Chapter 2
HUMAN BEING THE MACHINE

Philosophy of the Body

Despite some confusion, the Dreyfus’ phenomenological standpoint presents an authentic reservation to artificial intelligence, identified to abstraction climax, when he is speaking about an incarnation necessity. Let’s however notice that, after some millennia of body negation, philosophers had thus have a change of mind. Abandoning theoretical philosophy, the new philosophers can declare, as Molière’s plays fake physicians situating heart on the right side: “We have changed all that! [web]” and can justify simplicity and body. But didn’t they adopted the phenomenological model under pressure of natural or human sciences competition, a little as painters have adopted nonfigurative way since the photography invention? And AI adversaries are showing a wonderful unanimity on this matter, whether it is “by virtue of having a human body” (Weizenbaum, pp. 208-209) or the existence of an “artificial embodied agent” (Dreyfus, p. 250). This change doesn’t fail to surprise lay persons, who however rally with confidence (cf. Anthony G. Oettinger, in Dreyfus’ French edition, pp. XX-XXI). AI adversaries nevertheless do not deprive themselves of traditional pure idea valorizations in his examples. In what chess players, commonly models of intellectual abstraction, are bodily committed? Maybe by his mannerisms, habits, or superstitions, which is the press and psychoanalysts delight?

This AI negation principle for the sake of body importance summarizes every topic previously developed, justifying them by this lack of incarnation:

i) First, obviously, the analysis refusal is only referring to the holistic norm against steps breaking down in a task carrying out, as when Dreyfus (p. 179) puristly quoted the machine personalization (“he” instead of “it”). But it does seem incoherent to claim that human being (then  a machine) can do no matter what task, especially a complex one, without distinguishing some steps. This breaking down has his right place in learning. On the other hand, the refusal of organism individualization is utterly dogmatic, if grounded on the only argument this organism is artificial, because if there were two identical artificial organisms, like human twins, they would differentiate by their individual experiences. Even abstractly speaking, databases are containing specific information all right!

ii) The body representation refusal (Dreyfus, p. 235) is reaching the refusal of simple simulation too. While what is the difficulty to simulate a body moving in space, since it is enough to take account of information transmitted by sensors. Any robot or any simulation program can easily possess such external world communication devices. The rhetorical literary minded philosophers false questions actually have a self-hypnotic function.

iii) The artificial learning formal aspect is not any more considered, for apparently logical reasons, but tautological with the simulation refusal, by making depend “the work of the central nervous system […on…] the locomotive system” (Dreyfus, p. 303). But in the Simon’s thought framework, no one can see any problem, since it is enough to consider information brought by the locomotive system. But it is necessary not to forget the most trivial experience: if mathematics operations is depending on locomotive system (as works about child intelligence do indicate), conversely, it is also necessary to admit that computers are already able to calculate. This confirms reciprocally the autonomy of an abstracts system once represented – phenomenon on which Piaget has also insisted about intelligence stages.

iv) The formalization refusal also seems to result from the Searle’s simplicity principle, adopted by Dreyfus, challenging the representation necessity by: “our sensorimotor skills for coping with objects and people – skills we develop by practice […]” (Dreyfus, p. 53). Besides the fact this idea excludes, by petitio principii, any reasoning possibility, it is obvious that those without these intuitive capacities are quickly eliminated. Then, they aren’t forbidden to cheat a little by trying to understand the rule of the game of the nature, even by copying from the neighbor.

Actually, phenomenology seems to consist in an idealistic thesis summary, inverted and joined to current scientific researches fragments, carefully selected, by forgetting those contradicting it. It could be a consequence of an idealistic model converted to real by the late assimilation of incarnation Christian religion, which represent a contradictory assimilation for followers of its Nietzschean negation. Anyone can observe this crossed knowledge distribution by the professional or amateur theologians’ recent conversion to phenomenology. The idealistic recurrent framework can thus allow explaining the complication and the confusion of the phenomenological discourse by the difficulty of materialistic category acclimatization to its jargon [27]. The refusal to admit any aggiornamento is apparent to the insincerity of professional ideologues, on the principle: “We have always deceived, but the others were wrong to be right too early!.” As it was the same intellectuals or artists who were Stalinian, – if not Nazis –, this rationalization is familiar to them.

