The Death of Philosophy: Reference and Self-reference in Contemporary Thought

We owe the introduction¹ of the term “transcendental argument” to P. F. Strawson, who showed² how Kant, in the Critique of Pure Reason, used two types of argumentative systems to establish his theses: On the one hand, and most often, Kant devotes himself to a description of the way that our faculties, in harmony with one another, produce experience. On the other hand, and much less frequently, Kant has recourse to a type of argumentation that aims to show that if we do not accept a given concept, then we can neither think nor act as we understand ourselves to do. Thus, Kant justifies reliance on the concept of cause by showing that this concept is the necessary condition for our experience of succession. To show how certain concepts or series of concepts are necessarily implied in cognitive operations that we actually carry out is, in Strawson’s eyes, the nuclear structure of “transcendental argument.” Strawson uses this argumentative system in his own philosophical analyses, detaching this argument from its initial context in order to consider it as a type of valid reasoning, and thus reusable beyond the single context of transcendental philosophy. Thus, he means to show that we cannot differentiate the objects that we perceive if we are not able to understand ourselves and the objects as two entities coexisting in space. Following him, and again considering a general type of commonly executed cognitive operations (like, for example, perception of objects, reference to something, or even predication), Gareth Evans tried to bring out their necessarily implied presuppositions. In the same way, we have seen how Jürgen Habermas and Karl-Otto Apel use this type of reasoning when they try to bring to light the necessary conditions underlying our most common communicative experiences. Transcendental argument thus consists in bringing out the necessary presuppositions of an experience of thought or speech. Having recalled the argument’s definition, in light of the context of its birth (namely, Strawson’s revival of the Critique of Pure Reason’s reasoning), it seems permissible to formally analyze the argument in the following way:

In order that a transcendental argument be considered as such, the premise (p) must be a cognitive operation, like, for example, perceiving an object, establishing a link between cause and effect, or making any prediction.⁴ These are the cognitive operations that we perform daily. These operations thus refer to propositions of the type, to take up the examples cited by Strawson, “I perceive objects,” “I get my bearings in space,” “I communicate with others” and not to assertions about reality like “The earth rotates” or “Gold is a metal.” The initial fact or factum cannot be any old fact about the world but must be a cognitive factum. In contrast to ontological arguments that aim to establish a proposition that applies to the objects themselves and their properties (arguments that thus take the form of “s is p”), a transcendental argument must establish a proposition that applies to the concepts or representations necessarily implied in our thinking. As Michel Bitbol, who is interested in the use of transcendental arguments in quantum physics, has emphasized, “transcendental arguments [have] absolutely nothing to tell us from an ontological point of view.”⁵

In summary, a transcendental argument is an argument that does not directly apply to the reality of the world but to cognitive operations.⁶ It shows that human beings’ everyday representations would not be possible without a certain number of conditions that the researcher must determine (q, r, etc.). This is why a transcendental argument is often articulated in a negative way, with propositions of the type “We cannot do otherwise than to posit q” or “We cannot not posit certain conditions in order that certain representations become intelligible.” For example, the representation of an object presupposes the possibility of distinguishing it as an entity distinct from ourselves, just as the possibility of understanding a communication presupposes the possibility of distinguishing a speaker. It is clear, in such a framework, that a transcendental argument will be all the more convincing as the initial premise or factum will be difficult to challenge.

Given this first definition of transcendental argument, it goes without saying that it was the object of challenges, contestations, and objections, which justifies the now generally used phrase “the dispute about transcendental arguments.” This dispute has the advantage of uniting two paradigms that are taken to be opposed—the Continental paradigm (references to Kant) and the Anglo-Saxon paradigm, since, as Sandra Laugier recalls in a recent article, “debates about transcendental arguments … have dominated analytic philosophy at the end of the twentieth century.”⁷ And indeed, from Strawson to Jaakko Hintikka, from Dieter Henrich⁸ to Richard Rorty,⁹ from Barry Stroud to Apel, and even, quite recently, from Michel Bitbol¹⁰ to Elie Zahar,¹¹ there are numerous authors from both horizons who have taken part in this famous “dispute about transcendental arguments.” Against Strawson’s argument, three major types of challenges can be identified.

