Philosophie contre intelligence artificielle

The main interest of artificial intelligence (AI), whatever its result will be, is to sum up and to embody history of philosophy. AI is even the very last place in which these problems are explicitly and pragmatically discussed. Any philosopher can be challenged not to see the application of old philosophical debates, if he reads works on it. I even venture to consider AI, not only as the outcome of philosophy, but as its realization. Even its detractors admit this characteristic of continuation of classic philosophy (Searle, preface to the French edition, p. 3), via Heidegger (Dreyfus, p. 72), or Fodor (Winograd-Flores, p. 109).

Actually, there is no use repeating ritually that philosophy has abandoned the question of meaning, or the question of being. This traditional philosophy is equivalent here to the metaphysical state, according to the French sociologist Auguste Comte’s classification, in which the question lies in these ontological expired terms. On the other hand, it is easy to observe that philosophers have abandoned all other ambition that this compulsive ritualism, as Searle indirectly admit it too (p. 13).

The failure of philosophy under its metaphysical form, either classic or renovated, leads philosophers to claim that the efforts on these matters are vain. It cannot be otherwise when interferences of any alien to the sect are rejected, like these of AI research worker, Marvin Minsky. For Dreyfus (pp. 35-36) the Minsky’s knowledge engineering problem seems amount to the impossibility to realize the program of Husserl, yet in continuity of two thousand years of philosophy. Philosophical critics of AI do not only reject partial results of computer fields. Cumulating of some of partial results from its own philosophical tradition is also rejected, from Greeks to moderns like Hobbes, Leibniz, and Boole, or analytic philosophy contemporaries. Actually, many philosophers are ancestors of AI. Phenomenology’s ones cannot monopolize the title of philosopher except by arbitrary and falsifying selection. Like on official Stalinian photographs, this kind of process can gum whole sections of intellectual history. Hermeneutics in question then becomes simply a submission to a master whom any word becomes source of infinite interpretations. When the (knowledge engineering/philosophical) problem is to include any knowledge.

What is the present-days philosophical practice (except the moralizing comment of the news)? Its only positive side is essentially observed in recognizing correct erudite attributions. This erudition seems the only break with common sense. But what indeed could be a preliminary to elimination of archaisms rapidly becoming Byzantine quarrels on what Plato, Descartes, Marx, Heidegger, have truly said. The interpretive nature of all philosophical issues characterizes this practice in a self-centered way. Philosophy has to be objected the fact that its methods consist in accumulating references, without selecting them. Obviously, paraphrase presupposes a real work. But it rapidly turns into mimetic assimilation of the discourse or mannerism of the author. Here, the topic is a kind of capacity of self-possession. I have been able to confirm this hypothesis when I could observe that a forsaken author, put back on schedule of high-level examination (agrégation) for teachers, was first approached with a mocking patronizing attitude, and finally was toadily revered by the instructors. This referencialist nature is shown by this classic Ashby’s howler in the AI history, as Dreyfus reminds us very exactly:

“’Gelernter’s theorem proving program has discovered a new proof of the pons asinorum that demands no construction’. This proof, Dr. Ashby goes on to says, is one which ‘the greatest mathematicians of 2000 years have failed to notice […] [NOTE 8].’ The theorem sounds important and the naive reader cannot help sharing Ashby’s enthusiasm. A little research, however, reveals that the pons asinorum, or ass’s bridge, is the elementary theorem proved in Euclidean philosophy […and…] the first announcement of the ‘new’ proof ‘discovered’ by the machine is attributed to Pappus (AD 300).” (Dreyfus, p. 82).

This howler is obviously grounded on lack of methodical recourse to epistemological erudition by formal scientists. The philosopher gets a good deal, thanks to his referencialist specialization. But this specialty is also the origin of its classic professional bias claiming “Everything is already said.” Thus, we can actually observe, in AI, a tendency to resume everything each time from the starting point –, which is however a bit natural for a new field. While the philosophers’ methodological automatism explicitly demands this compilation of the comprehensive past works (Dreyfus, p. 100), with the risk to never produce any new results.

Anyway, the historical method is not so genuine in research work, because it is difficult to criticize in real time (before Internet) unknown works, or even secret – due to private research and intellectual or industrial property. Philosophy, on the contrary, usually discusses works already selected by history, and come into the public domain. Besides, it is necessary to notice that philosophical or erudite hermeneutics consists too in justifying all the erring ways of classic philosophical authors in a more than lax way. Thus, a few philosophers pitifully dishonor themselves each time the Heidegger’s case is being discussed again. A same indulgence (what sociologist Paul Lazarsfeld calls a “generous critique”) could be thus expected, especially for the case of under way works, like these of AI!

This admissible methodological function of accrued references management can then regress constantly to dogmatism, often by abuse of authority. The term philosophy itself and its derivations are themselves synonyms with this standing back imposture. This philosophical habit can only be a temporization when one cannot keep up (everything goes so rapidly nowadays).

The opposition of philosophy is simply understandable too in terms of refusal of AI treading on its questioning territory. Wondering about “What thought will think AI if it set itself as thought of thought?” (Daniel Andler “Preface,” in Dreyfus’ French edition, p. XIII) forgets to ask the same question for philosophy, whose meta-physic ambition is well known, but not as self-questioning! The pretension of philosophers to inform real (on the Aristotelian paraphrasic mode), and to tell what it ought to be, is notorious. But if their field is not grounded on examination of internal and external constraints, it becomes, literally, a utopia, designed in normative terms, in the totalitarian lineage. This pretension of philosophers to rule over sciences and to decree impossibilities is frankly as tiresome as grotesque. This normative philosophical idea naturally regresses to ground all programs authorization on religion, or on God himself. The amateur philosopher Weizenbaum seems to regret (p. 12) the time when the place human’s thought in the universe was grounded on a contract negotiation between God and human beings [NOTE 9].

The analysis refusal allows the philosopher to distinguish perception or human conceptualization from computer’s ones, which would be the result of traditional philosophy assumptions thinking in term of so called “atomistic concepts” (Preface of Anthony G. Oettinger, in Dreyfus’ French edition, p. XX). The operation is as classic as phoney. Besides, the term atomistic concept is a reconstruction of history, which has nothing to do with anything except a tautology: each concept is atomistic. The alternative would be that a concept would refer everything and its contrary.

The simplest traditional version of this philosophical refusal of analysis simply consists in refusing of determinings. For the post-theological metaphysics, only occupied with what is absolute, any determined item becomes a-philosophical by definition. A version of this refusal of determinings consists in applying a critical scope every times something is determined, according to a kind of mechanism thus formulated: “I understand it, therefore I’m against it,” leftist version of Franco’s principle during the Spanish war: “abajo la inteligencia! [NOTE 10].”

One of the keys of this anti-analytic standpoint is the fixing on the fetishist rejection of the microprocessor’s binary functioning, or on the fact that AI researches use ‘digital computers’ (cf. Dreyfus, p. 155). This analytic mode is opposed to mysterious characteristics of human mind:

“Patterns as complex as artistic style and the human face […] seems to require a special combination of insight, fringe consciousness, and ambiguity tolerance beyond the reach of digital machines.” (Dreyfus, p. 120).

But this dogmatic hate of the elementary stuff mixes up the digital and the symbolic. For a long time, and AI is only the ultimate proof, computers also process symbols. Binary digit (bit) does not mean more 0/1 than yes/no, or true/false. It is a simple two states code, which value is not either arbitrary than an analogue coding. Anyway, bytes are an eight bits code usually meaning letters (a, b, c…) and other signs (?, $, @, =, &…) what is only a way to write them, like in Morse, graphic variations, or other languages (and the other sounds). Besides, LISP programming language, specific to AI and processing lists of symbols, has been invented almost at the same time (1958/1954) than FORTRAN, specialized in calculus.

Dreyfus even develops in many occasions, in this anti-analytic standpoint, an apology of the analogue processing, against binary coding (pp. 160, 165-166, 187…). He agrees here Bergson’s speculations – which seem to come back in France, most possibly for the same reasons. R. Moreau, in his book (Ainsi naquit l’informatique [The Computer Comes of Age], p. 20), note that this track has been abandoned since Babbage. According to it, despite some specific usefulness (parallel calculation, complex simulation), the analogue processing does not represent a universal machine, does not provide the same precision, and does not memorize data.

We have to notice that the same refusal of binary explains the initial resistance to numeric compact discs that, today, replace for good analogue vinyl records. The discourse about the metallic sound and other dehumanizing word make wonder about the analogue humanity of the slide rule or of a sinusoid. This idea is obviously absurd. Complaints of music lovers can concern on the contrary the too good reproduction of the sound, obtained by the recovery of the original recordings. While vinyl records compensated the too large sound gaps because of specific difficulties of engraving. Therefore, it was the analogue sound that was artificial, to correct listening constraints.

Holism is the new name of the philosophical pretension to speak for the sake of totality. Philosophy should nevertheless have thrown over these Hegelian synthetic ambitions. But the weight of the tradition is certainly too strong. Unless it is simply difficult to do without this hardly secularized theological valorization.