Idealism, contradictory for apologists of the body, is elsewhere observed very precisely by Dreyfus, in Husserl himself, the phenomenology founder (Dreyfus, pp. 248-249). Only his late disciples have brought a correction. But maybe it is necessary to detect here a partial and distorted recycling of psychologists’ works like Piaget’s:

“Merleau-Ponty tries to correct Husserl’s account on this point […]. He argues that it is the body which confers the meanings discovered by Husserl. […] As Piaget remarks: ‘Perceptual constancy seems to be the product of genuine actions […]’.” (Dreyfus, pp. 248-249).

In both first philosophical cases, however, we can observe that learning is obviously neglected, while Piaget insists on it. As Piaget also precisely points out, Gestaltist fixist models of intelligence are contradicted by his genetic (i.e. concerning learning) models:

“Missing the genetic standpoint, ‘psychology of thought’ only analyze the final stages of intellectual evolution. Talking in terms of states and complete equilibrium, no surprise it amount to panlogicism.” (Piaget, La psychologie de l’intelligence, p. 33, my translation).

Anyway, by considering that a simple pocket calculator already undertakes operations judged intelligent for human beings, it is possible to consider that these skills are transmissible (to a machine or to a human being) without body constraints.

Consciousness of the Body or Self-consciousness

Then, impossibility a satisfactory body representation, or more generally of self-consciousness is opposed to AI. But, by opposing body to its representation, phenomenology merely looks like regression. It seems to claim inheritance of oriental or antique prohibition of body representation. Dreyfus (p. 25) appears happy enough about a lack of “serious theory” of this body representation, in 1968. But this lack rather characterized an interdisciplinarity absence at this period. Conversely, AI had already broke boundaries of traditional fields.

Anyway again, if common sense apology is essentially grounded on the possession of a body, as it is obvious that a robot has a body, it could have therefore developed the same reactions to a same environment. For an abstract program, it would be enough to simulate this environment, as all simulation programs are already doing it. The possibility of autonomy for an automaton could simply be defined in term of representations of internal and external states. This is absolutely the least condition, already mentioned by a robotics Spanish precursor, Torres y Quevedo, in the beginning of 20th century. Indeed, he did not believe in cognitive autonomy, but in his early time, it could conceive a perceptive and decision-making one:

“The main purpose of Automatics is that automatons are capable of judgement; that at any time, by taking account of impressions they get, or even of those they have got previously, they can carry out the required operation.” (René Moreau, Ainsi naquit l’informatique [The Computer Comes of Age], p. 29).

Actually, there is no obstacle to representation of its own body for a machine. The body is part of the world like any object. It is therefore easy to give a self-representation to any artificial protagonist in a natural or simulated environment. This simple condition is already established in the least simulation programs, or role play, which has to record players’ moves and characteristics for instance. And the minimum of autonomy could of course be about satisfaction of needs (to refill its batteries for instance).

An intermediate level of self-reference can also be considered as modeling animals’ behavior, or transitory human stages, or simply individual competence differences. This question is far from being solved in the human behavior case. In the famous prisoner dilemma of game theory, the topic is precisely comprehension of relationship between personal interest and reciprocity. This is equivalent to descriptions of egocentric stage in Piaget too.

The autonomy machines problem can also be expressed under the form of their ‘consciousness to be conscious’ impossibility. This characteristic is perceived as intrinsically human, albeit the meaning of this difference can be wondered at. Indeed, in both cases, it can simply concern the ability to produce statements concerning world or self-representation, whose proof for others is an explanation. If machine is able to be conscious of external reality, it has to be able to apply this process to self-representation. If it is able to represent the fact that ‘X think Y’ for anybody, it can be able to consider that X is itself. Elsewhere, this process is all the more easy to realize that AI principle lays on recursive functions application, on program ability to consider its own code as data, and then reprogram itself.

