¹⁴The exact definition of expectation as used in reward loss is not very clear in my view, and an important problem for future research. Not much attention has been paid on it, partly because in typical experiments, it seems obvious what the expectation should be, and there is little planning involved. In a prototypical experiment, an animal (or a human) is given the same (positive) reward several times for a given, simple behaviour, and then suddenly it is given less reward (this is called “successive negative contrast”). In such a case, the future expectation of the reward is simply assumed to be equal to the past reward. With longer plans in more complex environments, the definition will be less obvious. Clearly, there is a strong connection to the concept of a prediction, as discussed next in the main text as well as footnote 15 below. Furthermore, an alternative definition of reward loss might be developed using counterfactual contrast (Roese, 1997), formalized as counterfactual regret by Zinkevich et al. (2008), where the obtained reward is compared with what might have been obtained, if better actions had been chosen. If the agents form some kind of a society, even more options exist for defining the expectations. The agent might use information on what other agents get, and expect to obtain the same reward as others do. Such a “social” expectation might simply be based on probabilistic inference: If the other agents are similar to the agent in question, it is logical to expect that the agent in question will be able to obtain the same amount of rewards (Rutledge et al., 2016). Yet another, very different, form of expectation might be produced in a situation where the agent assumes a moral right to obtain something, assuming the existence of some ethical norms in the agents’ society (Dignum et al., 2000).