When decision theory was first developed, logical empiricism reigned. Accordingly, its developers eschewed terms and assertions that failed strict empirical conditions of meaningfulness. Although many versions of decision theory implicitly reject these rigid standards, Decision Space provides extensive discussion of an alternative view—contextual empiricism—which relaxes the standards that logical empiricism imposed on theoretical terms. Contextualism permits the introduction of intrinsic utilities, allowing Weirich to develop utility theory in a way that increases its unification and explanatory power.
Weirich advances versions of three different forms of utility analysis: intrinsic utility analysis, expected utility analysis, and group utility analysis. The analyses are unified by showing how, when suitably developed, each can be derived from the principle of pros and cons. This foundational principle dictates that options should be divided into pros and cons, with positive weights assigned to the pros and negative weights to the cons. The utility of each option is determined by the sum of its pros and cons. After developing each form of analysis separately, Weirich combines them to produce various forms of multidimensional utility analysis.
The fundamental concept of intrinsic utility is understood as rational degree of intrinsic desire, where objects of such desire are desired for their own sake, rather than for the consequences of their realization. Weirich assumes that an agent has a set of basic intrinsic attitudes (BITs), that is, objects of desire, aversion, or indifference, which cause all other intrinsic desires. By taking the utility that an agent attaches to a possible world as the sum of the intrinsic utilities of the BITs realized in that world, other forms of utility can be understood in terms of intrinsic utility, which gives it a central explanatory role in accounting for rational action. Intrinsic utilities can be used to compute the comprehensive utility of a world, that is, utility all things considered, because what is realized in a world is logically determined by the constitution of that world.
Intrinsic utility analysis assesses utility in terms of an agent's goals, whereas expected utility analysis considers possible outcomes, weighted by their probability. In Weirich's version, options are taken as possible decisions and utilities are assigned to conditional outcomes. First, expected utility is defined where states are independent of options and then extended to the case where states are dependent on options, with considerable attention focused on how to pick out the appropriate probabilities and utilities. The treatment here differs from many others in that Weirich takes probabilities and utilities as basic, rather than appealing to a representation theorem. Such theorems are often invoked to show that, given an appropriate preference ranking, there are probability and utility functions such that those preferences can be represented as maximizing expected utility with respect to the probability and utility functions. Weirich argues that the virtues of such reductions in probability and utility are overrated, and that greater explanatory power is to be had by taking probabilities and utilities as basic.
With versions of intrinsic and expected utility analysis in place, they are applied and combined in ways that illustrate the versatility and scope of the theory. One example involves analyzing the utility of possible outcomes in terms of the intrinsic utility of worlds. This joint analysis is made possible by the fact established earlier that the intrinsic utility of a world is equal to its comprehensive utility. Group utility is understood in terms of the utility of the individuals comprising the group, and is defined as a power-weighted sum of the utilities for the group members. Intrinsic utility analysis applies derivatively via the BITs of the individuals in the group, and expected utility analysis is shown to apply given the assumption of a group probability function. Finally, these forms of utility analysis are extended to situations in which decisions are made for others.
Numerous idealizations are made throughout in the book. These can be characterized broadly as requiring that agents have rational goals, that they have rational beliefs, that they make rational decisions, and that decision situations are themselves ideal. For example, the analysis of group utility assumes not only that the agents in the group are rational, but also that they can communicate and contract with each other without cost. Idealizations are needed for theorizing, though some of those made could be relaxed at the cost of a more complicated theory. To the extent that the principles of utility and decision are taken as regulative ideals, some degree of idealization is also appropriate. However, in the many cases where the idealizations fail to closely approximate actual conditions, there is some question about the extent to which they really can be of service in explaining rational action.
Introducing intrinsic utilities adds to the unity of decision theory, arguably increasing its explanatory power. However, it is not so clear that the explanations (which are achieved by rejecting operationalism in favor of contextualism) are analogous to those that justify contextualism elsewhere in the philosophy of science. A problem with intrinsic utilities is that they cannot be distinguished from extrinsic utilities by examining actual choices. It is one thing to reject operationalist theories of meaning, according to which all terms are to be defined by observable operations, but it is quite another thing to allow theoretical entities whose properties cannot be verified indirectly through observation. A plausible view maintains that electrons were widely accepted, not just because of their explanatory role as theoretical entities, but because it became possible to detect them and measure their charge. This case exemplifies a general principle not to accept entities, despite the explanatory power they may give to the theories in which they are embedded, unless they admit of more direct confirmation. Accordingly, contextualism is to be embraced in science, because it facilitates the development of theories whose consequences may eventually be confirmed, rather than simply for its potential theoretical gains.
While contextualism may permit the introduction of intrinsic utilities, the explanatory virtues of intrinsic utilities do not go along with the empirical verifiability that is sought in science. Weirich is clear that he is willing to forgo such empirical confirmation in order to gain explanatory power; however, the value of explanations divorced from empirical verification is open to question. The treatment of explanation in the book is rather thin, which is reasonable, given the focus on developing utility theory. However, this leaves considerable work to be done in assessing whether the explanatory gains made by introducing intrinsic utilities are worth what some will regard as their substantial cost. Nevertheless, Weirich has achieved an elegant and powerful unification of decision theory through his development of multidimensional utility analysis. The book is rich in details that will be of interest to specialists, but it is so clearly written that both the technical presentation and philosophical arguments are highly accessible.