In the programs current framework, pure knowledge one all right, any program already possess its inside knowledge. In CAL (Computer Assisted Learning) framework, it already can have a representation of the user; for instance, due to fact that he hasn’t yet the knowledge he is learning. In general, it seems although this negation of machine autonomy or negation of modeling represents a lack of understanding or a misinterpretation, like in the arguments below:

i) The refusal of exteriority or abstraction which constitute a representations of a situation (Dreyfus, p. 53). A text understanding program is no more in real life than a person reading a text, watching a movie, or observing other people acting. From this point of view, comprehension is an action external analysis. Usually, philosophers rather overuse this standing back apology about their own work!

ii) The challenging of acquisition of knowledge for a robot, by differentiating its primary stage from child’s one, because: “My plans and fears are already built into my experience of some objects as attractive and others as to be avoided” (Dreyfus, p. 266). The author’s idea is that a computer is passive, especially in some AI experiments, associating meaningless syllables (cf. Dreyfus, p. 110, and Weizenbaum, pp. 161-163 too). The psychological model of these experiences would be too Pavlovian according to him. But, are these conditioned reflexes meaningless for interacting body? Let’s not forget a conditioning disappears without regular positive sanctions, without satisfaction of waits (of course, simulation must be complete). Who can believe a robot could remain without information capacity about the dangers of the world. Knowledge only represented this way is then available in a robotic model of the world, contrary to this ‘already building,’ looking a lot like a spontaneous soul generation.

iii) Therefore, in order to deny the excessively theorist principle of a head without body, the principle of a body without head is built, concerning: “what human beings implicitly know about walking and eating simply by having a body” (Dreyfus, p. 23). Then, when humanity grounded on the body is argued, a paradox of a body without distinct mind is met in Gestaltist phenomenological theory, equivalent to an innatist programming. It seems to forget that a brain is also necessary for thinking. Out of dualism, the problem could rather be seen as an interaction. But brain, part of the body, doesn’t know anything directly, and information has to be encoded. This can be objected to the immaterialism of a direct intuition. And one of the ways of information processing, for human or computer, is communication by language. Which can be objected to solipsism, consequence of the former philosophical tradition, yet condemned by AI adversaries.

iv) An AI adversaries’ contradiction is also demonstrated in the unacknowledgement of satisfaction condition, like in the solving of phenomenological ‘mind’s intent’ by Minsky’s heuristics (Dreyfus, p. 244). On the contrary, it could be obvious, even for a non-DP man, that anticipation can be easily programmed. No matter what computer program already waits yes/no, at least, when user is questioned. In the same way, is-it possible to consider that an indication allows to make hypotheses or to eliminate some others – what about Minsky were precisely speaking about in the quoted passage. In language understanding, it is possible and necessary to proceed to such anticipation. It is therefore a petitio principii proceeding, not only a priori denying it, but not to notice it when under the nose! Phenomenology’s “mind’s intent,” in everyday language, is being called blinker.

The dogma that an artificial organism is not in a situation, phenomenon considered as intrinsically human (by petitio principii), brings elsewhere a contradiction, because it happens machine would be incapable classify a plethora of elements. More, this quantitative necessity is devalued in aid of the quality:

“Human handicappers are no more omniscient than machines, but they are capable of recognizing the relevance of such facts [the jockey’s mother death] if they come across them.” (Dreyfus, p. 258).

Of course, the factor legitimacy is determined in function of circumstances. The former example in the Dreyfus’ text was about the irrelevance of the weight of chess game coins. Oddly enough here, the author no longer imagines some limit cases he is fond of, like the well-known giant games. Well, a chess game program doesn’t take the weight in consideration in the game course. But an initial verification condition can be easily imagined for an autonomous robot of its prehensile capacities. And who can see how a human being, or a machine, would decide the information importance without learning. In human world, indeed a former experience determines the relevance of factors. And it is easy to understand that a simple statistical processing allows determining what are relevant factors and what is their weighting (in which weight is relevant under this statistical meaning). To conceive an element as a priori determining is precisely an external intervention, which was precisely the condition it was necessary to avoid, by hypothesis, for the program (cf. Dreyfus, pp. 264-265)!

