and became the main ingredient of individual decision-making under uncertainty. No matter that in many situations no frequency is available, that probabilities have been ignored for centuries, or that, according to Savage himself, subjective probabilities might not be well suited when the whole state space cannot be apprehended. The delusion that mathematical reasoning is more rigorous than any other approach made it tempting to reduce uncertainty to subjective probabilities. To embody this historical turn, Kay and King aggregate the standpoints of quite different authors such as Savage (1972), Ramsey (2013), de Finetti (1937), and Friedman (1976). In their words, these authors would have “won” what they qualify as a “forgotten dispute” against Keynes and Knight who advocated for a non-probabilistic approach to radical uncertainty. Kay and King, meanwhile, follow the latter approach. They argue that probabilistic models might sometimes be themselves ambiguous (e.g. the “two-child problem”) or that political decisions can be better addressed without probabilistic assessments (e.g. Obama deciding to kill Bin Laden). Hence, they emphasize that the neglect of radical uncertainty has political consequences that, altogether, explain why basing public policy on data alone is an intellectual failure.
6This part of the book is interesting for its broad history of probabilities, but their discussion of the so-called “forgotten dispute” would deserve much more development for at least three reasons.
7First, the originality of their argument is rather overstated insofar as subjective probabilities and their limits are still widely discussed both by mainstream and non-mainstream economists. Second, they promote Keynes's and Knight's contribution, but neither discuss what these authors have proposed, nor do they distinguish their stand-point on uncertainty. Had they dedicated more comments to Keynes’s Treatise on probability, they might have concluded that his analysis of imprecise probabilities in chapter XV is very close to the modern concept of ambiguity. As a matter of fact, in this chapter, Keynes argues that decision-makers may use intervals of probabilities to make their decision, a point that clearly echoes the multiple prior assumptions in this very large and established literature. Moreover, Keynes and Knight had essential methodological disagreements. While Keynes explicitly criticized subjective probabilities, it is not clear that Knight’s contribution was incompatible with the assumption that agents act as if they had subjective probabilities. For instance, LeRoy and Singell Jr (1987) argue that this is not the case. While the authors quote this last reference, I don’t see any attempt in the book to provide historical arguments showing that this is not the case.
8Third, their prospect on this debate between promoters of subjective probabilities and Keynes is, I believe, a little bit narrow. The authors reduce the subjective interpretation of probabilities to Friedman “as if” argument in favor of optimization or to Ramsey’s (2013) Dutch book argument. In the first place, the goal of Ramsey’s, de Finetti’s, and Savage’s subjective approach was to clarify the interpretation of probabilities. In contrast to the frequentist and the Keynesian (or “logical”) interpretation, their starting point was that probabilities are nothing more than a subjective state, and their main goal consisted in understanding the axiomatic conditions under which these mental states could be recovered by choice data. While they refer to the subjectivist point of view as an axiomatic interpretation of rationality, Kay and King do not discuss what axioms are made for (i.e. making a model falsifiable and identifiable). By omitting the methodological advantages of this tradition, they completely ignore the potential weaknesses of their unstructured conception of radical uncertainty. Can we identify when radical uncertainty is at work? What are its different forms? What can be said about its consequences? These questions would have been relevant to raise.
9In the third part (180 pages), Kay and King develop their own account of how individuals make sense of radical uncertainty, introducing their concept of reference narrative and their evolutionary interpretation of rationality. To make their point they provide a profusion of ideas and examples that can be articulated by using their Proposition 2 and 3 mentioned in the first paragraph. Proposition 2 states that deviations from optimization or Bayesian rationality should not be interpreted as irrational. According to them, rationality in decision theory is a small world reasoning, in which the agent faces a solvable situation. Yet, it is irrational to conceive the world as small when it is large; when it comes to large world problems, computing expected utility is not only impossible, it is a “lure.” For that reason, Kay and King insist on the role of evolution to cope with radical uncertainty. Nature has provided us with an “evolutionary rationality”: through biological and cultural evolution we have developed psychological mechanisms (heuristics, emotions, …) sufficiently efficient and flexible to cope with complex social situations, though they might entail pathological behaviors in specific contexts. Among these mechanisms is our ability to build narratives to account for “what is going on there”—a key sentence of the book, borrowed from Dick Remelt. According to Kay and King, narratives express our realistic expectations in large world contexts. Specifically, a reference narrative is a group of stories that help individuals understand what is at work in the world, and what they can expect from it. As for beliefs in economic theory, a reference narrative can be changed in response to disconfirming evidence. But contrary to Bayesian updating, these changes operate rarely and discontinuously.
