Introduction

There is no doubt whatsoever that when an agent involves herself in practical action she must have some reasons¹ that impel her to perform the aforementioned action while guiding the course of her doing. We can assume that reasons for action lie on facts that constitute a case for (or against) the performance of an action or come from previously accepted propositions that are the product of deliberation. Epistemic reasons are reasons for believing in a proposition through being facts which are part of a case for belief in its truth (taking into account ‘truth-related’ considerations). These two claims are controversial and their clarification may help the work of meta-ethicists.

To begin with, it is far from clear whether available epistemic reasons are sufficient to warrant belief since not only truth-related considerations figure among those determining the sufficiency of a specific case. In fact, the factors or principles which determine whether the case for the truth of a proposition is adequate to warrant belief are not necessarily reasons for belief. On the other hand, it has been argued that simplicity, elegance, explanatory power, or other such considerations govern rational belief or theory acceptance. But this is so, while theory acceptance is understood for what it is: acceptance of theories, regardless of belief in them. On these considerations, the simplicity of a proposition or a theory is not always a reason to believe it, although it might suffice to their acceptance. In this context, accepting a proposition is conducting oneself according to the belief that there is sufficient reason to act if we assume the true of that proposition. Aside of the known difficulties of formulating a satisfactory theory that combines appropriately epistemic reasons and practical reasons, capable of explaining in a rational way actions in the practical sphere, we have to deal with the problem of correctly identifying the source of such reasons. Acceptance of the proposition that P furnishes a good reason to act implies the claim that one is justified to act as if P; that is, that one has good reasons to act. But, as Williams argues, it is convenient to refine the treatment of reasons and be aware that they can be internal or external (^{Williams, 1981}).² In what follows, I shall examine Smith’s interpretation of the so called “internalist requirement of reasons” and discuss the construal of rationality surrounding such interpretation.

According to Smith’s interpretation of the “internalist requirement on reasons”, this requirement states that the desirability of an agent’s Φ-ing in certain circumstances C depends on whether or not he would desire that he Φs in C if he were fully rational. The internalist theorist distinguishes two models, namely the example model and the advice model, that might explain how a less than fully rational agent, who finds himself in circumstances in which a decision to Φ can take place, may deliberate in order to act in a way or another. In his paper “Internal Reasons” Michael Smith contends that the example model is wrong and makes an argument for the advice model (^{Smith, 1995}). He also endorses Williams’s suggestion as to what being a fully rational agent amounts to. As previously announced, in this paper I will pursue a threefold goal. I shall argue that Smith’s criticism of the example model is not conclusive, I shall make some critical remarks on Williams’s characterization of fully rational deliberative agents and, finally, I shall offer a formulation of the requirements a rational agent must meet to Φ in certain circumstances.

The example model and the advice model

To begin with, let us take a look at the differences between the example model and the advice model. Consider two worlds: the evaluating world and the evaluated world. To make things easier suppose the evaluating world is inhabited by the agent’s fully rational self (hereafter FRS) and the evaluated world is inhabited by the agent’s less than fully rational self (hereafter LFRS). We can reformulate now the internalist requirement in terms of this construal by establishing that the desirability of the agent’s Φ-ing in the evaluated world is fixed “by some processes and decisions that occur in the evaluating world.” (^{Smith, 1995}: 2). According to Smith, we have to imagine that the agent’s FRS looks from his world across the world of the agent’s LFRS, and forms a desire about what the agent’s LFRS should do. Presumably, the agent’s FRS takes into account his own set of desires (which should be rational) and the circumstances in which the agent’s LFRS finds himself. However, for the sake of discussion, we have to keep in mind that there is only one agent, although we have agreed to consider two aspects of him separated in two communicated worlds³.

The kind of separation instantiated in the two-worlds metaphor, though somewhat artificial has proved illuminating for the explanatory purposes of cognitivists and philosophers. It allows them to explain phenomena like metacognition and reasoning techniques like soliloquium. Let us see if it has the same positive outcome in theory of action.⁴ In the advice model, the agent’s FRS gives advice about what to do to his LFRS. In this model, the desirability of the agent’s Φ-ing in the evaluated world depends on whether his fully rational self in the evaluating world would desire that he (the agent) Φs in the evaluated world. By contrast, in the example model the desirability of the agent’s Φ-ing in the evaluated world depends on whether his FRS (which is in the evaluating world) would desire to Φ in the evaluating world. Notice that in this model the agent’s FRS is not giving advice to himself, but rather setting his own behavior (in his own world) as an example to be followed by the LFRS that inhabits the evaluated world.