1. The first objection comes from Barry Stroud in a 1968 article.¹² Taking up Strawson’s work, he attacks transcendental argument on the grounds that the latter does not manage to achieve the goals that it is assigned. Thus, for Stroud, transcendental argument had meaning only in Kant in opposition to skepticism, that is, for him, to philosophies that deny the possibility of positing the existence of a world beyond thought. I summarize Strawson’s argument in table 1.

Stroud objects that a skeptic will deny the first premise and the possibility of moving from the third proposition to the fourth. Indeed, hallucinations and dreams are two counterexamples that can be used in opposition to Strawson’s argument. Thus Stroud can conclude that it is possible to assert only that the idea of the perception of an object calls for the idea of the existence of objects outside the self. Consequently, the transcendental argument misses its objective. The first question in the debate is thus the following: is transcendental argument’s only purpose to challenge skepticism, itself reduced to the simple position of doubt about the existence of things outside ourselves? And must we conclude, with Stroud and later Rorty, that this purpose cannot be attained, that without an appeal to principle the argument loses any meaning and value?

2. The second objection concerns transcendental argument’s lack of distinctiveness, in that it may be perfectly reducible to a simple logical argument. Indeed, what does proposition 2 (“It would not be possible that p if we did not think that q.”) mean? It can be conjugated in the following way: proposition 1 (“That p.”) presupposes proposition 2 (“It would not be possible that p if we did not think that q.”), meaning that 1 is not true unless 2 is true. Transcendental argument is thus nothing other than a deductive logical argument in which proposition 2 is only a premise having the status of a postulate. Thus Alain Boyer, summarizing a widely held analytic position, asserts that there is no place in science for arguments other than

Thus, as Boyer concludes, “a reasoning is either deductive or non-deductive; in the latter case it is inductive. Tertium non datur.”¹³ The question here is thus to know whether there are argumentative systems other than simple deduction (subject to arbitrary postulates or axioms) or simple induction (a simple probabilistic assertion, always subject to falsification).

3. The third objection is transcendental arguments’ possible lack of fecundity. Transcendental arguments may in fact be powerless to help us discover new statements, propositions unknown until now. Indeed, because they almost always make recourse to a negative proposition of the form “You cannot say that x,” they can seem to fail to produce positive statements. Can transcendental arguments be anything other than simple machines for refuting past systems or propositions, and, if yes, what really enables them to do so? This is the third question that arises from this objection. These three objections can be presented in table 2.

In the first place, the initial factum is very different from Strawson’s “mental facts” or Apel’s “facts of communication.” This initial factum is not a mental given (like “I perceive myself as distinct from objects”) nor an empirical given (in fact, we communicate) but is rather a claim inscribed in certain types of statements—a claim to truth and to universality implied in scientific or philosophical statements. It is thus a question of examining a demand at work in the structure of sentences and not of starting from a fact that is considered to be indisputable. In this sense, transcendental argument here does not repeat Kant’s stance—which has been criticized since his time, notably by Karl Leonhard Reinhold, for whom the Critique of Pure Reason’s structure is clearly hypothetico-deductive because it can be analyzed in these terms: if mathematical and physical propositions are valid (the initial factum), then this can be only under certain conditions (intuition, concepts, etc.). But this does not start from a fact that could be considered primary or originary; it is a matter of determining under what conditions certain types of statements (statements that make a universality claim) can be consistent. In a parallel fashion, this does not start from a fact that could be considered true (the discipline of mathematics) but from the truth claim.

The type of argumentation employed can therefore be expressed like this: (scientific) validity claims or (philosophical) truth claims require that a certain number of rules or conditions, that an explanation can bring to light, be respected. Thanks to this formulation, we can understand why this is about a type of transcendental argument, for it is understood in terms of the conditions or requirements indispensable for a given x. At the same time, it allows us to precisely define its difference from the canonical formula, for the factum in question is a class of statements, those that make universality and truth claims.

