Kantian Economics

“Kantian” Economics

Minhao Benjamin, CHENa April 2009

a

benjamin.chen « at » polytechnique.edu

2 ETHICS AND DECISION-MAKING Individuals are often presented with choices in their daily lives – what to have for breakfast, which college to attend, who to marry, etc. Some of these choices are decidedly trivial in that they do not involve much conscious deliberation. The choice between sausages or pancakes can well be decided by a simple appeal to the senses. Other choices, however, could require a more vivid imagination and might involve considerations of regret. These are usually decisions that greatly influence the course of our lives (or at least we think they do), and which most people make only once or twice in a lifetime. In these situations, we tend to take a trip into possible futures, allowing our minds to explore the potentialities of each choice and anticipate the wrenching regret we might experience if things do not turn out as we had hoped. The moral dilemma occupies a position that is distinctly different from the cases we have previously described. When we mull over a moral question, we search for the “right” course of action. Classical utilitarians would tend to argue that the moral act would be the one that maximizes total utility where, in the immortal words of Jeremy Bentham, “each person is to count for one and no one for more than one”. The moral assessment of a choice is therefore an unadulterated function of its consequences. On the other hand, there are philosophers who assert that such a consequentialist approach to ethical decision-making is deeply dissatisfying and that we should privilege intentions when evaluating the moral worth of an action. The Kantian Categorical Imperative which commands us to “act only according to that maxim whereby you can at the same time will that it should become a universal law” is situated within this philosophical tradition. The permissibility of an action depends on the maxim upon which the agent acts. THE PRACTICE OF MORAL JUDGEMENT It is important to emphasize that as practical tools, both utilitarianism and the Kantian categorical imperative are normative frameworks that address questions residing within a predefined moral structure. It would be a mistake, in my opinion, to extend these frameworks to more general settings. In choosing to attend school, we do not usually consider whether this would be the best social use of our time – i.e. whether this is the action that would maximize total utility. Similarly, even the most committed Kantian agent would not try to decide between apples and oranges by appealing to the categorical imperative. The practice of moral judgment precludes moral naïveté on the part of the ethical decision-maker. Barbara Herman suggests that we should include in the description of the Kantian agent what she calls “rules of moral salience”. These rules are often acquired as part of a moral education, and allow the Kantian agent to grasp the morally relevant elements of his circumstances and intended course of action. Rules of moral salience can also impose a burden of justification on certain types of actions. For instance,

3 The agent in the deceitful-promise example understands that normal prudential or instrumental justifications of actions will not do in the case of deceit. This is the mark of his “conscience”. It seems he already knows that deceit in the pursuit of (mere) self-interest is usually not permissible. So he comes to the CI knowing that, unless his situation is morally exceptional, it will not be judged permissible. Issues of excuse and justification do not enter moral thought only after an action has been done; they are part of the way a normal moral agents explains the permissibility of his actions to himself.b

It is important to note that rules of moral salience do not constitute prima facie duties. They serve to bring to the agent’s attention the relevant details of his situation that require moral consideration. The agent then weighs these elements against each other and the fruit of his deliberation is captured in a maxim which is then presented to the Categorical Imperative for evaluation. Only after passing through this acid test is the proposed course of action considered to be permissible. MODELLING ETHICAL BEHAVIOUR In examining Kantian ethical behavior by the economic agent, I will follow the moral theory developed by Herman. While Herman acknowledges that “the claim that… [rules of moral salience are] necessary follows from the structure of the CI as a practical principle of judgment (italics mine) and not from the correctness of [her] textual interpretation”c, such a claim is precisely what we require for a coherent model of the Kantian agent who conscientiously seeks to apply the Categorical Imperative. We will begin by defining the set P as the set of all possible statements about the world. Since expression is necessarily constrained by language, P is a countable set. Accordingly, we will also set [0, 1] as our canonical state space. The canonical state space captures all the possible circumstances the agent can face. Of course, only some of these circumstances would be morally relevant, and we denote by F = P '×I , P ' ⊂ P , I ∈{0,1} the description of all such relevant features. F will usually be finite. We will also define an action space A which we shall assume for the moment to be finite and independent of α ∈[0,1] ). Finally, we assume that the agent’s underlying preferences over A are complete, transitive and therefore admit a utility representation. For any α ∈[0,1] , the agent first solves the standard utility maximization problem: max U α (a ) for all a ∈ A . We then introduce the function RMS : A ×[0,1] → P( F ) which describes the agent’s pre-existing moral knowledge. If RMS (a*, α) = {∅} , the candidate action b