Smith argues that the example model is wrong. Since he presents his case by reworking an example by ^{Watson (1975/1982)}, let me quote the situation in extenso:

Suppose I have just been defeated in a game of squash. The defeat has been so humiliating that, out of anger and frustration, I am consumed with a desire to smash my opponent in the face with my racket. But if I were fully rational, we will suppose, I wouldn’t have any such desire at all. My desire to smash him in the face is wholly and solely the product or anger and frustration, something we can rightly imagine away when we imagine me in my cool and calm fully rational state. The consideration that will motivate me if I were fully rational is rather that I could show good sportsmanship by striding right over and shaking my opponent by the hand. In that case, does it follow that what I have reason to do in my uncalm and uncool state is stride right over and shake him by the hand?’ (^{Smith, 1995}: 3)

Smith grants that a FRS might have the desire of shaking hands with the victorious opponent, but he denies that a LFRS could have a reason to do this, even if his FRS sets that desire as an example. Smith feels that, on the contrary, given the state of anger and frustration which precisely originated the desire of smashing the opponent in his face, the LFRS must avoid physical proximity with his opponent, so that he can avoid acting upon a desire that was created out of anger and frustration. What the LFRS has reason to do in this case, Smith argues, “is to smile politely and leave the scene as soon as possible.” (^{Smith, 1995}: 4) On Smith’s account, apparently the example model fails because the LFRS would be incapable of following his FRS example, due to the particular circumstances in which he finds himself.⁵ The advice model would prove workable, since what the FRS would advice his LFRS on the spur of the moment would probably be to leave the scene quickly.

Several precisions need to be made here. First of all, let me call to your attention the distinction that must be established between an agent’s FRS and the same agent’s LFRS. Although neither of those is adequately defined in Smith’s paper, we can agree that the FRS is free of emotions, is always in a “calm and cool” state; deliberates rationally (this will be discussed later on) and is capable of reasoning in such a way that his decisions and desires can be set out as examples to be followed by his LFRS. On the other hand, the latter can have desires that stem from emotions; that is, can have desires and act according to them, in a way that would not be approved by the FRS. Let us put aside for the moment the fact that human nature precludes the possibility of a world of purely FRSs, or even the possibility that we can desire and consequently act all the time as one of them. Let us focus on Smith’s example. What prevents the FRS from advising his LFRS that he shakes hands with his opponent? Smith contends that the FRS would advise his LFRS to leave the scene quickly because the physical proximity might tempt him to act upon his previous (irrational) desire, that is, that the former would give such advice for practical reasons, a plausible account. However, nothing prevents the FRS from advising his LFRS to show good sportsmanship. After all, if the agent is deliberating; that is, if the LFRS is giving a second thought to his violent initial desire, why he could not follow the best possible advice? Smith seems to believe that the LFRS is so irrational⁶ that his FRS has to compromise and settle up for a less undesirable desire, instead of conveying the best one in his advice. Furthermore, he seems to think that deliberative rationality (including evaluation of background conditions, for example) will lead to the FRS to forget what he would do (were he to act) and advice something intermediate. But Smith’s distinction between the example model and the advice model is not sharp enough if it is going to be grounded only on what the FRS would do if he were to act in certain circumstances and what he would advise his LFRS in the same circumstances.⁷