Thus defined, how can this transcendental argument answer the objections I’ve enumerated? Against Stroud’s and Rorty’s arguments, I can show that it is not at all about establishing some sort of realism against skepticism.¹⁴ Against the attempt to reduce the argument to other, more classic forms of reasoning, I can show the irreducibility of this reasoning. From there, I can establish its fecundity.

First of all, to respond to the skeptics, I must say that transcendental argument is not about trying to secure the existence of objects outside ourselves.¹⁵ Transcendental argument has no immediately ontological value but, on the other hand, has an epistemological bearing insofar as its concern is to examine the validity claim inscribed in nucleo in any philosophical discourse. Transcendental argument concerns the status of a discourse that makes a truth claim, not the limited contents of this or that philosophical or scientific proposition. In other words, from physics to economics, from mathematics to philosophy, from biology to sociology, a speaker claims to say something true and cannot assert that in the framework of the practice of his science he claims to say things that are entirely false (even if in fact he can make false statements, he does not claim to do so without giving up the particular status of his discipline, which is to want to say the truth). It follows that if the skeptic wants to challenge a transcendental argument thus reformulated, he must understand that the question does not concern the relation between our representations and exterior objects but rather the relation between a speaker and what he says when he wants to hold a discourse that makes truth claims. Can a skeptic go so far as to say that everything he says is false, including the proposition that he just uttered that “everything is false”? It is thus the status of a discourse that is at stake in this renewed version of the transcendental argument.

Next, to the objection that the argument lacks originality or distinctiveness, we must recall that transcendental argument rests upon the ground of the congruence between what is said (the contents of a discourse) and the procedures employed to be able to say what is said. The argument here does not consist in drawing a consequence from a given premise (deductive reasoning). Nor is it the generalization of an empirical given. The cognitive process required by this type of argument is thus to reflect, at the same time as one articulates the effective contents of a proposition, on the procedures of utterance underlying one’s statement. The argumentative process consists in reconstructing the conditions upon which a proposition—or a series of propositions—acquires meaning, coherence, and consistency. Each proposition refers in fact to a network of more or less implicit statements and conditions for its consistency. These are the propositions tacitly presupposed “to be able to say what is said” that transcendental argument means to methodically reveal in every philosophical system. Argument thus comes back to find the grounds intrinsically and implicitly attached to an assertion that makes a truth claim. It is thus a matter here, as Jaakko Hintikka put it, of directly confronting the question, “How do we know that we know?”¹⁶—or, to put it differently, “How can a scientist say what he says?” It is clear, in doing so, that this is not a matter of drawing a logical consequence from an initial axiom nor of generalizing a given experiment. Consequently, it appears that transcendental argument, thus defined, possesses a distinctive and original structure—in any case, it cannot be folded back into the modalities of reasoning of the form modus ponens. For all that, this argumentative structure cannot be relegated, as those who hold to a strict logicism would have it, to the group of propositions that are invalid because they are neither deductive nor inductive. Tertium datur [a third choice is possible], we could say, in that transcendental argument thus conceived is “totally new,” and, as Bitbol notes, its employment “could considerably alter our understanding of what constitutes physical theories”¹⁷ and, as a general rule, of what constitutes any system of propositions that claim the truth of what they say.

To complete my response to these objections, I must address the third criticism—is transcendental argumentation anything other than a simple machine for refuting opposing statements, such as skepticism, critique, strict positivism (scientism), etc.? Is it doomed to establishing a list of what cannot be said without ever indicating what must be said? This objection, by far the most formidable, requires that we tackle the other major aspect of transcendental argument’s fecundity.