Writers, philosophical amateurs, and even DP men or computer scientists have adopted this analysis refusal, sublimated in holism. Thus, Weizenbaum invoke Huxley’s spirit when he ventures in questioning operative validity and analytic method of science, for the sake of a holistic profession of faith as trivial as extremist (Weizenbaum, pp. 127-128). Huxley, keen amateur of parapsychology, shown, like his colleagues, a heavy concern in scientifically grounding observations or theorization of these parapsychological manifestations. Parapsychologists too never lack claiming titles, statuses, or university publications [NOTE 11].

Thus, holism of current phenomenology especially works on some Gestaltist clichés, still impressive for sensitive souls. It is necessary not to ignore these references, as Newell and Simon did imprudently by neglecting the overall approach of the Gestalt psychology, on pain of Dreyfus’ sarcasms (p. 320, note 34). But what is the holists’ use of the Gestalt theory? It opposes to data processing in general, and especially to AI, the impossibility of programming every feature considered as essentially human. For this purpose, they untiringly remind the same worn-out few curiosities, against the analytic capability:

Traditionally, the phenomenological idea is grounded on a generalization of perceptive behavior by the Gestalt theory, which actually has brought the totality phenomenon to light. But human vision is still grounded on elementary signal encoding, breaking down, even if there is an ulterior recomposition. This process is equivalent to a plain television way of work. Television whose picture emphasizes the problem raised to philosophy by cinema and its movement breaking down, very root of Bergson’s philosophy holistic glosses.

The phenomenological discourse wonders on the tritely classic mystery of not seeing hidden faces of objects (like a cube), nevertheless “once we have experienced these further aspects, they will be experienced as copresent” (Dreyfus, p. 241). This mystical presentation is open to misinterpretation. Perception does not inform us, indeed, at any time, on all the characteristics of any object we see in front of us. It is very obviously the characteristics of visible faces, and general, physic, logic rules, which allow us to make deductions. More, besides the fact that memorization is accessible to computer this kind of reconstruction is easily programmable. Nowadays, object programming allows the properties inheritance: i. e. cat has feline’s characteristics and unclear features, otherwise known, or deduced by analogy. But it is always necessary to confirm deductions by observation, on pain of excessive generalization or dogmatism – which are not even necessary to consider as specific to philosophy.

The area of chess game gives an appropriate ground to insinuate the more or less Zen theory of total vision that allows experiencing a brutal flash of understanding. The chess grand master, thanks to these capacities, can thus possess an overall vision that gives him abruptly the solution (Dreyfus, p. 102). It seems phenomenological to receive an insight when “zeroing in on this area he discovers the unprotected rook” (Dreyfus, p. 105). A canonical Gestaltist interpretation could have rather spoken about typical configurations, recognized as such, or anticipated, what prevents us from programming heuristics (empirical knacks), generated by learning or by frequency analysis for instance? But, even in the case of an overall configuration, significant elementary conditions are necessary. Besides, anyone knows the systematic studies analyzed by grand masters, and their team of counselors. How is it possible to do otherwise, especially in chess? The holistic/Gestaltist myth is merely an updating of the explanation by magic. It is also necessary to note that genius chess players or high level sportsmen generally succumb to sirens of parapsychology, grigris, and other suckers’ traps.

The holistic assumption (“The whole is superior to the sum of parts”) is also a matter of cliché. At least, it has allowed at Marxists, who had a tendency to overuse it, an easy retraining to phenomenology. Former materialists, they end up in idealistic ranks! This thinking thus allows a negation of the possibility of any specific computer program – what very seems in contradiction with the yet phenomenological locality principle: “Each [of this programs] is a small engineering triumph an ad hoc solution of a specific problem, without general applicability.” (Dreyfus, p. 46). This locality principle tells us nevertheless that, even if one perceives totalities, one does not live in the Large Whole. It seems this idea arise out of the solipsist idea of a pure mind. Body, beloved of phenomenology is yet situated in an analyzable and differentiated environment, in space, time, acts, tastes, etc. The holistic regression, stemmed from traditional philosophy, has a hard life. If individual differences did not exist, and if transmission of thought existed, partisans and AI adversaries would commune in a same wholistic knowledge (like the Theillard de Chardin’s noosphere). And it would not be necessary to write books.

A processing option of the part and the whole consists in saying that the whole gives the parts their meaning. But, the fact that an element derives his meaning from totality does not mean that totality does not do the same from its elements (it is only a matter of personal preference). Elsewhere, this bias causes/induces the classics puzzles, in a parody of system analysis. Each element of a theory depends on “the truth of the whole theory” (Weizenbaum, p. 141). With this system, no concept could have ever been defined, but their whole theory could. Does it mean that an overall theory can be absolutely isolated itself, what relativize its wholeness, making it part of a vaster theory (cf. “a partially explored territory,” idem p. 142)? How can it be established with this lack of comprehensiveness?

This dependence of the element from the whole can be specified by the mechanism of relations, reminiscence of the Marxist vulgate, or of oriental mythologies. Indeed, students attentive to philosopher certainly are speechless in front of such Zen principles, like “in the case of the rhythm […] there is no such thing as a syncopated beat, for instance, existing all by itself.” (Dreyfus, p. 313). [NOTE 12]

But this Gestaltist dogmatic idea seems to deny the element that it mentions. It is obvious rhythm is composed of elements. If relations determine rhythm, this characteristic is shared with numbers, or words, both depending on semantic systems. But all of them are analyzable. Besides, relationships themselves are identifiable as elements.

In its opposition at current science, holism develops, like any sect, something like a strong feeling of persecution as much more developed the faithful is sure to detain the truth:

“Such an [holist] alternative view has many hurdles to overcome. […] Western though has already committed itself to what would count as an explanation of human behavior […] according to universal laws or rules.” (Dreyfus, p. 232).

The search for an alternative to western sciences does not prevent from resentment not to benefit from its stamp. This western scientific thought has nevertheless taken time to elaborate the thought, which is denied here. It is in that sense that this critique rather seems to get back to Rousseauism, to Third-Worldism (by Marxist tropism), or to archaism. This characterizes the universality of western thought is precisely the fact that it does not debar by rights [NOTE 13]. If this process could emphasize ethnocentrically its own influence, it is solely due to archaic residues. Either perfection cannot be demanded, or speaking from an external standpoint, while denying this universality.

This holistic clichés could simply be admitted as equivalent to speech mannerisms with no consequences, if they wouldn’t enter in a permanent exaggeration, emphasized by mutual-quotation. Dreyfus shows an (astonishing of delirium) holistic example, which “is worth quoting in full,” as he said himself! Rudolph Arnheim, professor of psychology at Harvard, speak about an Evans’ attempt to make solve a test by a computer concerning simple figures (typical IQ test in which the figure below has to be found among three possibilities). He also opposes the machine to the human processing way in this kind of questions:

“What happens when a person is confronted with a figure such as Figure [3]? The reaction will vary somewhat from individual to individual as long as no particular context calls for concentration on specific structural features. By and large, however, the observer is likely to notice a vertical arrangement, made up of two units, of which the upper is larger and more complex than the lower; he may also notice a difference in shape. In other words, he will perceive qualitative characteristics of placement, relative size, shape; whereas he is unlikely to notice much of the metric properties from which the computer’s reading of the pattern must set out, namely, absolute size and the various lengths and distances by which this individual figure is constructed. If one asks observers to copy such a figure, their drawings will show concentration on the topological characteristics and neglect of specific measurements.

Confronted now with a pairing of A and B, the human observer may have a rather rich and dazzling experience. He may see, at first, fleeting, elusive resemblances among basically different patterns. The overall figure, made up of the pairing of the two, may look unstable, ungraspable, irrational. There are two vertical arrangements, combining in a sort of symmetry; but these two columns are crossed and interfered with by diagonal relations between the two ‘filled’ large circles and the two smaller, unfilled shapes. The various structural features do not add up to a unified, stable, understandable whole. Suddenly, however, the observer may be struck by the simple rectangular arrangement of the four smaller figures: two equal circles on top, two equal squares at the bottom. As soon as this group becomes the dominant theme or structural skeleton of the whole, the remainder – the two large circles – joins the basic pattern as a secondary, diagonal embellishment. A structural hierarchy has been established. Now the double figure is stable, surveyable, understandable, and therefore ready for comparison with other figures. A first act of problem solving has taken place.

If the observer turns to Figure C, his view of this new pattern is determined from the outset by his preceding concern with A and B. Perceived from the viewpoint of A, C reveals a similar vertical structure, distinguished from A mainly by a secondary contrast of shapes. The family resemblance is great, the relation comes easily. But if C is now paired with D1 the resemblance looks excessive, the symmetry too complete. On the contrary, a comparison with D2 offers too little resemblance. D3 is recognized immediately as the correct partner, the missing fourth element of the analogy, if the relation between A and B had been properly grasped before.

This episode of perceptual problem solving has all the aspects of genuine thinking: the challenge, the productive confusion, the promising leads, the partial solutions, the disturbing contradictions, the flash appearance of a stable solution whose adequacy is self evident, the structural changes brought about by the pressure of changing total situations, the resemblance discovered among different patterns. It is, in a small way, an exhilarating experience, worthy of a creature endowed with reason; and when the solution has been found, there is a sense of distension, of pleasure, of rest.