Herman, Barbara. “The Practice of Moral Judgment”, The Journal of Philosophy, Vol. 82, No.8 (Aug., 1985), pp 418 c Ibid, pp

4 is trivially permissible (i.e. there is no moral content to the decision problem) and all such action in argmax U α (a ) can be an outcome of the decision-making process. If RMS (a, α) is non-trivial, then there is a moral question to be resolved. The agent will then provide the maxim upon which he proposes to act. For the Kantian agent (and not merely the ‘glib’ talker), the maxim that is furnished will not be an object of choice. The maxim can therefore be identified with, and represented by, (a*, α, RMS ( a*, α)) d. It only remains for us to define the mapping CI : (a*, α, RMS (a*, α)) →{0,1} . If the outcome is 1, the action is nontrivially permissible, and can be a possible outcome of the decision processe. If the outcome is 0, the action is forbidden to the agent. If there is no element of argmax U α (a ) for which CI (( a*, α, RMS ( a*, α))) =1 , the agent has to solve the optimization problem 1 max U α (a ) for all

a∈ A \ A

where

A1 ≡

argmax U α (a ) , a ∈ A , and the cycle repeats.

Since the action space is assumed to be finite, the number of iterations will also necessarily be finite. 1) Suppose the iteration stops and we have at least one ‘permissible’ action in the action space). Then, the Kantian agent will maximize his utility by choosing amongst the alternatives that are permissible. Since the underlying preference ordering over the initial action space was complete and transitive, the preference ordering over the final action space would also be complete and transitive. 2) Suppose the iteration stops and the final action space is empty. Then, it is vacuously true that the preference ordering over the final action space is complete and transitive. We can therefore say that with complete and transitive preferences over the initial action space, the Kantian agent behaves as if he were maximizing a utility function. This result should not surprise us since the agent started with complete and transitive preferences, and the Categorical Imperative merely imposes certain constraints on the agent’s choice set. The above formalization also allows us to identify one of the potential shortcomings of Kantian decision-making – we cannot ensure that the We include RMS ( a*, α) since in the case where we would like to have several non-identical agents, each agent may identify different particulars of his case as being morally relevant, even though they face identical circumstances. e For the purposes of this discussion, we make no normative distinction between a trivially-permissible and non-trivially permissible action. d

5 deliberative process will result in an actual moral choice. It might well turn out that in certain extreme situations (and where the choice of possible actions is limited) no action will be sanctioned by the Categorical Imperative. Indeed, it is often forgotten that even if we accept the impermissibility of telling a lie to save a life, this does not directly imply that telling the truth (or indeed any other conceivable action) is morally permissible. This is in contrast to the utilitarian approach where starting from the same assumptions, a solution is always guaranteedf. “KANTIAN” ECONOMICS What do Kantian ethics and economics have to learn from each other? Laffont (1975) argues that “to induce people to behave in a ‘Kantian’ way may be a good solution to some problems”g, where by “problems”, Laffont refers to the classical question of optimal resource allocation. Roemer (2006) proves the existence and Pareto-optimality of “Kantian” allocations but also concludes that “Kant’s categorical imperative only implies cooperation (solidaristic behavior) conditional upon the rewards to cooperation being sufficiently great, perhaps a sobering thought for philosophical Kantians who believe that Kant’s categorical imperative implies a strong kind of solidarity”h. At first sight, the Kantian agent appears to have gained a toehold in economic theory. Yet, I would venture that the formalization of the Categorical Imperative in both Laffont (1975) and especially Roemer (2006) rests on a misunderstanding of Kantian ethics. Laffont (1975) “postulate[s] that a typical agent assumes (according to Kant’s moral) that the other agents will act as he does, and he maximizes his utility function under this new constraint”. Following Laffont’s lead, Roemer (2006) formalizes the Categorical Imperative in the following way: There is a set of n agents, each with a utility function u i : ℜ n+ → ℜ . The argument of the utility function is denoted by ( L1 ,..., Ln ) . L−i ≡( L1 ,..., Li −1 , Li +1 ,..., Ln ) . We call the profile of utility functions { u i } monotonic if for all i either u i is strictly monotone increasing in L−i or u i is strictly monotone decreasing in L−i . Definition : Let { u i | i=1,…,n} be a monotonic profile. An allocation of effort ˆ =( L ˆ1 ,..., L ˆn ) with the property that no agent would prefer that all agents L change their efforts by the same factor is a Kantian allocation. The Kantian philosopher would, I think, object to the preceding characterization of the Categorical Imperative for two reasons, both of which are intimately linked to each other. f