First of all, I should note that if, in fact, transcendental argument is particularly illustrated by refutations like “you cannot say what you are saying,” these refutations cannot be considered to be only negative, since the work that delimits the set of statements that we cannot accept and the systems to which we cannot adhere is, in the end, constructive. For example, with this argument, we are now in a position to reject the two key stances of contemporary philosophy—skepticism and scientism. Given that contemporary philosophy is torn between a certain form of strict positivism (the cognitive sciences, naturalism, all the way to various materialisms)¹⁸ and an increasingly radical relativism (Feyerabend, Rorty), we can now propose a third, positive solution. Thus even if my argumentative practice—which consists in showing that a discourse x self-destructs—can appear to be only negative, its yields are no less considerable because, as we have seen, the currently most common and widely shared theories must be abandoned. The argumentation that takes the form “You cannot say that x” is heavy with positive propositions in that it clears the barriers that obstruct the horizon of thought today.

In addition, I can also show that transcendental argument can move from the proposition “You cannot say x” to another that, despite the double negative, is heavy with positive propositions, namely, “You cannot not say that x.” I have illustrated this passage from the negative to the positive in my works on Fichte, through (among other examples) the theory of affection (called the theory of stimulus), the theory of the relation between the finite and the infinite, and even through certain practical statements, even (and including) his most concrete views, such as his condemnation of the death penalty, which rests entirely on the employment of this kind of reasoning. The very production of a given proposition within a given system (in this case, Fichte’s or Hegel’s) attests to the fact that transcendental argument produces positive statements that take the form of “You cannot not say that x.” I should probably insist on this style of praxis or of the regulated production of statements, a general style of argumentation that is born in the distinctions of its starting point. I have noted that in Fichte, the starting point—namely, identity as congruence between what is said and what one does—is a task to be accomplished, an end to pursue, and not a psychological or factual given. I have also noted that the most immediate characteristic of the science of knowledge is that it is not defined as the description of an x but as the construction of itself. A construction because to reflect does not mean to reflect like a mirror or to give an image of something anterior but rather means to produce concepts and propositions as a task that one freely sets for oneself (the noncontradiction between the Tun and the Sagen). We thus can see how to argue is, stricto sensu, to produce something.

But to be sure, if the whole of Fichte’s system demonstrates that we can move from a negative statement of the sort “You cannot say x” to a positive statement of the form “You cannot not say that x,” then I must—beyond the production of this or that concrete proposition of Fichte’s—demonstrate this capacity with more contemporary examples.

Confronted with this challenge, I could be tempted first of all to retort to those who complain about the argumentation’s negative character that I have nevertheless demonstrated the necessity of positing the concept of truth. Indeed, the skeptic cannot say that truth does not exist without destroying the contents of this proposition. On the other hand, to say “truth exists” undergoes the test of self-application of a proposition to itself without difficulties (“truth exists” and “I claim that this proposition is true” do not mutually exclude each other, unlike in the previous case). It will probably be replied that the term “exists” does not mean anything here. To avoid this verb, full of misunderstandings, it suffices to say that the notion of truth cannot ever be consistently eliminated from our scientific or philosophical discourses. To put it positively, scientific and philosophical discourse cannot not posit the concept of truth—even if it cannot immediately determine its contents. (Is it an intuition and a concept, empirical verification, some sort of a priori, etc.?) We can clearly see that despite an apparently negative formula “to not be able not to posit that x,” the analytical gain is considerable. That said, one could retort that this proposition is too general to be useful. My law of reflexivity would, of course, be acceptable, and the idea of truth absolutely inescapable but unusable in concrete and specific contexts. In a word, the reflexive law could be accepted but would give rise to the question, What is it good for in concrete terms? To answer this question, I propose to address, as a case in point, a problem in the “philosophy of mind,” a problem that is taken for concrete since the problem is to know how our mind functions—the brain, or reason? I approached this quantitatively important aspect of current philosophy in part 1, and I recalled—in order to overcome them—Herbert Feigl’s physicalist thesis and (at greater length) Geoffrey Hodgson’s evolutionary view. In so doing, I thus brought up two of the foundational paradigms of contemporary philosophy of mind—physics and biology—but neglected the third paradigm, the sciences of the artificial. It is on this third paradigm that I would like to test the utility of transcendental argumentation that comes from the reflexive a priori. I shall do so by comparing this argument to what the founder of the paradigm of the sciences of the artificial, Herbert Simon, has to say about reason.¹⁹