None of this is true for the computer, not because it is without consciousness, but because it proceeds in a fundamentally different fashion. We are shocked to learn that in order to make the machine solve the analogy problem the experimenter ‘had to develop what is certainly one of the most complex program ever written.’ For us, the problem is not hard; it is accessible to the brain of a young student. The reason for the difference is that the task calls for the handling of topological relations, which require the neglect of purely metric ones. The brain is geared to precisely such topographical features because they inform us of the typical character of things, rather than of their particular measurements.” [NOTE 14] (in Dreyfus, pp. 140-142).

This gloss obviously looks like the exercise of a literary competence [NOTE 15], even to fine art esthetic criticism, that allows all divagations. It looks a lot like a personality test – what give an idea of their interest –, in contrast with an IQ test, which was the example studied. In this very example, the conclusion is grounded on the obviousness of the solution that alone allows the immediate overall vision. But, it is likely that the simplicity of this configuration has been chosen for the presentation of results of the program, and no for the step, formally identical with figures that are more ambiguous.

Besides, it is possible to represent the problem in more comprehensible terms: being given a figure A (small circle included in a larger one, small square under the two circles), and a transformed figure B (small circle, small square underneath included in the large circle), find the figure D which is transformed from the figure C (small square in a large triangle up, small circle down), among D1 (small square included in a large triangle up, small circle down), D2 (large triangle up, small circle down, small square at the right of the triangle), or D3 (small square up, small circle included in a large triangle down).

Furthermore, IQ tests are known to choose this figurative strategy for cultural neutrality concerns, contrary to one of the anti-IQ’s arguments (used by Weizenbaum). Therefore, what should have been favoring esthetes, usually figurative rather than formal, seems to encourage them to delirium. Obviously, this mannered way to consider an IQ test will produce catastrophic results. This can explain difficulties of some subjects possessing this literary or esthetics cast of mind that would make them dream about presented figures. I have to confess I am contemplating henceforth relativizing my excellent results at this kind of diversions. Albeit…, it rather seems to me than a discourse like the latter come under a cognitive pathology.

How more specifically understand this holistic delirium? It seems that all the argument of professor Arnheim is distorted by the obviousness of his discriminating human/computer will:

i) The abuse of terms with blurred pretensions for human beings (“by and large; fleeting, elusive resemblances; look unstable, ungraspable, irrational; sort of symmetry; family resemblance”), while it concerns simple geometrical figures! Who believes human being doesn’t immediately recognize a circle, a triangle and a square?

ii) The computer perceived as analyzing “absolute size and the various lengths and distances” for its pattern recognition (possibly from measures, certainly). It rather has to analyze angles or proportions, and no the raw measurements. As I make it above, these forms can just as well be described, because the initial problem really is, here, analogy between two groups of figures! The problem is not the recognition of circles, squares, and triangles, but that of their layout (and its transformation). The size can therefore be expressed qualitatively (large-circle, small-square, etc.).

iii) The overall terms abuse characterizing the holistic idea, supposed human (“vertical arrangement ; overall figure ; the dominant theme or structural skeleton of the whole ; unified, stable, understandable whole ; basic pattern ; contrast of shapes”). The author reaches the caricature by telling that between C and D1: “the resemblance looks excessive, the symmetry too complete” only to say that it happens to be the same configuration!

iv) The ad hoc paraphrase of the Gestaltist understanding flash would rather stay to demonstrate (“rather rich and dazzling experience; Suddenly, however, the observer may be struck by the simple rectangular arrangement; the relation comes easily; D3 is recognized immediately; the flash appearance of a stable solution whose adequacy is self evident”). This flash consists here in recognizing that the big figure is lowered between A and B, and that the operation has to be reproduced between C and Dx.

v) The incomprehensible opposition between human processing quality (“This episode […] has all the aspects of genuine thinking; It is, in a small way, an exhilarating experience, worthy of a creature endowed with reason”) and the absolutely contradictory defamation concerning the computer (“the problem is not hard; it is accessible to the brain of a young student”). This inexplicable insincerity reveals a trait that does not honor the human side.

If Dreyfus hasn’t made a mistake while bringing this text, parodic of his own standpoint, it is difficult to conceive the internal logic of a similar reasoning. The most indulgent criticism could be to consider it is about fetishism for explanations in holistic terms. Arnheim’s reaction could have been a kind of exorcism of the diametrically opposite Bobrow’s approach commenting the GPS (General Problem Solver) program processing. The mechanism of (holistic) belief would then simply have consisted in strategies of avoidance of some taboo words, whose Bobrow’s transgression had here exceeded the crisis point:

“GPS treats sub-objects through its goal-subgoal organization. That is, GPS avoid looking at structures on a given level by decomposing them into smaller structures tied to subgoals.” (Dreyfus, p. 138)

For such a rate, in only two sentences, of terms that can be perceived as an anti-holistic provocation, an unbewitchment operation is supposedly necessary!

What means holism is well understood: the whole is different from the sum of parts. But the problem is – since always – how to apply a so much general slogan to real situations. Does it bring anything else than simple recognition signs between supporters? A practical example shows the limitations of an overall concept, especially in human sciences context:

“The overall population evolution doesn’t teach very much on local population: for instance, during the last thirty years, while the total population of France increases 20%, this growth has exceeded 100% in more than a thousand townships, […] and in more than 10.000 townships, it decreases one quarter on average.” (Hervé Le Bras, “How the population evolves in France? Urban filaments?,” Interviewed by Guitta Pessis-Pasternak, Liberation, Wednesday 1993 April 14th, p. 34).

Here, the reference to the Whole is thus not superior to the sum of parts, contrary to what was heard in the sixties. It even concerns, in our example, a loss of information. The hypothesis that this holistic idea is a paying back the investment made in Marxism is borne out again. It elsewhere happens that this term holism purely and simply replaces totalitarianism that has obviously been discredited by Stalinism. This represents convergent evidences to signal the origin of Hegelian lexical influence. And the phenomenological regression stay idealistic or pre-Marxist, because, as would say the master, philosophers live in a wonderful world, out of the necessity of technical manufacturing and commercial distribution of things surrounding him. He just observe the result of this organization:

“Human beings are somehow already situated in such a way that what they need in order to cope with things is distributed around them where they need it, not packed away like a trunk full of objects, or even carefully indexed in a filing cabinet.” (Dreyfus, p. 260).

On the contrary, the real world is organized (thanks to computers). And things are labeled on supermarket shelves – happily for us. This situation can be called ecosystemic principle, because it is obvious that the alive individual is found inevitably in a habitat allowing his surviving. And this situation is the result of an analytic elaboration in a modern society, what can be called supermarket principle.

This effective lack of consideration of numerous material constraints assumed by others represents, we can imagine, the imaginary universe of the womanly novel ideal of the 19th century. These situations could be considered as real, although limited to the elite. But we hope these fantasies will soon disappear from literature, thanks to the equality of women by the work (women who still are under a residual identity crisis on these questions). Let’s note for now this situation is equivalent to a Scarlett O’Hara’s world, surrounded by slaves, who could have gone to the supermarket instead of her.

This Heideggerian’s readiness-to-hand idea is accompanied by the bizarre idea of refusal of representation, which seems to reduce human condition to animal one: “You are thrown on what people loosely call your ‘instincts,’ dealing with whatever come up.” (Winograd & Flores, p. 34). The purpose of the operation is simply to exclude reason, and its traditional manifestations in philosophy. The pretext seems to be a praxis again allowing Marxists’ retraining. But why not to admit that action is itself a distinction [NOTE 16]? What can mean that the representation allows action, or accompanies it? Elsewhere, if: “In driving a nail with a hammer […] I need not make use of any explicit representation of the hammer” (idem, p. 33), you’d better have an explicit representation of the distinction between the nail and your fingers.

This ecosystemic availability of the “readiness-to-hand” world, that is obviously technical in our modern societies, has also to be considered as an elaboration, even in older/former societies. More generally, this condition is the anthropological fate, and the foundation of nature/culture opposition. That could have been obvious to who is imbued with “human world.” But the holistic romantic tradition is contradictorily over-valuing the idea of nature. In social sciences, the treatment of these questions is equivalent to functionalism, solution produced by the anthropologist:

The philosophical paradise, before the Fall, rather characterizes the child, who evolves in a protected environment, reminded by Weizenbaum (p. 211). But the small human has to learn rapidly how to be autonomous and how to master his environment, indeed with his body, but also explicitly and consciously. Especially, he has to organize his contacts with others, and to read labels in supermarkets.

Actually, the holism principle looks simply like organicism, which conceived the society as a whole without malfunctions. It does is evident that an organism is different, even superior, to the sum of its organs. Albeit we do not see very well what it would become without some of them. We can notice it is simply a matter of preference for a bodily or biological symbolism. Then the modern term of holism masks its own symbolic without possessing the expressiveness of Organicism, neither the neutrality of system. By opposition, this idea of system lies on a more abstract and possibly mathematizable symbolism. And it can be generalized to biological, mechanics, natural, artificial contexts, or to anything, as information theory does!

System analysis, implemented by cybernetics and carried on by AI, is precisely the neutral, fetishism free, recycling of totalities, or interactions between entities. It is here a first point that shows the vacuity of the phenomenological opposition to artificial intelligence.