The final judgement is empirical too: it won’t be necessary the betting computer always wins, but it only has to be more competitive than a human player, in order the project is validated. However, it will always win less than a cheat. Eminent human superiority on machine! And anyway, it is obvious computer is the ideal tool to deal with this kind of problems with a large number of variables, because the mass of information would rather be a handicap only for human being.

Are we Programmed?

The traditional immediate opposition to AI possibility is known to be computer programming by an external human intelligence. It is a Dreyfus’ leitmotiv, who strengthens his opinion by quoting Selfridge and Neisser, pattern recognition experts (p. 98), Samuel, reputed chess program author (p.110), or by pointing that the distinction between “essential and inessential operators […] is introduced by the programmers” (p. 116). This argument is much older, since it finally concerns Descartes’ (1596-1650) [28] animal-machine notion. This externally programmed machine idea was also agreed at some data processing precursors, like Lady Ada Lovelace [29] (1815-1852), Babbage assistant (1792-1871) in the programming of its punched card calculator. Later, Torres y Quevedo (1852-1936): “ […] challenging Descartes’ thought according to whom automatons could speak “reasonably,” wrote that Descartes was lost by the idea that automaton had to do reasoning itself, while it is its builder who reasons for it.” (René Moreau, Ainsi naquit l’informatique [The Computer Comes of Age], p. 29).

Actually, there is another way to consider the programmer’s intervention. It can simply be equivalent to any problem processing from any given starting point. That can cause sometimes a few confusions as Dreyfus actually point out about Miller neglecting data pre-treatment in Newell, Simon and Shaw’s experiments (Dreyfus, p. 187). But besides this mistake, this is an absolute legitimate device in an experimental framework. Contrary to the holistic myth, it is indeed possible to process any problem independently, by considering that the already present knowledge isn’t a problem. And it seems the whole problem is equal to something looking a lot like the sum of parts. More generally, intermediate steps are cumulative, and possible conflicts between particular resolutions are themselves particular problems. It is thus possible, in machine comprehension simulation, for instance, to short-circuit some steps: specially when computers were slower, it was possible to work in batch processing instead of real time; it’s possible to work directly on written document bypassing the oral computer analysis phase; or on correctly spelt words by eliminating inputs correction; etc.

This analysis type outcome, which is the general scientific research method in any field, will simply allow obtaining a model of the necessary device basis for machine autonomy. Let’s notice it is equivalent to assumptions analysis. By opposition at the ineffable intelligence idea, formal or material (in a robot), AI can be considered, as a model or a simulation of human behavior. AI is then simply a way of posing problems on human representation, or on his knowledge system. Let’s notice the interest of a model to realize tests, due to lack of wide human experiment possibilities, or simply due to historical experience reproduction impossibility. But it is also necessary to note that AI allows integrating several fragmentary experimental systems with a more or less natural language interface. Now, what holism could precisely reproach to (chemical, medical, behavioral, environmental, political, economical, etc.) modelings was precisely the limitations of its fragmentariness? The AI adversaries’ standpoint is therefore contradictory again when they refuse in the same time this division of labor, and its AI solution. But it is true they are certainly disturbed by the fact that human and machine limitations are formally identical, in the strongest AI hypothesis (cf. Jacques Pitrat, “Discussion,” in Dreyfus’ French edition, p. 433).

This thesis of external programming also raises the question of elementary control of arguments counterbalancing. It is perfectly possible to produce the hypothesis computer is not intelligent if it is programmed, providing the demonstration human being is intelligent because he is not programmed. This hypothesis does not derive from the other, contrary to these rhetorical distorted processes. In lack of such a balance, the classic mistake is made consisting in only getting up the case against [30].