First of all, I should recall that Simon’s field of investigation goes far beyond the single domain in which he has been honored.²⁰ Compared, because of his multiple scientific competencies, to Leonardo da Vinci,²¹ Simon undeniably embodies the anachronistic figure of the complete humanist who, having mastered his epoch’s fundamental fields of knowledge and having innovated within some of them, does not for all that disdain wondering about the definition of reason and of science. In this sense, Simon would probably have been described, in a different time, as a philosopher, in light of the guiding question that defined his domain of investigation. Indeed, he wrote in the preface to Reason in Human Affairs that the question of how we can understand individuals’ rationality is the very heart of all his reflections: “The nature of human reason—its mechanisms, its effects, and its consequences for the human condition—has been my central preoccupation for nearly fifty years.”²² Starting from the behavior of “economic agents” (firms, and then the individuals that compose them), Simon has in the course of his work expanded the scope of his investigations to encompass human reason in its entirety. His project is thus not only to determine agents’ possible behaviors in an economic context—that is, to grasp only one aspect among others of human behavior—but in fact, as is attested by the sentence I just quoted, also to discover the “mechanisms” of human rationality in general. This detail is significant in that it alone entitles me to examine Simon but not any economist or any sociologist or any other specialist in some aspect of humanity. Indeed, if the economist confines himself to only the domain of economic rationality while asserting that this rationality is not at work in other (moral, political, scientific) domains, then transcendental argument cannot be used, even indirectly, to analyze these propositions. To say something about the rationality of economic laws is not necessarily to understand all rationality, for example, the rationality of moral agents or scientific researchers. Different demands (investigation into self-interest, happiness, or the better argument) could preside over the behavior of an acting human. It follows that only the economist’s assertion that he indeed is speaking of “the nature of human reason” makes the comparison possible. Such is surely the case for Simon, who in a first approach defines human rationality as a “limited rationality.” How should we understand this first characterization?

Rationality, from a general point of view, means that human beings can, most of the time, give reasons for their decisions. Unlike animals, humans can enter into a process of justification for their behavior. As Simon writes, “By a weak definition of rationality, virtually all human behavior is rational. People usually have reasons for what they do, and if asked, can opine what these reasons are.”²³

Nevertheless, this rationality is limited, in that humans find themselves in a particular environment in which they do not control all the elements, thus they cannot calculate every possibility; moreover, they can delude themselves about what they can do or what they are:

Of course, as Freud (and laboratory experiments) have taught us, people may deceive themselves; the real reasons may be different from what people suppose they are . . .

To say that people have reasons for their actions means that there is a connection between the actions and the goals (values, utility functions) the actors possess. The actions increase the likelihood that some of these goals will be achieved. However, even in behavior that we would call rational, there may be serious gaps between action and goal achievement.²⁴

Individuals thus make decisions as a function of limited “goals or values,” on the basis of a series of finite inferences and observations, which could be false. Even if the inferences are false, the behavior would be called rational.

This definition of rationality could appear banal, in that it simply tells us that a human, unlike the Leibnizian God, cannot calculate, foresee, or know everything, and therefore that his choice of the “best possible” is not perfect but is necessarily limited by his finite capacities. This generic definition nevertheless has considerable importance at the narrow level of economic debate and particularly as a critique of market theories stemming from Léon Walras, all of which presuppose the “utopian hypothesis”²⁵ of “perfect rationality.” This very general definition is made more precise by what Simon terms “procedural rationality.” Here again, this definition must be understood in opposition to the neoclassical economic theories in which Homo economicus must ideally effect all the possible calculations to make the best choice. If this type of rationality, which Simon calls “substantively rational” (and which, again, amounts to the rationality of Leibniz’s God), is presupposed, then it is sufficient to know the environment (the possible worlds) to anticipate an agent’s choice. Conversely, for Simon, we should consider the procedures by which individuals make this or that decision. It only makes sense to revise the rationality at work in the decision-making procedure in situations when agents’ choices are difficult:

Briefly, given a particular end, humans employ a procedure to choose a situation that is most likely, in one’s eyes, to lead to that end. And yet for Simon, these decision procedures can be scientifically analyzed; they are predictable, even if our rationality is limited, our environment is complex, and we are likely to deceive ourselves.