A special version of holism can be characterized by the different versions of philosophical infinity. The Gestaltist idea that uses totalities (actually, local totalities), to perceive parts, then becomes an absurdity consisting in claiming the already realized infinity of human knowledge, as the linguist Bar-Hillel did (Dreyfus, p. 215). This illusion of infinity is difficult to sustain, because knowledge is not only definite, but it is rather reduced for some people, contrary to the quantity of stupidities that it seems utterable. Actually, much more tritely, phenomenology, in its daily utilization of examples, incarnates the philosophical archaism of infinity in its list form. Anyone understands what infinite intellectual effort is accounted for philosophers used to wander in generality:

“Anyone in our culture understands such things as how to sit on kitchen chairs, swivel chairs, folding chairs; and in arm chairs, rocking chairs, deck chairs, barber’s chairs, sedan chairs, dentist’s chairs, basket chairs, reclining chairs, wheel chairs, sling chairs, and beanbag chairs – as well as how to get out of them again.” (Dreyfus, p. 37).

Let’s note first, on the contrary, it is notorious that some people do not know how to sit or how to clamber out of a beanbag seat, or to unfold a reclining chair. But the entire problem is to know whether this list is really infinite. That can practically be limited in distinguishing it with other lists. The only infinity can simply consist in the idea that a new seat can be invented. Elsewhere, the following example does show how indetermination problems are also raised for human being:

“To take Winograd’s example: […] ‘The fact that last week Rusty flew to San Francisco would be expressed by describing the event as a typical instance of Travel with the mode specified as Airplane, destination San Francisco, etc.[NOTE 17]’ […]. But etc. covers what might, without predigestion for a specific purpose, be a hopeless proliferation. The same flight might also be a test flight, a check of crew performance, a stopover, a mistake, a golden opportunity, not to mention a visit to brother, sister, thesis adviser, guru, etc., etc., etc.” (Dreyfus, p. 50).

This kind of unwarranted speculations seems the only infinite in the circumstance. On the contrary, anyone can observe that the problem stated to AI allows solving a questioning on human communication. The program to trigger is ‘Ask Rusty,’ or in a text, ‘limit speculations according to what comes before or after, or to default values (or to probabilities if to guess)’.

These precise examples are however useful. Anyone can see how proceeds this philosophical inhibition, which obviously cannot stand the limitation principle of the extension of a concept. Dreyfus seems to apply the primitive numeration principle: one, two,… a lot. He is able to count here until one, two,… twelve types of chairs, and it come immediately to… infinity. The French computer scientist Jacques Pitrat did notice it too:

“H. Dreyfus has the lucidity to see precisely the huge difficulty of the problem. But he makes a mistake in confusing very difficult and impossible […]. Yes, human brain uses a great deal of knowledge. But this mass of knowledge is certainly not infinite. We have there the same drifting from “a lot” to “infinity” […].” (Jacques Pitrat, “Discussion,” in Dreyfus’ French edition, p. 434.).

The philosophical infinity is simply held in check by this epistemological closure of micro-worlds. AI research workers have precisely represented this notion when they met this infinity problem. But the example of philosophical difficulties in extension definition comes from very high: the obscure Kantian opposition between synthetic and analytic concepts actually covers this problem. Cassirer has tried to explain why extension was analytic and weight was synthetic, he concluded extension was analytic because of the classic scholastic definitions (cf. Umberto Eco, “The quest of the perfect language,” Lectures at the College de France, 1992-1993, untranscribed passage in the published work): the “big dogmatic sleep,” whose Kant has confessed being victim, has left him narcoleptic aftermath. With artificial intelligence, this infinity problem simply solves by a comprehension definition. Insurmountable theoretical difficulties are not raised, since it concerns, very classically, to clarify characteristics or specific differences.

These micro-worlds can suffer critiques which remind those against the frequent inanity of models, “oversimplified and had hoc cases” (Dreyfus, p. 119). But this is the case of all school exercises, in human training. They can in both cases (human/artificial micro-worlds) have an experimental or pedagogical value. This resolution of micro-worlds by Schank’s scripts [NOTE 18] precisely allows some simulating practical and repetitious human knowledge, as it is admitted, to be however immediately denied (as usual):

“No doubt many of ours social activities are stereotyped and there is nothing in principle misguided in trying to work out primitives and rules for a restaurant game […]. But […] going to the restaurant is not a self-contained game but a highly variable set of behaviors which open out into the rest of human activities.” (Dreyfus, p. 42, my italics).

Here, we can again observe an illustration of the holistic refusal of any analysis, grounded on the supposed infinity. Dreyfus seems to look for what is missing, instead of recognizing what is brought. Is it about an eternal discontent trait: a half-full bottle is always half empty? And this relation then defines the subject and not the object. Is it about dogmatism, which does not tolerate discourses free from ritual phenomenological signs? Competition on reserved area is worsened when to deal with plain knowledge engineers, daring to face philosophy, like Minsky facing Husserl’s noema (cf. Dreyfus, pp. 35-36).

For instance, the restaurant script does not say it is necessary to breathe, for normal human behaviors are undertaken by other default scripts, or by properties inheritance mechanisms. This ordinary knowledge characteristic is programmable, and becomes the object programming new standard. It is also necessary to note that scripts can represent a representation of sociological level. They allow representing of cultural differences by distinct scripts. The micro-world is thus a good modeling of daily actions in a given cultural setting. It also models general cognitive expertises, as this human psychology study is showing:

“To anticipate an event, to repair one’s bicycle, to make one’s budget or to train one’s program action, it is no need recasting all causality and time, to revise all admitted values, etc. The solution only has to extend and to achieve already grouped patterns, even if it is necessary to correct the patterns with minor mistakes, and especially to subdivide and to differentiate, but without rebuilding them in a whole.” (Piaget, La psychologie de l’intelligence, p. 46).

This is also the result already acquired by artificial intelligence research workers, Papert and Minsky, whom again specify this epistemological closure limiting the number of categories to master for possessing a competence on a specific area:

“They conclude: ‘But [he list] is not endless. It is only large, and one needs a large set of concepts to organize it. After a while one will find it getting harder to add new concepts, and the new ones will begin to seem less indispensable’ [NOTE 19]. This totally unjustified belief that the seemingly endless reference to other human practices will converge so that simple micro-worlds can be studied in relative isolation reflects a naive transfer to AI of methods that have succeeded in the natural sciences.” (Dreyfus, pp. 11).

On the contrary, for what sake such a micro-world concept – stemming of case study – would a priori be refuted. When someone has the guts to tackle the ordering task, his observation of a limitation deserves a better consideration. A prejudice is precisely all the more unjustified that it is denied by the experience in question! Dogmatic denials or irony indicate a psychological resistance mechanism.

The phenomenological illusion can therefore result from the claim that knowledge is always incomplete (this incompleteness is actually only linked to the progress of our knowledge). But this apparent carefulness can be nevertheless opposed to the alternative of a total ignorance. Successive knowledge stages ought simply to be distinguished. This general incompleteness of human knowledge is precisely understandable in term of closure and not infinity, what is finally reluctantly admitted:

“Schank […] is right. Normally in reading a story we do not suppose that a person who enters a restaurant for a purpose that does not involve eating is preparing to eat.” (Dreyfus, p. 44).

We can see the result of this quest for other mistakes or gaps ends therefore to a fiasco. The failure is total, at least, in the differentiation between human being and machine. Therefore, a lot of phenomenological glosses for nothing!

Elsewhere the all-or-nothing demand is quite contradictory from those refusing binary solutions (expressed by 0/1 in computer’s world)! This is certainly the philosophers’ idea of logic. However, mathematicians know these digits (0/1) are also equivalent to limitations expressing probability. Even if this knowledge is not yet distributed enough in academic fields or in public at large, more often manipulating percentages, which only are another conventional presentation (from 0% to 100%, the interval is continuous and decomposable).

A relative validity of anti-analytic arguments is the necessity to examine all the sub-problems combinatory, in AI in general and especially in expert systems. This combinatorial explosion is often emphasized for it leads in the large numbers area. In the fable in which a sultan wants to reward the inventor of chess game, this one put a grain of wheat on the first square, two on the second, doubling until the sixty-fourth. This gives 2 to the power 64, that is approximately 1,84 x 10 to the power 19 (18,4 billion of billion grains), what exceeds a lot the world production, even current one. Then, the argument is encountered that we are rapidly in front of an amount superior to the number of atoms in the universe. This scientific arguments suckers’ enthralling operation forgets obviously to consider that the binary capacity of sixty-four bits [NOTE 20] is precisely equivalent to the former lese-sultan’s crime. If it is easy to generate, with a binary combinatory, a tree possessing a large number of descendants, doubling at each level, it is easy to find an information according to the same principle, with equivalent indexes. Each billion information is simply accessible in thirty levels; each covered in a few milliseconds.

Dreyfus perceives the combinatorial notion as a limitation of computers. And he is actually right if considering that a computer has to try systematically all the solutions. To explain gaps between strategies of chess programs and human players (26,000 Vs 200 moves checked) Dreyfus is bound to consider “what William James has called ‘the fringes of consciousness’” (p. 103). Therefore, the only advantage to the computer is not to forget any possibility, as emphasized by Newell (idem, p. 101-102), what is nevertheless good to mention. But how to claim that a man never miss an interesting move. If it weren’t the case, no grand master would ever lose. Or precisely, that would situate chess game in an entirely deterministic context. All of this is rather about a rationalization of grand master, or maybe about a mystification of the credulity of whom has a belief to justify. Human being is also made of this.

