¹⁵(Clark, 2013). From the viewpoint of mathematical theory, it might actually be more appropriate to talk about predicted reward instead of expected reward in the definition of reward loss. While these are often seen as the same thing—prediction being an expectation of a future quantity—the concepts are not equivalent. In particular, in machine learning theory, a prediction can be considered more general than expectation: a sophisticated prediction will also include an estimate of the uncertainty involved in the prediction, in addition to the mathematical expectation. This is relevant here because it seems that the certainty of the prediction affects the level of frustration. I would claim that if you are completely certain that you will get chocolate (say, 5 pieces), but then it turns out you don’t, the frustration will be greater than in the case where there is only some chance of getting any (like the example in the main text, 10 pieces with 50% probability). Crucially, in this example the expected amount of chocolate, in the sense of the mathematical expectation, is the same in the two cases, and only the uncertainty changes. Such an effect of uncertainty should be taken into account in the definition of reward loss. I will not do that completely rigorously in this book because such theory seems to be lacking at the moment; however, closely related developments on uncertainty will be found in Chapter 10 and footnote 19 in Chapter 14 (see also the text preceding that footnote).