This follows from standard assumptions on utility functions and classical microeconomic analysis. Laffont, J-J. 1975. ‘Macroeconomic Constraints, Economic Efficiency and Ethics: An Introduction to Kantian Economics.’ Economica, vol. 42, pp. 430-7 h Roemer, J. 2006. “Kantian allocations.” Cowles Foundation Discussion Paper g

6 The first red flag is the conspicuous absence of maxims and motives which lie at the heart of Kantian moral philosophy. The definition proposed by Roemer (2006), bears a distinctly consequentialist flavor. The apparent contradiction lies in the fact that Roemer’s Kantian agent performs the counterfactual universalization, but then proceeds to maximize his utility based on that assumption whereas for Kant, the moral agent is invited, through universalization, to verify that the action he proposes to undertake, for the reasons given in his maxim, does not lead to a contradiction in conceivability. The second point of contention is that the categorical imperative as a procedure does not reveal to us the ‘optimal’ action, but only tells us if an action we propose is permissible. Elster (1989) makes the distinction between what he called “everyday Kantians”, who cooperate if and only if universal cooperation yields a Pareto improvement over universal defection, and “true Kantians”. He is careful to note that One reason one cannot will that all will act in a certain way might appear to be that it could be worse for all if one did so. This is not Kant’s view, however. Rather, he argued that one cannot will Xi if the notion of all doing X harbours a logical or pragmatical contradiction.j

Despite Elster’s caveat, however, the label attached to the thinking behind “everyday Kantianism” seems like a rhetorical sleight-of-hand given what we understand of the foundations of Kantian moral philosophy. As such, it is quite a surprise to see Roemer (2006) claim that his “formalization of that (i.e. Kantian) premise shows that the solidaristic behavior it implies is quite conditional – upon the rewards to co-operation being sufficiently high”k, with the concluding advice that “those philosophers who call themselves Kantians, and believe that this identification justifies a solidaristic world-view, I suggest, should take heed”l. It could perhaps be suggested that philosophical differences aside, the formalization proposed by Roemer (2006) adequately approximates the action of Kantian moral agents. It is difficult to see how this can be the case. Let us return once again to the well-worn example of the prisoner’s dilemmam that Roemer uses to demonstrate limited cooperation under the Categorical Imperative. In this example, the strategy space is defined to be [0,1] , where a strategy of p means that the agent will cooperate with probability p. Yet, depending on the features of the problem that are judged to be morally salient, the application of the Categorical Imperative can yield different outcomes. To illustrate this point, I shall return to Benjamin Constant’s notorious example of the murderer at the door. i

X here seems to refer to an action, but if we want to be precise, it is the maxim of the action that is willed as universal law, a point that has been often been gleefully seized upon by critics of Kantian moral philosophy. j Elster, Jon. The cement of society: A study of social order. Cambridge: Cambridge University Press 1989. k Roemer, J. 2006. “Kantian allocations.” Cowles Foundation Discussion Paper l Ibid. m We might of course inquire as to what a Kantian moral agent is doing inside a prison cell, but let us leave that question aside for the moment.