However, contrary to the so-called stupidity of this combinatorial strategy, it is perfectly possible to limit the number of results or research operations. Traces of learned paths can be recorded for formally similar researches, or to organize research order according to particular criteria (what any databases administrator is paid for); and in general, any knowledge designed to get optimization can be accumulated (besides the growing microprocessor quickness and memories capacity).

In history, the combinatorial explosion has already raised problems that had precisely motivated mathematical formalization of the question. Umberto Eco has reminded us that Raymond Lulle, in search of a perfect language, had felt dizziness in front of the large number produced by the factorial. This point is interesting for more than one reason. Besides the origin of the problematic of rational calculation, Eco observes this mystical dizziness with the large numbers and explains it by the insufficient categories to conceive a new reality. Elsewhere, the young children still feel this dizziness today (phylogenesis be still in ontogenesis).

Finally, it has to notice that the whole human intellectual production (any discourse, fiction or scientific hypotheses) seems equivalent to the expression of a combinatory. In the literary area, like in every day life, actions could be interpreted as those generated at random. And very often, personal or other experiences cannot be observed as useful to avoid reproduction if the always same mistakes. The supposed privilege of human being is again impeachable.

Actually, the philosopher is satisfied when he find a regress to infinite. A metaphysical trance state is observed with such a discovery (Dreyfus, p. 222). This state is simply an inhibition shock. In data processing, some loops produce this result, when a function of the program is calling itself. Programmers know too well this risk in their professional practice. Nevertheless, they also know it is necessary to define stop conditions to escape from it. Philosophy can be reduced here, operatively, to a prehistory of data processing, which still ignore the method defining a success condition! Human relationally speaking, we understand now the lack of acknowledgement for AI research workers’ local successes.

Besides this computerized solution, Dreyfus reminds himself a common sense one, in which the limitation of the explanation is the competence of the speaker. The French philosopher Pascal had very well understood this way of suppression of the regress to infinite risk – or obsession:

“This order, the better among men, does not consist in defining or demonstrating everything, but to stay in this happy medium not to define clear and widely known things, and to define all others.” (Pascal, L’esprit de la géométrie, paragraph n° 22).

It is easy, indeed, to notice that the regress ends when the subject or the speaker is satisfied. The interest of this solution is precisely it takes into account the existence of real speakers situated in time and space, with their particular competence, according to phenomenology’s principles. But it should have been noticed that the problem stands for human like for computer [NOTE 21]. Any explanation given by a computer can, under these conditions, be considered as satisfactory. On demand, it can be used as basis to a supplementary explanation. It is precisely the hypertext principle, which can provide all necessary explanation about any term already included in an explanation. Thus, any text includes its own dictionary or even its teachware (teaching software). The interest in hypertext developing is the ability to define explanation levels, and not to repeat already provided explanations: a same support is both a specialized paper and its popularization. This is obviously not possible for books, or for a real teacher, situated in time and space, in a real class. In the better case, this would be the situation of the tutor, if omniscient, or if we all were Alexander the Great in his time, enjoying lessons from Aristotle.

The problem reminded by Dreyfus could however be equivalent to an occasional limitation of computer research. It is indeed possible that DP men or computer scientists have tended to be immediately researching ultimate breaking down. The very denomination of primitive, like Roger Schank’s ones, could indicate it. However, the fact that composed or primitive concepts can be chosen does not contradict at all a local solution. Then, denying this ultimate characteristic, when the purpose is satisfied, become absurd: this is all the difference between demonstration (satisfaction of goals in a means-end framework), and contradiction or regress to infinite (as a cognitive distraction).

The mentalists’ big mystery is the introduction of intentionality. This word is a real mess that elsewhere covers exactly the notion of spirit, in the out of fashion ways:

“’Intentionality,’ by the way, doesn’t just refer to intentions, but also to beliefs, desires, hopes, fears, love, hate, lust, disgust, shame, pride, irritation, amusement and all those mental states (whether conscious or unconscious).” (Searle, p. 16).

The phenomenon to represent or to explain is what tradition could have called spirit, soul, and consciousness. Thus phenomenology carry out a passage from the relations to the world of things, objectivity to anthropocentric finality, or to animism, even to theatrical typecasts grounded on archaic psychological categories. If a universe without consciousness will be meaningless, always according to Searle (p. 15), it is a pun on the term meaning: law formalism, or anthropocentric finalism. Mentalism makes confusion between meaning for human beings, albeit navel-gazing, and meaning of the universe in itself, i.e. physical laws, existing long before the existence of human beings!

Actually, intentionality can be most often reduced to subjectivity. However, like in the following: “[i] I can feel my pain, and you can’t. [ii] I see the world from my point of view; you see it from your point of view” (Searle, p.16), it can rather be called individuality (i) and relativity (ii). In both cases, as it can be acknowledged, there is no distinction with material objects: when a stone falls, all stones don’t fall (individuality), and a context exists too for all objects (relativity).

The problem that obsesses phenomenology is obviously grounded on emotions. This particularity is simply linked to personal experience of the alive being in which information is about the subject, and only him. The culture is simply, from this point of view, the result of a memorization (more or less selective), shared by a group, thanks to communication. This reality is formalizable in both cases by a memory system. However, an individual can indifferently deal with information concerning himself or not, including information with an affective cost for others, without being really concerned. Commiseration, solidarity, or any other feeling, can be itself very formal, or fake. Real sharing can elsewhere be a matter of hysteria: a pain (cf. Searle) is shared metaphorically. This hysteria is precisely about confusion between real and fiction. As Searle said so well (p. 47): “The metaphor only becomes harmful if it is confused with the literal sense.” From which possibly a drifting to a pathological form.

Taking intentionality seriously remains an effect of academic leniency that can make us multiply the false concepts. Anyone can realize it when the philosophical meaning of intention is opposed to the supposed vulgar one:

“I need to introduce one or two technical terms. […] Intentionality. To say that a mental state has intentionality simply means that it is about something. […] Intending, in the ordinary sense, has no special role in the theory of intentionality.” (Searle, p. 60).

But the high philosophical technical intentionality, impossible to hold in state, finally, most often, amounts in its trivial meaning, or generalized into a not very operative concept:

“In the theory of action, there is a fundamental distinction between those actions which are premeditated […] and those actions which are spontaneous […which…] I call an intention in action.” (Searle, p. 65).

More, this reduction to consciousness incidentally depreciates the role of the context:

“If I am going for a walk to Hyde Park, there are any number of other things that are happening in the course of my walk, but their descriptions do not describe my intentional actions. So for example, I am also moving in the general direction of Patagonia, shaking the hair on my head up and down, wearing out my shoes, and moving a lot of air molecules.” (Searle, p. 58).

It is obvious, for instance, that shoes wearing, as negligible it can be on a walk, is the only reason for their replacement, and for the place of all replacement in a budget. The ruffling of hair justifies too hairstyle or hats. But these contingencies don’t deserve philosophy. All the banality of the process lies here. The term intention, which usually implies thinking, has simply been generalized to material, automatic, or unconscious actions. This could most possibly be explained by the opposition to the behaviorist’s study of automatisms.

This mysterious mentalism can be clarified more specifically. It is one of the Searle real interests. His phenomenalist enlightenment makes us rediscover, behind the big words of philosophical jargon, the simplicity of phenomena:

“I decide, for example, to raise my arm and lo and behold my arm goes up. But if our thoughts and feelings are truly mental, how can they affect anything physical?” (Searle, p. 17).

Let’s notice, first, that the problem lies in the body/mind dualist framework, although reputed expired by the same authors. But, let’s especially notice too the banality of problems raised by the philosophical mind. The argument of the immediate universality of these questions leaves a kind of uneasiness in front of the negation of all scientific knowledge, and in front of the lack of any analysis method. For the most, can we however admit for these elementary questionings the fact that sciences no longer, for a long time now, care to answer to naive questions, children’s and ignorant persons’ ones. In this meaning, natural sciences have at least the blame to leave the field open for archaic speculations and for charlatans selling miracle solutions.

One of the problems is also the appropriation of scientific knowledge by philosophers to distort them from their matter and associate them to traditional mythical worldviews. Simple misinterpretations are observed then, as when Searle (p. 63) considers the classic experience realized with frogs in natural science class as an evidence of Mentalism. This choice is particularly surprising given the fact that this electrical stimulation could be used as foundation of behaviorism (stimulus-response couple).

In the same way, mental causation really look like parapsychology, to which have precisely yielded all those having wanted to generalize the pragmatic functions of mind to mental causation of mind, like William James, Henri Bergson, etc., until John Searle. These analyses are obviously exclusively grounded on the evocative power of words, which have from this point of view, a mental causation. Out of these divagations, no one can see where is any obstacle for AI. Causality of the brain is the production of representations. These representations are phenomena easily simulated since the fact of a (formal) program ruling a (corporal) movement is perfectly known, in the robotics case. If what represents could cause, there would be no obstacle to consider this feature is shared by a computer program. The entire problem is actually reducible to a simple psychologization, whose we have, especially in Searle, the ultimate evidence. The classic AI example about necessary operations for a trip, in the continuation of Herbert Simon, which can be used to represent a problem in means-end term (to book one’s ticket, to prepare a suitcase, to call a taxi…, to enter the plane, to attach one’s belt, etc.). Then, Searle perceives this plain actions sequence in psychological terms, even in frustrated desires terms! When Simon sees a means (transportation, money, etc.), Searle sees frustrations “I don’t want to spend money; I don’t like to stand in queues at airports; I don’t want to seat in airplane seats […etc.]” (p. 66). When Simon considers (more or less aware) reasoning practices (means-end), Searle is psychologizing. Since brain causes all! And the analysis in term of desire can then be read as a verbal alliance with psychoanalysts in “the adjudication of conflicting desires.” (Searle, p. 66).