Notice that Smith’s argument presupposes that either the FRS or the LFRS might act (there is not doubt as to the fact that they both can have desires). Even granting that the FRS’s acts are merely counterfactual⁸ as they appear to be in both models, some problems arise. Let us go back to the squash game for a moment. We devote a lot of time in our culture to teaching the rules of social games. Although we promote competence we also point out sportsmanship, and we expect the principles of good sportsmanship to hold everywhere, from school to the professional fields. It is true that we see a lot of violent, irrational or uncivilized behavior in sport games but we condemn such instances of irrational behavior. We expect good sportsmen to act according to the most rational desirable attitude (shaking hands). To foster this kind of behavior, we insist on self-control, which is an alternative name for deliberation in the case we are discussing. Probably the defeated contender in a squash game would have (momentarily) the desire of smashing his opponent, but if he is capable of deliberating, there is nothing that prevents him from shaking hands instead of smashing the face of his opponent. If the FRS is in control of the situation, that is, if he can act after deliberation, he would surely shake hands, but if he is not in the position to actually act, nothing prevents him from advising his LFRS of doing what he, the FRS would do in the same circumstances. If this account is correct, the distinction between the advice model and the example model becomes artificial, a matter of mere ways of presenting the facts, since we can have the same story we have been telling with the two models, just by appealing to instances in which the agent shows good sportsmanship and self-control and instances in which his behavior betrays a total lack of self-control.⁹

In the practical sphere (real world) many people act exactly as the FRS would do. Take the cases of politicians, professional sportsmen, and many more public figures. It could be argued that the motivation for their actions stems from wrong or selfish desires (keeping a good public image, for instance) but this does not prove that there are no cases of good rational deliberation. Smith argues that the FRS would advise his LFRS to leave the scene of the game, regarding the circumstances in which the LFRS finds himself. I already suggested that, on Smith’s account, this might be so because upon rational deliberation the FRS thinks that leaving the scene is the best thing to do. However, we can conjecture different reasons to support such a conclusion, and not every reason will be compatible with Smith’s view. We can conjecture, for example, that the FRS knows that his LFRS is incapable of doing what he would do because his LFRS can hardly control his emotions; we can conjecture that the FRS does not trust his LFRS willingness to follow hard or unpleasant advices, but we can also speculate that from the fact that an advice is something that one is not forced to follow, the FRS gives the one he thinks his LFRS is more likely to accept.

The two worlds metaphor

It appears that the two worlds metaphor in which the example and the advice models are developed is misleading. Smith says that the FRS looks across his world (the evaluating world) to his LFRS (who is located in the evaluated world). Both, the FRS and the LFRS have desires, but the LFRS’s desires are subject to evaluation, and the criterion of evaluation should be the FRS’s own desires. The first inconsistency that strikes me in the example model is that the FRS’s desire takes place in the evaluating world but it is also aimed to the evaluating world. But presumably, the evaluating world is not a world in which anyone, not even the FRS, can act. Acts are carried out only in the evaluated world. One can only wish that the LFRS would act following the example. However, things are not better on the side of the advice model. In this model the FRS would form a desire (in the evaluating world) about what his LFRS should do in the evaluated world, and then he advises his LFRS to act accordingly. It is obvious that in both models we have cross-world communication, so this fact cannot account for the difference between them. How are they different anyway? We lack a clear-cut answer, but it is obvious that at the end of the day both models can mix up together. Perhaps the source of this confusion could be the fact that in both models we are dealing with the case of a self who deliberates by reasoning with himself, that is, with one self who sometimes behaves up to the highest standards of rationality and sometimes behaves less rationally, regarding the circumstances, but who could do his best given enough time and right training in rational deliberation. It seems, then, that the separation of the self into such components is not that productive for the theory of action.

Before moving on to the last section, let me make one more comment. On Smith’s account the “example model of the internalism requirement thus gives us the wrong answer in cases in which what we have reason to do is in part determined by the fact that we are irrational” (^{Smith, 1995}: 4)¹⁰. In a charitable reading of this paragraph we would have to understand that he is considering the LFRS desire of smashing his opponent in the face as a desire formed out of sheer irrationality. But we have to remind ourselves of the fact that though we are not fully rational, this does not entail that we are irrational, simpliciter. We may act, sometimes, in irrational ways but that does not make us irrational beings in the same way that failing to reason in a fully rational mode (or committing mistakes during the process) does not entail that we are incapable of engaging in good reasoning practices. Quite differently, rationality and irrationality are not the horns of a dilemma, they are the standards with which we judge and classify actions and processes which we perform and that we consider worth of critical analysis.