7 Kant insists that it is morally impermissible to lie to the murderer about the whereabouts of your helpless friend. However, Korsgaard (1986) argues that If murderers standardly came to the door and said:” I wish to murder your friend – is he here in your house?” then perhaps the universal practice of lying in order to keep a murderer from his victim would not work…But the murderer is not likely to do this, or, in any event, this is not how I shall imagine the case. A murderer who expects to conduct business by asking questions must suppose that you do not know who he is and what he has in mind. If these are the circumstances, and we try to ascertain whether there could be a universal practice of lying in these circumstances, the answer appears to be yes. The lie will be efficacious even if it is universally practiced. But the reason it will be efficacious is rather odd: it is because the murderer supposes you do not know what circumstances you are in – that is, that you do not know you are addressing a murderer – and so does not conclude from the fact that people in those circumstances always lie that you will lie.n, o

Thus, without being precise about what is at stake in the prisoner’s dilemma (and even if we were), it is impossible to endogenously model what action a Kantian agent would or would not be allowed to take. Furthermore, it is not clear if non-trivial mixed strategies can be a “Kantian” solution unless both cooperating and defecting are morally permissible courses of action. Electing to defect 50% of the time (or to defect if a coin toss comes up heads) runs into the same moral problem (assuming one exists) as choosing to defect. As long as there is a positive probability of defecting, the Kantian agent will require that the maxim under which he proposes to defect be tested by the Categorical Imperative. It is therefore not clear to me that the approach of Laffont (1975) and Roemer (2006) satisfactorily capture the essence of the Kantian Categorical Imperative. Even if we allow them the latitude that Elster provides in his definition of “everyday Kantianism”, it would seem that the claim of philosophical affinity between this mode of thinking and Kantian ethics is tenuous at best. WHAT IS LEFT? It would seem that economics has little, if anything, to teach the Kantian philosopher. As a distinctly non-consequentialist approach to ethical decisionmaking, Kantian moral philosophy seems invulnerable to economic critique.

n

Korsgaard, Christine. “The Right to Lie: Kant on Dealing with Evil”, Philosophy and Public Affairs, Vol. 15 No. 4 (Autumn, 1986), pp. 325-349 o It would seem that what is included in the maxim would generally depend on our moral education and what we perceive to be morally relevant to the ethical dilemma. Thus, I am unsure if Korsgaard’s argument is fully valid – how many of us, transported into this hypothetical scenario, would consider the question of common knowledge to be morally relevant? Nevertheless, this highlights the importance of circumstances in determining our maxims, and whether or not an act will be judged permissible by the CI.

8 On the other hand, as I have tried to show previously, Kantian ethical behavior can enter into economic models in the form of constraints. These models might help us understand how economic allocations would differ if some agents were to employ the Kantian Categorical Imperative in ethical decision-making. Nevertheless, the usefulness of such models is very much limited by the fact that the agent’s maxim and set of actions which are permissible have to be exogenously given. Furthermore, from a practical perspective, it is not clear if sufficiently many of the individuals who participate in the marketplace adopt the Kantian Categorical Imperative for it to be significant to economic analysis. Finally, it is possible to claim that even in “everyday” economics, ethical considerations have been implicitly assumed. For instance by enforcing a budget constraint, we do not allow economic agents to rob or steal endowments. While this is an interesting argument to pursue, it is not clear if our motivation for such behavior is Kantian (allowing theft under some extenuating circumstances would destroy the institution of property), utilitarian (everyone would be “better off” if I did not steal), rule utilitarian (society would be “better” off if we, as a rule, did not steal) or simply prudential (I would be very likely to get apprehended and punished if I stole). The truly sobering conclusion we are, perhaps, forced to reach is that economics and Kantian philosophy have little to gain from conversation with each other.

REFERENCES 1.

Elster, Jon. The cement of society: A study of social order. Cambridge: Cambridge University Press 1989.

2.

Herman, Barbara. “The Practice of Moral Judgment”, The Journal of Philosophy, Vol. 82, No.8 (Aug., 1985), pp 414-436

3.

Korsgaard, Christine. “The Right to Lie: Kant on Dealing with Evil”, Philosophy and Public Affairs, Vol. 15 No. 4 (Autumn, 1986), pp. 325-349

4.

Laffont, J-J. 1975. ‘Macroeconomic Constraints, Economic Efficiency and Ethics: An Introduction to Kantian Economics.’ Economica, vol. 42, pp. 430-7

5. Roemer, J. 2006. “Kantian allocations.” Cowles Foundation Discussion Paper