The opposite hypothesis that human being can be programmed by computer has to be considered too. This doesn’t concern obviously a robot takeover. That reciprocity rests on the consideration, even present in AI adversaries, that our tools influence us. Besides Weizenbaum, who insists a lot on this point, when it suits him, it represents an element of the Heideggerian framework, wondering: “How our tools are parts of the background in which we can ask the question what is to be human.” (Winograd & Flores, p. 163). Data processing in general has this very characteristic, in its current phase, to aim friendly use. These two authors elsewhere also note that management programs do use accountancy terminology instead of an external language (p. 176).

Therefore, human being inside a culture is the result of his interaction with his tools. But this principle of human personality construction has to consider all our technological tools and not only artisan ones, like in the Weizenbaum’s beloved Mumford’s model. There are no reasons to stop the time before industrial revolution, like the Amish, or like the implicit rationalization of some backward-looking considerations in which: “the possibility of an ‘electronic library’ […] may make easier for a reader to find a book on a specific narrow topic, while reducing the ease of ‘browsing’ through shelves of loosely related material” (Winograd & Flores, pp. 166-167). Who knows why such a computer research is not possible – elsewhere “browse” is the term used (now) on Internet for the same operation –? As much that a physical support like book excludes a simultaneous presence in too numerous shelvings, while a cross-search allows many associations. The computer indexation advantage allows a lack of artificial limit: physics, budgetary, or competence (or arbitrary) librarian dependency.

Instead of limiting automation to industry and technoscience, all our conceptual tools of natural sciences, formal sciences, or human sciences, can be considered. From this point of view, AI is only a more complete integration of these tools, since it can also simulate human personality construction itself, or to assist it by a (more or less transparent) interaction.

Ritualization and Automation

The idea of programming is also common to human being and computer because no one can ignore this term is used for long in the human being case. Programming or automatism ideas are commonly used, in the conditioning meaning or to describe behavior ritualization. But this programming idea seems to be perceived differently, by AI adversaries, according to the biological or computer nature of its support (Winograd & Flores, pp. 99-100).

However, the question of biological programming (idem, pp. 100-104) is not so trivial. And the behaviors selection by evolution or learning can be simulate too. But it is certainly necessary to record the fact a large number of human activities are grounded on carrying out rules formalized by other people or by tradition. Human life is widely represented by this routine reality. Therefore, most of human beings would not be intelligent most of the time – what can also be acceptable. Human beings and computers cannot be distinguished on this point.

This programming idea also answers to the problem of action, equivalent, for a subject, to finding a direction in a reputed incoherent world, with more or less adequate conceptual systems as rules. Traditional and elementary strategy consists in grounding action on checked elements: precedent, routine, tradition, known technique. A new situation is often perceived as a risk. It is the same when an already known situation has produced an unexpected result. It can be emphasized too this ritual aspect could as well be called a procedure, or an algorithm, when formalization is known. Common consciousness is aware of the algorithmic nature of these situations, nevertheless human, as the French popular refrain reveals:

In the same way, when Dreyfus entitle one chapter of his book: “Orderly Behavior Without Recourse to Rules” (pp. 256-271), the ideas of habit or ritual come in mind at once, especially if we consider a rite as a reproduction of an action without mastering its explanation. A restricted meaning could limit its definition to religious phenomena, or secular, having some solemnity, but a lack of scientific operativeness [31]. Other behaviors could be called habits, automatisms, or quirks. But it does concern the same field of repetition. In both cases, the means-end relationship principle, questioned by our philosophers, can be considered as a demarcating criterion with algorithms, as ritualization is opposed to the idea of aware and reflective action, or rationality. When ritual is illusory, it can be called superstition, which researches a result and sometimes get it, but by useless ways. Examples are numerous out of the religious reference area:

i) Study of animal behavior: when, in an experiment, pigeons reproduce a behavioral useless extra sequence to obtain the food ration.

ii) Data processing: when a result is obtained by methods uselessly complicated, using heavy rules. It is the case of all nonoptimized programs.

iii) Management: business (or family) over-equipment, especially by fashion effect. Computer can sometimes be an illustration of it (cf. Weizenbaum, pp. 27-33).

iv) Languages: jargons, useless overcoding, or fake concepts, whose human sciences have made a specialty. Etc.