Analysis of the environment is given as a response to stimuli:

For complex decisions, instead of automatic reactions (action-reaction, stimulus-response), a “deliberative procedure” must be employed. The rational agent investigates “alternative solutions” and adopts one of them, not because it is in itself the best but because it appears “satisficing,”²⁹ solutions that Simon defines as “solutions that are good even though they are not optimal.” In a word, it is a matter of subjectively noticeable solutions, even if they are not objectively and in themselves (from the point of view of a calculating God) the best.

But, it will be asked, how are we to analyze and predict—that is, to scientifically deal with—these decision procedures, which can initially seem to be numerous, or even infinite? Simon proposes to do so beginning with artificial intelligence, a domain on which his innovations have incontestably left a mark.

The mind, like the computer, can be conceived as a stock of symbols that can be combined with one another. The symbols are “physical patterns (e.g., chalk marks on a blackboard) that can occur as components of symbol structures (sometimes called ‘expressions’).”³⁰ Thus, a well-stocked symbol structure can be brought into relation with other symbol structures; it can also be transformed in order to create structures more appropriate for the environmental stimuli. These symbols are “physical symbols” in that they have a physical substrate, the brain—computers and brains are thus equivalents, because the physical substrates are “fabricated of glass and metal (computers) or flesh and blood (brains).”³¹ Despite their differing substrates, the computer and the brain thus have the same skills at their disposal:

Having established this equivalence, the computer can serve both as a norm—in the sense that certain programs can improve decisions (for example, expert systems that function as assistants)—but also as an image, a representation, a photograph of human intelligence. Artificial intelligence thus enables a description of the ways in which individuals make decisions:

How can this definition of human rationality and its comparison with computers be of interest to transcendental argumentation? Because it cannot be brought to completion without contradiction, and because the demonstration of this impossibility enables me to posit another definition of human rationality. Let’s look at these two points.

First of all, I should generally say that three parameters are clearly to be considered in Simon’s discourse; these are the economist’s discourse on agents’ rationality, the contents assigned to agents’ rationality, and reality (even if this reality is understood as a set of rules or conventions). These three parameters multiply the economist’s tasks: On the one hand, he must state what reality is (for the scientist)—is it a set of arbitrary conventions or a set of immutable and necessary laws? This is the classic question of realism or of reference. On the other hand, he must define individuals’ rationality (limited, unlimited, determined or undetermined, a pure calculus of self-interest, etc.) Also, the economist must determine how the economic agent understands or represents reality (because this understanding is not necessarily the same as the economist’s). Finally he must try to understand the relation between the rationality that he employs qua scientist and the rationality that he attributes to the individual or the economic agent. If the question of “human reason” is Simon’s “central problem,” he must not define this reason as if his own scientific discourse were not also an expression of it. Here we confront the problem of self-reference.

In view of these tasks (which I can specify without having to get into the details of analysis, nor having to test the proposed mathematical models’ falsifiability), what can we say a priori? That human rationality and the rationality employed by the economist must be congruent. The economist’s rationality, through its very employment, must not enter into a contradiction with what he says about human rationality. In a word, the scientific (economic but also sociological or anthropological) discourse must make sure not to invalidate itself by attributing a rationality to humans that its own discourse contradicts. Here, indeed, we find my law of reflexivity, which can now be defined as a reflexive a priori because we can say before any experience of a theory to be analyzed that the status of its utterance and the contents of its statements must not enter into conflict.