Indeed, we can admit that these psychologizing ideas do characterize some subjectivities, albeit, we could consider them as pathological, and epistemologically false by their negation of the pragmatic means-end schema. However, isn’t it possible to integrate these attitudes to learning, in this means-end model, by taking an account of failure or mistake situations? Incapacity to succeed is obviously frustrating, and interpretable in more or less animistic archaic terms (misfortune, bad influence, or even bad intentions of the machine). In addition, a situation is the more irrational than it consists of interactions with other pathological mentalists – what is known to cause/induce rationalizations (environmental pathologies according to therapists).

This phenomenological mentalist idea appears in the ritual manifestation of a few clichés, used as abracadabras to maintain archaic ideas. In AI adversaries (and in philosophy in general), the opposition to computers intelligence lies first of all on a mechanism of self-persuasion consisting in talking a lot of fine words and transforming some formulae into clichés: “If one is sensitive to the central meaning of theses interconnected intentional terms […]” (Dreyfus, p. 8); “ […] to learn, in a sophisticated sense of this world” (idem, p. 109); “[…] when viewed in a broader perspective” (Winograd & Flores, p. 31), etc.

This technique lies in classic rhetorical valorization by these magical terms (abracadabras). It is traditionally frequent, for preachers or ideologists, to use terms that brings in the discourse the orthodox comment to be told on its matter. Communist hackneyed phrases were customary of this process. It is necessary not to forget that many (most) academics have had to assimilate it to undertake their career. There is always something to have been gotten out of it.

Reciprocally, these signs can be used as indicators of these biased ideas, to say the least, sometimes until a pathologic state. Some signs do not deceive, and some high initiated know how to perceive them:

“The editor’s business is to recognize a madman at glance. When the topic is Templar knights, it is almost always a madman.” (Umberto Eco, The Foucault’s Pendulum, p. 71).

“There is very few serious books about Rosicrucians. How to recognize a serious book on this matter? Simple, if it is about Rosicrucians, it is not serious.” (Umberto Eco, “The quest of the perfect language in European Culture History,” Lectures of the College de France, November 12, 1992).

In AI adversaries, some abracadabras are encountered, representative of some borderline experiences, becoming leitmotivs for current theories:

The various super-powers issuing from a fringe consciousness, or intentionality, can be recognized, like the radiological vision, which make see through walls of a house (Dreyfus, p. 104). Intellectual confusion is easy to understand here too: we do not see through walls, we know what is behind. And according to what we know about some similar houses we can imagine what it is behind the walls of houses we do not know. This process is easily programmable. It hasn’t anything mysterious, and does not concern extrasensory perception. A computer can very well deduce that, in any houses, even of an unknown type, some people can feed themselves, sleep, etc. And the cases it could be mistaken are the same foreign cultural situations in which a human being would also be mistaken. Moreover, in the same way, a suited program could then record these new experiences and take them in account in ulterior reasoning (even more surely than a human being).

Heideggerian and phenomenological philosophy in general have brought us the cult of poet’s. He becomes the new prophet (what make his former arrogance absolutely unbearable then). Facing impossibility of total comprehension or perception of “his entire belief structure,” Weizenbaum (p. 193) turns again to this old cliché. But the saving poets’ tirade (versus social sciences) is a logical absurdity: if a poet understands something to human being or to society, he is a psychologist, sociologist, anthropologist, etc. The corporatistic monopolies have not the power to abolish reality. And poets can enjoy a usurped reputation. Elsewhere, they are in general sufficiently conformists (court, official poets) to fall in the oblivion when the regime has changed! The myth of the nonconformist poet is simply grounded on an alternative conformism, or on the literature professors’ future one who appropriates a past revolt when being classic.

The inter-contradiction between AI adversaries appears here when Dreyfus notes the calling to the left and right brain abracadabra by Weizenbaum as “the latest ‘scientific’ version of the Platonic dichotomy” (Dreyfus, pp. 64-65). Abracadabra indeed, but it is not because an element of the platonic theory allows it to be run down that the very phenomenon of scientificity can be questioned. Only the usage as abracadabra has to be criticized, as it can be used to justify no matter what.

The holistic dogmatism, refusing all division, obviously turns down the analytic micro-worlds. Nevertheless, they would be accepted if they would express this cabalistic virtue encountered in holograms (Dreyfus, pp. 20, 25, 51), which stands against the information-processing hypothesis. Albeit this author admits that it concerns a simple hypothesis, it is always good when it is so convenient. Holograms rather can simply be considered as the last avatar of holistic philosophy, when overused.

The movie-lovers know very well the famous example, reminded by Dreyfus (p. 125), of the impassive actor, interpreted in a different way in association with three different scenes. In fact, this experience doesn’t prove anything else than the fact that actors are overestimated, and than the scenario is all. It is even surprising that Dreyfus hasn’t claimed that the actor Mosjoukine had actually had an ambiguous expression, containing the three expressions; either for he knew the purpose of the operation, and he was a brilliant actor; or he ignored it and had an intuition or an anticipation, thanks to fringe consciousness, even maybe to quantum reversibility. Imprudently, I had put forward the prediction above, but this abracadabra of the Russian actor is immediately followed by a gloss for denying deduction from partial elements. And in which it emerges that: “The expression is not deduced from the traits; it is the organization of the eyes, the mouth, and so forth.” (Dreyfus, p. 125). This really seems to mean that the three expressions are present by phenomenological magic.

The current physics is today largely used to justify all the most traditional irrationalist theories. In France, on France Culture (cultural French state-owned radio broadcast company), Michel Cazenave was the leader of this trend, with the Conference of Cordoba – in which, incidentally, broadcasting has sometimes shown that these interpretations rested on verbalist misunderstandings. Conversely, it is necessary to grant Weizenbaum that his summary of the Heisenberg and Gödel’s original references do not yield him to these more or less delirious interpretations. He makes very precisely clear that:

“Werner Heisenberg […] did not thereby falsify Leibniz’s conjecture. But he did show that the major premise was unattainable.” (Weizenbaum, p. 221).

However, these principles denying total deductibility therefore contradict his chapter five (pp. 132-153) resuming the classic epistemological criterion of theory as deduction. But contrary to the epistemological current vulgate, the (macroscopic) determinism is not contradicted, as Weizenbaum very well points out, only an absolute human knowledge outlook is denied. Elsewhere, the still invoked intuition is the psychological equivalent to empiricism. Anyway, as Pamela McCorduck also remarked, these knowledge limitations are as about human being than computer! And their use as argument by AI adversaries is not therefore legitimate.

A special mention has to be made to the abracadabra reigning in the epistemological area, which can be called Copernicus’ fantasy. The matter at issue is not only the fact that Copernican legend is used as a leitmotiv in this field, with possibly some misinterpretation. It concerns the fact that any theory a few different, if not an original one, is presented as a Copernican revolution.

However, this reference from AI adversaries has a supplementary sophistication degree, because partisans of AI are assimilated to dogmatic tempting to secure their prejudices in front of difficulties, like former Ptolemy’s cosmology partisans, “they try to save their theory by adding a few epicycles” (Dreyfus, p. 114, see also p. 173). Thinking to be Copernicus, Newton, Einstein, Bohr, etc. is easy, but a priori, the dogmatic resistance to a new thesis would rather be the AI adversaries’ standpoint, without prejudging the rightness or the falseness of their standpoint. And maybe, in general, scientists should accept first to be Ptolemy.

The critique of assumptions has become one of the current philosophers’ competences. Philosophy, and the reigning epistemological vulgate, has succeeded to impose this principle of critique by prior hypothesis analysis. Dreyfus claims this method by reference to Kuhn’s paradigms (Dreyfus, p. 39). In the most general way, assumptions critique is inspired by Kant, who challenged Newton’s statement about not making hypothesis: “hypothesis non fingo” in his new theory. This Critique is grounded on an enlightenment of the actually undemonstrable assumptions (here: uniform rectilinear movement, persistence of movement, fall of bodies), grounded on other observations or speculations. Unfortunately for Kantian philosophers, this interpretation is false, because Newton is merely referring to Cartesian’s epistemology, in which divine intervention was guaranteeing world coherence. Newton, like the Laplace of learned legend, doesn’t need a Deus ex machina hypothesis.