If the repetition idea can have a derogatory meaning (other devalued names: repetitious, common place, idée reçue, parrotlike, etc.), it is especially because of anti-robotization innuendo, often creeping within descriptions. To illustrate this passive behavior reproduction disapproval, an anecdote with (true or false) pedagogical purpose, is transmitted in methodological speeches: the story of a mother always cutting the roast before to put it into the oven. Some day her daughter asks her the reason why. As the mother got this habit from her own mother, she seizes the opportunity of a visit to ask her. Searching her memory hard, the grandmother remembered her youth, when her mother’s oven was too small to contain a whole roast for the big family.

But of course, something repeated is not inevitably false, contrary to the double bind becoming another common place adopted as well without critical hindsight. As what makes common place a mistake is the fact we knows (or we believe) they are false.

Let’s notice this ritualization situation also product a cognitive activity of rationalization. These more or less valid exegeses cannot be questioned as for their humanity. A body of professionals (or enlightened amateurs), and institutions, can be composed in organized religion, or more generally in sects. Indeed, it is always possible to find a social function to a ritual. It is equivalent then to the levi-straussian justification of the myth, in which any explanation is better than uncertainty, and represents a consolation. This rationalization appears to be the only difference between human being and machine (but nothing forbids simulating it).

Computer Assisted Artificial Learning

The issue of external programming can also simply be equivalent to learning. From this standpoint, human or artificial system learnings are equivalent, at least concerning information supply. Despite Weizenbaum’s irony about knowledge as “organization of ‘facts’” (p. 207), most of what can be explicitly known by human beings already can be encoded for computers one way or another. Knowledge can be tritely considered as equivalent to a database content. And, since Aristotle until the Encyclopedia, science is what is teachable. Since Encyclopedia, we have entered the era in which knowledge revealing has taken the explicit form of instructions for use (although, it is necessary to take the trouble to read it). Thought or technique, are no longer equivalents to the classic form of Zen initiation. This last impregnation learning mode actually proceeds with the least explicit communication possible and despises clarifying. This attitude, usual in philosophers, had already been noted by sociology:

“It seems the encoding of the pedagogical work doesn’t escape the ambiguity generally characterizing the academic discourse on academic rules. Indeed, the rule, immediately stated, tended to be lowed to the subordinate rank of a technique (“just massage”).” (Louis Pinto, Actes de la recherche en sciences sociales [Research Proceedings in Social Sciences], n° 47/48, p. 21).

The similar refusal of popularization (by authors like the French philosopher of sciences Lévy-Leblond for instance) is grounded, it can be granted, on a logical argument: there is no second-rate knowledge, people know or people do not know! And therefore, a real scientific training is necessary. This idea is referring at the lamentations about low cultural level, or low scientific knowledge distribution, even with graduated persons. However, the same authors challenge encyclopedism and interdisciplinarity under the dilettantism anathema. But even this logicism is obviously absurd. First, because there is, anyway, a scientific knowledge distribution, no matter what is told, more or less voluntaristly or spontaneously, due to existence of popularizing people. Despite mistake possibilities, nowadays, a good part of the world population is being acknowledged of many scientific developments, like blood circulation, schematized by Harvey only in the 17th century. And it is in the same way for most of knowledges, especially medical like this one.