Rationality and acting rationally

Let us deal now with the characterization of rationality. As mentioned it earlier, nothing has been said about the meaning of the expression ‘fully rational’. Smith fills up this gap by endorsing Williams’s similar analysis of the subject of internal reasons. Any agent that we are to consider fully rational in the practical sphere has to meet three conditions:

Smith believes that Williams’s conditions (i) through (iii) provide a fairly accurate account of the idea of what it means to be practically rational.¹¹

However, he disagrees with the role Williams’s account confers to imagination in the process of deliberating correctly, and for this reason he thinks those conditions require supplementation and amendment. I am not going to enter into the details of Smith’s argument for this contention since my main interest here lies in the characterization of practical rationality that both Smith and Williams share.

At first sight this characterization seems to work fairly well. There is a glass in front of me that contains gin and petrol (a rather unusual mix, but certainly a possible one), but I mistakenly believe it contains gin and tonic and I form the desire to drink from it. Here condition (i) is violated. To Φ in this case (i.e., to drink from the glass) would be objectively undesirable, and my desire to do so can be explained solely because of the fact that a false belief misleads me into desiring what I should not if I were not mistaken.¹² But suppose I do not risk myself into guessing what the glass contains, and consequently I do not form any desires (in particular, I do not desire to drink from the glass). Does this mean that drinking the stuff in the glass would be objectively undesirable? Absolutely not. Because it may turn out that the glass contains “the most delicious drink imaginable” and my failure to desire to Φ in this case has to be attributed to the fact that I lack the relevant true beliefs. In this instance, violation of condition (ii) explains why I may fail to desire something that I would, if I were fully rational in the sense of (i) through (ii). Finally, deliberation aimed to the formation of new desires requires the interaction of beliefs and desires and condition (iii), because even meeting conditions (i) and (ii) we may fail to Φ in cases in which Φ -ing will still be desirable.¹³

Nevertheless, conditions (i) and (ii) indeed require some clarification. If they are to be considered workable, then they have to be relativized to particular cases. We cannot seriously hold (i) and (ii) in a general non-relativized way simply because there is no human agent that has all relevant true beliefs (about all possible objects of desire) or that totally lacks false beliefs about some thing or another. But conditions (i) and (ii) are still problematic, even in a relativized construal. Let me suggest two ways in which this problematic feature can be illustrated: 1. No matter how simple a possible object of desire might look, the set of beliefs about it seems to be always open-ended. It may be the case that we cannot have all the possible true beliefs about such an object just because some tiny aspect of it has escaped our attention. In other words, whether or not a human agent has all the relevant true beliefs about something is always a disputable fact. 2. Condition (i) seems to contradict the basic tenets of some widely accepted theories of rationality. Mutatis mutandis, in Popperian critical rationalism, for example, rational deliberation is a process in which the deliberating agent tests his beliefs against experience.¹⁴ If there is a conflict between the belief and the relevant part of experience then the agent rejects the false belief and replaces it with a new one, which in its turn has to be tested again. There is a long and ongoing process of correction (elimination of error) that refines the set of beliefs and that provides the agent with the best possible set of beliefs (one containing as few false beliefs as possible). But in all stages of this process we might still have (or form) false beliefs. To sum, a working definition of rationality does not have to exclude false beliefs, on the contrary, it has to accommodate them and provide a working method to suppress them. Granted, false beliefs might mislead the agent and lead him to make the wrong choice or desire what is not desirable. But in the long run, the agent will be able to refine his set of beliefs and make better choices, desire what is desirable, and act accordingly.¹⁵

This way of looking at the question of rational deliberation seems to give a better account of two well known facts. The fact that we devote a great amount of time trying to teach our children to desire and act in ways we consider right, and the fact that in certain circumstances even mature individuals sometimes hesitate and look for advice, not from their FRSs but from someone who might have more experience, more knowledge or simply more ingenuity about practical matters. Regarding the distinction between a FRS and a LFRS, it seems that the qualifier “fully” could be dropped out with no harm to the overall argument. If one is reluctant to do that, then I suggest to interpret that qualifier in a weak sense (which anyway would be compatible with Smith’s account since he attributes a certain degree of irrationality to the FRS). In this weak sense, a fully rational agent is one who is able to test critically his reasons (through rational deliberation) and upon reflection is able to make the necessary changes in his subjective motivational set, form the right desires and act accordingly.