In the field of human phenomena study, we have too these critical assumptions. Sociological tradition even builds on such a refounder standpoint, in which “the sociologist should set anticipated notions about fact aside and face the facts themselves” (Durkheim, Les règles de la méthode sociologique [Rules of Sociological Method]). But it is necessary not to forget this precaution, in its time, aimed the introspective philosophical method. To the resumption of this idea by the French sociologist, Pierre Bourdieu, under the form of a whim: “there is no immaculate perception,” we could oppose the question of an immaculate conception existence! Another French sociologist Raymond Boudon sums up this critique of assumptions idea under the name of Simmel’s effect, which it then would rather be called Kant’s effect. And he uses it as an explanatory principle of mistakes as he considers “they derive from sound arguments contaminated by implicit scopes” (L’art de se persuader des idées douteuses, fragiles ou fausses [The Art of Self-Persuasion: The Social Explanation of False Beliefs], p. 15). This approach gives an easy leading role to German sociological tradition, possibly by erudition or pedagogy effect, without questioning itself too. However, the Boudon’s approach aims to restore rationality of daily discourses, contrary to intellectual tradition – (Bourdieu) – denying consciousness or liberty to subject. This Boudon’s book central thesis is precisely a point of convergence with some artificial intelligence works. As it is observed, the role of assumptions can actually make obvious a cognitive problem when a rationalization rests on excess premises (cf. Jean-Louis Desalle, Scientific Meeting of the ARC, 7. 12.1992). But this problematic still concerns indifferently human being and machine.

This philosophical assumptions analysis competence produces a foreseeable enough discourse, automatically tacked on. Dreyfus devotes thus the second part of his book to the assumptions analysis of AI. He identifies four of them: biological, psychological, epistemological, and ontological assumptions. Actually, they could be reduced essentially to his refusal of an allegedly binary processing (cf. pp. 155-157).

Dreyfus has invented a biological assumption of AI that would impose to the brain an information processing “some biological equivalent of on/off switches” (Dreyfus, p. 156). The phenomenological reductionist obsession makes him insist on what was a provisional neurobiological model of the brain. But that hasn’t anything to see with concerns or assumptions of AI. Anyone can even see (on the previous page!) that all is about a Dreyfus’ misinterpretation about the quotation of Newell and Simon:

“Newell and Simon introduce one of their papers with the following remark: ‘It can be seen that this approach makes no assumption that the ‘hardware’ of computers and brains are similar, beyond the assumptions that both are general-purpose symbol-manipulating devices, and that the computer can be programmed to execute elementary information processes functionally quite like those executed by the brain.” [NOTE 22] (Dreyfus, p. 155).

The quotation, although the original is available, precise it concerns only a functional simulation. The obsessive fixing on digitalness of computer processing cause/induces, as we can see it, a total lack of understanding of any other reality. But for research workers of AI, this biological assumption doesn’t exist, since this demand of likeness structure/function is the philosophers one (cf. Dreyfus, p. 168). This demand benefits possibly from the concerned complicity of neuroscientists through connectionism. This (common) lack of understanding is obviously grounded on the confusion between functional and organic.

The psychological assumption concerns – strictly speaking – information processing. Albeit here too, the famous “bits of information” contaminating this psychological assumption are still useless. It is certainly a consequence of analysis refusal, and in fact, of any thinking, perceived as neutral and impersonal, like in classic rationalist tradition, with no protagonists’ involvement (Dreyfus, p. 156).

It has to be noticed too that information processing is oddly enough considered as a victory of mentalism on behaviorism (p. 164), allowing to get rid of experimental psychology. This can be understood when philosophical psychology neglects rational knowledge too. While the phenomenological project lies too much on the idea of commitment in a situation to accept any mastering of situations with explicit thinking. A precursor of this negation of experimental psychology or rationalism can be also found in the work of Jean-Paul Sartre, Esquisse d’une théorie des émotions [The Emotions: Outline of a Theory], small 66 pages handbook providing an excellent productivity for philosophy students in order to prevent many disturbing readings.

The obsessive refusal of all explicit or calculable form is also risking here to lead to reject knowledge itself, because is it a better knowledge ability (except the Boolean and binary obsessive reference) than the epistemological assumption, its clarifying possibility, and its logical expression (Dreyfus, p. 156). A more flexible definition of knowledge, enough to include inexplicit social practices especially observed by sociology, cannot nevertheless deny a clarifying possibility. Dreyfus had accepted, on principle, against Searle, human sciences possibility, but he is refusing their reality when he is questioning more especially: “(a) that all non arbitrary behavior can be formalized, and (b) that the formalism can be used to reproduce the behavior in question.” (Dreyfus, p. 190). It is happening that this negation has been contradicted (i) by the philosophical-practical tradition of the Encyclopedia first characterized by a formalization of arts and techniques; and (ii) by any industrial robots or machine tools.

The ontological assumption simply refuses any analysis or any logical breaking down of reasoning or behaviors (Dreyfus, p. 156 and p. 207). The opposite assumption is that anything is in everything, and that contexts, magical word (or abracadabra) are encased one another. But the simple identification of any context is isolation of a reality all right, whatever its name is context, situation, event, etc. The main consequence, at the period this Dreyfus’ book was written, was the fact that: “ […] the problem of artificial intelligence becomes the problem of storing and accessing a large database” (Dreyfus, p. 209). Today, these technical problems are very largely exceeded, even if their importance was already relative, formally speaking.

The contribution of the Kantian’s critique of assumptions actually turns out to be perilous to handle. As indicated by the simple ambiguity of the term critique itself: analysis too often become stigmatizing: “The assumption was taken to be self-evident – an axiom seldom articulated and never called into question” (Dreyfus, p. 225). Anyone can see here that Dreyfus put forward the assumption definition itself to the one he disapproves. This lack of neutrality in the assumptions approach and in their critique does seem to show that the concept (or what it is referring for it is not so clear) is not yet assimilated. It is what is precisely characterizing an archaism, when use of a new notion is regressing to the former one: here, dogma [NOTE 23].

Another present possibility of the term assumption is referring to operative concepts or even ad hoc ones (which has the meaning of abracadabra in epistemology). That doesn’t mean a provisional recusal is legitimate. Weizenbaum (p. 199) aims for instance the use of the notion “[interlingual] conceptual base” by considering it is ad hoc. He’s referring to Schank’s semantic primitives, whose reality is widely accepted, even if some persons prefer a largest choice than his eleven primitives, or are leaving free the reduction level [NOTE 24]. More, this gratuitously skeptical attitude diplomatically doesn’t criticize the Chomsky’s syntactic or semantic primitives. Elsewhere, his innateness human structure thesis is the Weizenbaum’s liking. But it doesn’t become true for all – excepted on the ad hoc mode that it has to be something allowing language, since there is such a thing.

Anyway, the operative aspect of definition is common to any concepts, scientific or not. They refer inevitably to a perceived reality under any denomination. The distinctive feature of a scientific concept is precisely to accept the provisional aspect of this delimitation. The more or less strong symbolization of terms only belongs to mnemonic processes, like in physics (where symbols are abbreviations) or mathematics (where variables are x, y, constants a, b, etc.). Weizenbaum is shrewdly noticing that is not necessarily the best idea and that is corrected in data processing, for programs maintenance reasons, as one can “use whole words to denote the variable […] The equation d = at²/2 when transformed into a program statement may thus appear as distance = acceleration * time**2)/2 [** is use to indicate exponentiation].” (Weizenbaum, p. 146). It is therefore an advantage of the computer programming all right! Its initial (often recurrent) esotericism has been cured, thanks to memories capacity increase allowing a more explicit notation.

Another possibility of critique of assumptions simply refers to Popper’s criterion. It demands not demonstration, but the possibility of refutation to deserve being called scientific. This reference is explicitly used against “the psychological assumption [which] seems not to be an empirical hypothesis that can be supported or disconfirmed, but a sort of philosophical axiom whose truth is assured a priori” (Dreyfus, p. 174). But even in the case of a refutability demand, it is good to insist on the import fact that no matter what thesis can be legitimately studied, and a metaphysical thesis stated, even for Karl Popper. And for this, it has to be more developed that would allow a challenge on impressions of its adversaries. Elsewhere, it certainly seems resting on a few arguments to its partisans [NOTE 25].

The French epistemologist Bachelard characterizes false ideas as epistemological obstacles. But obviously, these false ideas are legitimate for their contemporaries, or to those professing them. By definition, ideas aren’t put forward by those who believe them as illegitimate. And reciprocally, ideas judged illegitimate by contemporaries can be found confirmed thereafter.

If what characterizes modern science is free hypothesis study, it is obviously not the study of illegitimate hypothesis. But even if these hypotheses are preliminary, their liberty is contradictory with the famous idee fixe of monomaniac scientist according to Einstein. The permanent idea selection could be equivalent at random, because for a fertile idea, many others will be eliminated, or would be epistemological obstacles! Even the Feuer’s treaty, Einstein and the Generations of Science, could therefore be an a posteriori construction, because such ideas like relativism (in Einstein’s environment) always existed, and are always available. The so-called hypotheses, besides afterwards rationalizations, can always claim priority: as for astrology, all predictions exist in philosophy (as Weizenbaum is reminding us!).

From this standpoint, Dreyfus should have been noticing that composition of these preliminary hypotheses threatens to lead to a combinatorial explosion, formerly opposed to AI! He would be therefore led to find heuristics, which he challenges too, or led to have intuition, which look a lot like physicians of Molière’s sleep-inducing capabilities. It is often easy to extricate one self, by a verbal trick, from the problem that is precisely to solve.