And yet if we apply this point to Simon, it seems that human rationality is entirely determined, that is to say, predictable, reducible to mechanisms that a computer could accomplish. Alain Boyer underscores this point, writing about Simon’s conception that “the ‘self satisfying’ agent seems unable to be anything but an automaton following rules or routines, almost incapable of any critical and reflexive look at the routines in question.”³⁴ This determinist conception is also noted by John Searle:

I can now demonstrate that this determinism is unsustainable in this form. If we envision rationality in terms of routines or automatic functioning, of a stock of symbols that combine together according to precise algorithms that we can reconstruct, if we are thus able to predict future combinations, then the scientific discourse must understand itself in the same way. But Simon never claims, when speaking of science, that it is a sequence of routines or a regulated combination of stockpiled structures, and even if he were to do so, it would make his discourse and his activity futile. As a pure sequence of routines, science could not claim to be anything other than a mechanical repetition, always starting over. From this, I can say that the act of reconstructing the scientist’s rationality is a necessary exception to his theory of rationality—Simon’s theory of rationality is not universal because it cannot be applied to its inventor. Because scientific rationality, through its very employment, goes beyond what it says about rationality, it is self-contradictory. And yet—a significant fact—Simon himself recognizes that he cannot explain the rationality at work in the sciences and in particular in the discovery process. He forcefully rejects Karl Popper’s view that discovery is, in the final analysis, irrational but admits that science has not yet achieved a demonstration of the mechanisms that preside over the process of creating new representations. He even speaks of a missing link with respect to the process of creating new representations: “The process of discovering new representations is a major missing link in our theories of thinking and is currently a major area of research in cognitive psychology and artificial intelligence.”³⁶

It will be said that Simon answers my question here in arguing that we will be able to account for this process in the future. But—apart from the character of this kind of proof by “induction from the future”³⁷ that I have already discredited—it turns out that the model he uses (of computation as a sequence of routines or a regulated combination of stockpiled structures) normally makes the discovery of something new impossible! With his model, Simon prohibits innovation and, at the same time and in the same way, states that a later groundbreaking discovery will demonstrate it. Paradoxically, Simon does not ask future science to confirm his hypothesis but rather, indeed, to demonstrate its falsehood. As it happens, this is not the only “remainder” that Simon acknowledges in his theory. The problem of determining ends and values also remains in suspense. If rationality is a procedure for achieving an end, who decides the contents of the end—for example, Nazism or perpetual peace between nations? However, I am not emphasizing this drawback here but rather the contradiction at work in the discourse of justification.

My analyses’ conclusions are not merely negative, even if they arise from a specific challenge to the contents of a theory x, in this case, a theory in which human thought is reducible to a computer’s functions. These conclusions include an immanently positive side, which I can now present.

1. I can establish that determinism is, in certain frameworks—in this case, of hypotheses concerning the definition of human rationality—impossible to articulate. Determinism is not only, as Kant would have it in the Critique of Judgment, infinitely improbable from the perspective of certain kinds of phenomena, it is false and this falsity is demonstrable through argumentation. Here we see the difference between this system, which enables a demonstration of determinism’s falsehood, and the current discourse, more or less coming out of the Critique of Judgment (a discourse that I’ve already analyzed), which denounces determinism in the “social sciences” either from “everything happens as if,” that is, from a heuristic hypothesis, whose internal contradiction I have already shown, or from a moralizing viewpoint (determinism—in favor of individual self-interest, etc.—must not be accepted, because it is not good), a viewpoint that Christiane Chauviré has quite effectively criticized in her article “Pourquoi moraliser les normes cognitives?”³⁸ Between the relativism of “everything happens as if” and the moralism of “that ought not be so,” another route remains open for the philosopher: demonstrating that a proposition is quite simply false.

2. Apart from the impossibility of articulating a total determinism, we can also easily show, from the reflexive a priori, why other theories about “human rationality” are impossible: to thus reduce human rationality to a pure calculation of a strictly individual self-interest (a case frequently assumed in economics) amounts to a simultaneous claim that the scientific discourse that maintains this proposition is also the product of pure self-interest—which means that there are no general values, and we can neither discuss nor argue with someone who implicitly does nothing but express a preference or individual self-interest. Briefly, there are many frameworks and theses that can be analyzed from the perspective that I have called the reflexive a priori.