10Proposition 3 states that humans are social animals and communication is a pivotal feature of decision-making. Since a community in which cooperation is, to some extent, more efficient, natural selection has made us inclined to cooperate. Here the authors refer to the evolutionary account of cultural dynamics proposed by Spencer, Dawkin, or John Maynard Smith. According to them, because of its loose connection to nazis or white supremacist eugenics, this interpretation has encountered fierce resistance. Yet, for Kay and King, this paradigm provides powerful explanations for the emergence and the persistence of a wide-ranging variety of inclusive institutions—in contradiction with a far-rightist vision of human nature. Among these institutions is language, which allows us to communicate, as well as to construct our thoughts, suggesting that communication and rational deliberation have common grounds. Humans' reasoning skills arise, before anything else, from the necessity to communicate, and intelligence is a collective phenomenon. This explains why narratives are efficient. Narratives are stories we use to convince humans, including ourselves, that the complexity of the world can be regimented in a simple and vivid depiction. In that sense, narratives are useful, but they can also be a trap when they are narrowly interpreted as truth. The same goes for models. For that reason, policymakers should not restrict themselves to the narrow view of a single ex- pert. Several experts will come up with different stories to account for the same social phenomena and there is no clear-cut way either to favor one of them or to aggregate their opinions. Be they based on probabilistic (or mathematical) reasoning or not, many of these stories can provide a partial model of “what is going on there.” More generally, the collective aspect of intelligence implies that policymakers should not always conform to the experts’ views because the ability to depict or explain a phenomenon may differ from the ability to choose a well-suited action. A weaker way to state these two propositions might be that cognitive re- sources provided by culture should not be interpreted as what makes individuals deviate from it; rather, they are what constitutes human rationality. This is a deep and interesting point. Yet, I am not very convinced when the authors argue that optimization contradicts this point. First, it seems that they interpret optimization in a narrow sense, i.e. the maximization of an expected utility. But the concept of optimization is much more general than that. Optimization simply requires that the decision-maker use criteria to rank some mental or physical entities. For instance, the authors suggest that narratives be powerful drivers of decision-making. But not all narratives are, and some are more efficient than others. This is precisely what explains cultural evolution: some prevailing narratives are replaced by more convenient ones. This is also why, since Aristotle’s Poetics, many books have been written to explain how to tell a good story. Why not use optimization to describe how people select what counts as a good narrative? At the end of the day, optimization simply implies the existence of an ongoing selection process that guides the decision-makers in her decision. Of course, one might argue that optimization is not necessary. But I am not sure to understand how we can think of individuals guided by a fully external selection process as rational. In any case, I do not see any passage in the book in which the authors seriously raise this issue. The point I have just made is confirmed by the fact many economists (including Akerlof and Snower, 2016; Loewenstein in Chater and Loewenstein, 2016; Bénabou et al., 2018; Eliaz and Spiegler, 2020) have already discussed the notion of narrative in a context in which agents are optimizing. Unfortunately, here again, the authors do not quote them, they only mention Schiller who sees this pattern as a rationality failure.
11The fourth part (79 pages) is dedicated to applying the conceptual framework introduced in the third part to discuss many well-established economic models (in finance, insurance, or macroeconomics) and their connection with radical uncertainty. Their main claim is that these models are powerful narratives that partially capture important real-world mechanisms. While it is worth knowing them, they should not be taken too literally, for their key concepts (such as the efficient market hypothesis in finance or rational expectations in macroeconomics) rely on a misled conception of risk and uncertainty. Not only do they ignore radical uncertainty, but they require that the meaning of risk be shared by every agent. Yet, Kay and King claim that many economists, indoctrinated by their economic education, take these models too seriously. They argue that the consequences are twofold. First, by wrongly pretending to be able to forecast the future, these economists create a public demand for deluded prediction. Second, by focusing on prediction, they blur the main issues, i.e. understanding what is going on there. I think the authors give a fair account of these models, their advantages, and their limitations. Yet, in the light of their critical and partial account of rational choice under uncertainty theory in part II, it is surprising that they spent so many pages discussing these models in such a benevolent way.
12The fifth part (44 pages) is an attempt to draw what the economists should do and how they should see themselves. According to them, the role of economists is to produce practical knowledge to help policymakers, public servants, and, more generally, any citizen understand what is going on there. They should also adopt a reflexive attitude towards this knowledge. This means that they should collect data, obviously, but they also need to wonder how these data are collected. They should use models, of course, but they must see these models as tools and nothing more. It is hard to contest these claims. But after dedicating so many pages to the collective nature of intelligence, the authors could also have proposed less individualistic recommendations. Academic and political institutions play a key role in that story. How could we modify them to promote a healthy relationship of economists, policy-makers, or the public to radical uncertainty? What epistemological, sociological, and political processes drive the misunderstanding of radical uncertainty? The authors do not raise these questions.
13To conclude, the book contains some interesting ideas and is entertaining. It is worth reading for its articulated criticism of applied economic models. For all that, it should be taken neither as an introduction to the concept of radical uncertainty, nor as a political proposal in order to deal with it. The book provides a poor account of what has been done on this concept. It is built on an artificial opposition between what the authors point out and mainstream economics. As a consequence, its own proposals are loose both theoretically and historically and, in my opinion, not very useful for the current analysis of decision-making under radical uncertainty.