With a small effort, we can also consider the accusatory usage of the assumption notion as clumsiness. It could be classified under the professional bias influence of suspicion philosophy (Marxist, phenomenological or structuralist), having marked academics training. It would be therefore necessary to consider productions in question as stylistic automatisms, rectifiable by an ideological-context-free processing other automatism, because very often, the so-called assumptions analysis is merely equivalent to the so-called philosophers’ incapacity to escape this suspicion framework (“who is speaking, from where?”). To deny legitimacy of those without the same assumptions, it seems work this way: (i) “this assumption is not demonstrated,” (ii) “then it’s false,” and (iii) “therefore mine is right” [NOTE 26]. This way of thinking avoids undergoing the only assumption validation criterion, in the pluralistic framework, consisting in comparing results (and not to minimize them, or to refuse partial results). The anti-IA discourse himself is proceeding by (phenomenological) assumptions adoption pretending to deny different (analytic) assumptions, without any common scale or neutral arbiter.

One of my philosophy professors claimed once that philosophy consists in giving an ultimate word. And contrary to the Popperian refutability principle, this suspicion philosophy simply consists in building a discourse on which refutation cannot have any hold. This is all right a way to be sure to have the last word. The escape in ineffable can thus be reduced to this rhetorical strategy refusing any contradiction possibility by an elusive discourse.

If anything can obviously be criticized by anybody, it is possible finally to question a little the philosophical assumption of assumptions analysis, at least as preliminary condition:

i) Even if this analysis is possible, nobody is obliged to question his own assumptions. Elsewhere, Dreyfus’ book isn’t either an assumptions analysis of phenomenology.

ii) An assumption is precisely, due to its definition, what is not questioned. It is also, since Kant, what others are questioning for you. This establishes the Kantian paternity for suspicion philosophy.

iii) This assumption of reciprocal questioning possibility is simply equivalent to free debate. It is obvious that dogmatism reigning in scientific/philosophic discourse may not to prepare to this standpoint.

iv) Rather than assumption, subjectivity or relativity (individual or collective), it would be preferable to speak about parameters (or parameterization). This elsewhere identifies parameters for formalization, or validity conditions. The interest is to preserve relativity under the free hypotheses variation form, giving up the rigid philosophical all or nothing tendency.

Anyway, the assumptions of others total negation process represents a preliminary philosopher dogma: no needed to know AI to pull out of one’s top hat this critique of assumptions abracadabra. But that is not excluding a contradiction in not admitting the utilization of an “individual’s beliefs system” (Dreyfus, p. 46) by partisans of AI, when Schank grounded his analyses on it, because the critique of assumptions discourse is himself in contradiction with the dogma that thought is not analyzable, and grows in practice “without ever having to represent ourselves, our body as an object, our culture as a set of beliefs, and our propensities as a situation ÷ action rules” (Dreyfus, p. 53). In short, partisans of AI would be the only persons functioning with assumptions. If their adversaries’ insincerity wasn’t so obvious, they could admit the limited validity of an individual working type.

What is remaining of philosophy today can be wondered, out the persistence of eternal old questions, like consciousness or mind presence in a world made of matter or particles (cf. Searle, p. 13). Former philosophies, and precisely Cartesian dualism, had brought answers to this kind of questions. Some people no longer consider them as valid because of the scientific inflation bringing partial answer, or changing the nature of questions. However, the negligence of these questions by all sciences, and their technocratic contempt can explain the philosophy success in maintaining them. This mistake, or this fault, can lies in the lacking of connecting these scientific admitted/recognized answers to human being’s natural questionings.

The astonishment idea, beloved of philosophy, rests most possibly on a professorial practice bias. Teachers seem will to generalize the effective reality of the discovery situation. With another tendency, inspired by Socrates, that would want the pupils or students to rediscover at any time each thought steps. The philosophical bias of problem conservation appears then under the archaism form. Here, the emotion comes before a solution. But on principle, it is difficult to believe in Santa Claus when one has grown.

Religions or residual philosophies are then equivalent to the archaic persistence of former answers or former ways to question. The mechanism of myth, according to Claude Lévi-Strauss’ principle, lies in the will to answer to all persistent questions of each step of knowledge development. Then can be explained diverse and various superstitions, beauty or virility old miracle remedies, renovated in parallel medicines, paranormal or parapsychology, or some versions of ecology, the whole renovated on the cultural identities mode.

AI adversaries are instinctively using the recourse to mystery that has always been a facility of theological discourse secularized into parasciences. It naturally leads to the strictly speaking magic evocation, which is not neglected by phenomenology. The phenomenological concern, legitimate a priori, to re-establish subjectivity is accompanied by an extremist negation of material objectivity, or cultural intersubjectivity. It only remains a string of phenomena appearing and disappearing to individual consciousness like by magic. (This process can be understood as formalization of the cognitive stage, described by the Swiss psychologist Piaget, when the child doesn’t yet perceive any continuity in the passage of a thing behind a screen and its reappearance on the other side!).

In the last analysis, Dreyfus can be paid back for the favor of his first paper against AI, when he associated it to alchemy, because anti-IA the discourse, phenomenological or philosophical in general, turns regularly out to parapsychological solutions. Several factors already mentioned compete there, and the philosophical method of enigma accumulation leads to sensationalism quest. Despite its pretensions, it can be amounted to the exploitation of a “incredible but true” journalistic style – common, true, to many sciences popularization. For those ignoring the fact, we know now, since the movie Dead Poets Society, that poetry or philosophy are used for driving adolescents who take them too seriously to suicide.

[NOTE 8] W. Ross Ashby, “Review of Feigenbaum’s Computer and thought,” Journal of Nervous and Mental Diseases.

[NOTE 9] Actually, Weizenbaum is right to generalize research of principles at religion and Law. But, speaking about contract concerning religion is confusing gloss and reality. On the contrary, religions or traditions are the most arbitrary dogmatist place. And this would be a serious misinterpretation and an inversion … to reduce rationality at religion. The possibility of the churchy alliance of “individuals of quite different mind, temperaments, interests and training.” (Weizenbaum, p. 11), means therefore that individuals in question have to be a Catholic, Protestant, Jewish, Muslim, Buddhist. They could be atheistic, but virtuous, if they have the same opinion than the former.

[NOTE 11] Holism of these parapsychological fields often stop with defamation of other parallel concurrent sects. In a way, the TV serial X files happens to be the only ecumenical place. So much craps are gathered for understandable scenario reasons are keen to create either a lack of credibility up to a big laugh, or a very soon paranoid epidemic.

[NOTE 12] We all know the Zen original reference of this puzzle. Oddly enough, nobody seems to link it with this other pedagogical practice consisting in splaping the pupil in the face in order to make him enjoy the non-theoretical (phenomenological) virtue of answers.

[NOTE 13] Of course, some inhabitants of western countries, or even some western countries do make debarring laws. But they aren’t in that sense western culture members. Actually the notion of regression is necessary in order to caracterize behaviors contradicting a classification: for instance, a physician making witchcraft isn’t a physician.

[NOTE 14] Rudolph Arnheim, “Intelligence Simulated,” Midway, University of Chicago, juin 1967, pp. 85-87.

[NOTE 15] We can remind the Claude Lévi-Strauss’ Tristes tropiques, in which he exercises at a scholar description of a sunset.

[NOTE 16] French phenomenologist Merleau-Ponty give to the hunting tiger a representation of matter permanency, which seems a little excessive, although in the Socrates’ philosophical tradition.

[NOTE 17] Bobrow et Winograd, “An Overview of KRL,” Cognitive Science, vol. 1, 1977, p. 8.

[NOTE 18] Scripts allow to define, for a program, the distinct actions that represent a sequence like, in the restaurant script: to book, enter, sit, choose dishes, eat, pay, exit, etc., with possible variants (free or according to some constraints).

[NOTE 19] Marvin Minsky et Seymour Papert, July 1970, Draft of a Proposal to ARPA for Research on Artificial Intelligence at M.I.T., 1970-1971, p. 52.

[NOTE 20] 64 bits = 8 bytes. A bit (binary digit) is the elementary computer unit (0 / 1) regrouped generally in byte (8 bits), which can be used to code a letter or a sign (a… z, A… Z, 0… 9,,”,” *, /, +,%, $,… etc. ). The RAM memory of a current computer (beginning of 1996) contains generally 8 Mo (Megabyte, or millions of characters). A hard disk contains approximately 400 Mo to 1 Go (Giga-byte, or billion characters).

[NOTE 21] Wittgenstein summarizes this Pascalian solution in asserting that regress to infinite can stop. Nevertheless, according to Dreyfus (p. 204), this feature seems to be kept aside for human beings, for the sake of the myth of a priori human /machine distinction.

[NOTE 22] A. Newell and H. A. Simon, Computer Simulation of Human Thinking, The Rand Corporation, P-2276, 20 April 1961, p. 9 (Dreyfus’s italics).

[NOTE 23] On the contrary, let’s notice it is possible, in a generous hermeneutic way, to consider dogmas as assumptions, in some former uses.

[NOTE 24] Like Bernard Levrat, for instance, is reminding us in Language (French) journal, n° 87, pp. 47-54.

[NOTE 25] Is this again a Marxist reminiscence? If he had claimed that it is not necessary to believe what peoples say about themselves, would it be necessary to believe what their adversaries say about them? What can explain particular communist trial procedures!

[NOTE 26] J. P. Bronckart (in Théories du langage, p. 172) had noticed too this kind of rationalization in Chomsky: “A new mode of demonstration, consisting first in stating a conditional hypothesis, […] presenting then this hypothesis as a probability, and finally, to close the matter by “observing” that facts have demonstrated the hypothesis and have transformed it into an absolute principle."