3. Furthermore, these considerations make it possible to outline several concrete positions: thus any political discourse that aligns itself with an ostensibly immutable necessity (divine right, or the necessary laws of economics) or with an ostensibly necessary rationality (like a pure calculation of individual self-interest or an accumulation of routines and automatic functions) can be rejected. We see here that the reflexive a priori, far from yielding no general propositions, leads to a concrete stance, founded in reason. This stance is adopted through consideration of what can be said and what is impossible to say without contradiction or self-refutation. I have demonstrated this for the hypothesis of an entirely determined rationality; I have also shown it to be the case for a conception of rationality as a pure calculation of self-interest. These assertions are not without concrete consequences for an understanding of how a society, a business, an institution, or a state is organized. Consequently, there are many varied fields of application for this general epistemological thesis.

4. At the end of this analysis, I can also propose a schema (an interpretive grid) to enable a consistency test for a hypothesis or a discourse in the human sciences or the sciences of mind, such as Simon’s thesis about human rationality. Indeed, three factors are to be considered: the economist’s discourse, the contents or the form of rationality attributed to agents, and reality, whether this latter be understood as a system of arbitrary rules or conventions (according to a more or less Wittgensteinian thematization) or as a series of immutable laws (according to a classically positivist or even naturalist thematization). If we return to Simon’s theory, for example, we obtain the following concrete schema:

Two kinds of lacunae are immediately recognizable in this schema: The first—which I could show (with the same schema) to apply to many “concrete” thinkers—is to maintain that reality is a proof of one’s hypotheses’ validity (this is why Simon holds that his hypothesis is more viable than Milton Friedman’s or an Aztec myth, which Paul Feyerabend could say does the job just as well), even though one also claims that this reality is not knowable. This first lacuna corresponds to the classic problem of reference. The second is clearly the contradiction that appears between the definition of “human rationality” (Simon’s very ambition, in his own terms) and the use of a scientist’s rationality. This lacuna corresponds to the problem of self-reference—and the failure to consider this problem has led to the serious crisis situation that I analyzed in part 1.

And yet we see that this position is rich with concrete views, with positions on particular propositions in particular human sciences, without which the philosopher would necessarily have to transform himself into an economist and become a rigid specialist, in Nietzsche’s sense of an “expert” concerned with only one-thousandth part of the leech’s brain. In my view, the philosopher can determine which general discourses cannot be maintained in a logical and consistent manner. Having said that, I will certainly be accused of putting philosophy in a position “above and beyond,” so that philosophy presides a priori, as queen of the sciences, over all the sciences with a uniform incompetence. This is why, in order to explain my point of view, I should specify things with Simon as my example. Economics, to put it crudely,³⁹ has had two great phases or directions: In the first, at its moment of birth and modeled after physics, economics is understood as the identification of natural laws. In the second, born from the aporias of the first, economics, as well as identifying the laws of reality (physical laws or systems of conventions), must wonder about agents’ rationality (how do individuals act and interact with one another?). It follows, on the one hand, that, like physics, economics raises general epistemological problems (hypotheses’ realism or antirealism)—debates that philosophers can contribute to without for all that claiming a position “above and beyond”; on the other hand, in its latest developments, economics particularly tends toward the problem of the nature of rationality (is reason an appropriate response to a set of exterior stimuli? is it a simple calculation from individual self-interest? a product of genes? adaptation to an environment?). Here, too, philosophers can have something to say in this debate, without “arrogance” (see, on this point, Popper’s discussions of particular human sciences’—for example, psychoanalysis’s—status in light of his epistemology). From reflexive a priori as the congruence between a statement and its utterance, philosophers can show how a given thesis x about rationality cannot be said. That philosophy can converse with this or that science without for all that becoming one among them is thus clearly demonstrated, and its fecundity is thereby shown.


1 That p.	“I perceive an object.”
2. It would not be possible that p if we did not think that q.	“I cannot perceive an object without conceiving it as distinct from myself.”
3. We must think that q.	“It is necessary to conceive the existence of objects outside myself.”
4. It is true or it is necessary that q.	“Objects outside myself exist.”