The problem raised here is rather a correction of educators’ illusions. They sometimes seem to imagine that pupils record every word, precisely like computers! This scholar disappointment reminds us the religious one when cleric noticed that the obediently mumbling of canonical dogma left traditional superstitions or heretical interpretations remain. These academic prejudices rather mean that teachers don’t do their knowledge controlling job, and/or they usually evaluate pupils according to subjective, intentional, overall, etc., criteria; i.e. by having their favorites. Or these professors maybe imagine that teaching of a fashionable or critical theory exempt them from teaching basic knowledge, which would therefore be acquired by thought transference, and preferably by those flattering them by adopting their current fads.

Let’s notice that the very scientific knowledge distribution has removed to science its mystery – precisely so little scientific! Therefore, the world disenchantment has stroke science in its most spectacular form, “incredible but true!” style. This form can be admitted as narrative dramatization. Scientific texts do not have to be inevitably boring like some people seem to think of them. This soporific approach seems to be the rule in Europe by opposition to the USA. This boredom symptom could characterize the literary approach of science. A course lived under the aspect of a chore could have associated this characteristic to scientific work [32]. Adversaries of science then can worship boredom, this apparent sign of scientificity.

More comprehensively speaking, the explicit external programming allows making a model of learning, by studying what is necessary for moving up from one level to another. This hierarchical conditions of possibility interlocking is precisely equivalent to programmed education. This programmed progressiveness of acquisitions can happen without computer. It is simply equivalent to any well-designed school system, what can be utopian all right! And the identification of minimal learning conditions, necessary assumptions or primitives, is once again experienced by the AI opponent on the obsessive regressive mode (Dreyfus, p. 256). This standpoint logically results to the negation of any pedagogy. Isn’t it again a reminiscence of the Hegelian logicist paradox in which Marx and Engels aiming at Feuerbach, in their third thesis of The German ideology, stating that “it is essential to educate the educator himself” [33]? As we can see, such an approach specially regresses in the classic philosophical chicken and egg enigma. Poverty of dialectics!

The computer can precisely be an excellent way of control of these minimal, optimal, learning conditions. This Simon and Schank’s interpretation of “a robot that is to have a childhood” (in Weizenbaum, pp. 202-203) is also the Turing’s former standpoint which doesn’t deserve to be considered on the sacrilege mode. Automatic learning simply aims to allow knowledge input in declarative form (by interactive dialogue), rather by procedural or algorithmic programming, especially to save work. Winograd and Flores (p. 132) notice too that H. Dreyfus and S. Dreyfus (in Mind over Machine, 1985) consider that only the novice stage is using rules to acquire a competence. But what does authorize them to minimize this first level? Wouldn’t it be rather necessary to consider that an ability to create or to use rules, once established, is applied to the next stages (in which we are again in a novice state)? Ultimate levels of expertise could be characterized by the oblivion of these acquisition mechanisms, or simply by ritual sclerosis. Then what phenomenology seems to describe is only the established automatism, as a property of biological system.

Numerous philosophers also seem to believe having an overall or intuitive comprehension of Heisenberg’s relation of uncertainty, of Gödel’s theorem, and other abracadabras. The reference to quantum mechanics or other uncertainty principle, besides its frequent irrelevance, is also an imposture in pedagogical terms. Like for Latin of Molière’s fake physician, one could utter today: “Do you speak quantum physics? No?… So, as Niels Bohr said….” But mathematicians or physicists truly understanding this theory have had to learn it by elements. And more generally, school programs are actually organized in steps. An overall comprehension would be a huge saving for the national education budget, or the UNESCO one. This refusal to reckon the collective or individual (phylogenic or ontogenic) progresses in term of steps can only lead to the conclusion of the bad master (see in the following chapter).

Bringing external rules to light is the way pedagogy works, but also the way pupils do, contrary to what philosophers seem to believe. The assistance by a language grammar, or by any formal systems, represents the characteristic mean of human learning. Chimpanzee is capable of learning, and invention, but doesn’t help its babies, when it sees them fail (as reminded by André Langaney, director of the anthropology laboratory of the Musée de l’homme in Paris).

Social and Professional Reproduction

In this same programming standpoint, the sociological interpretation of the social role as reproduction can be reminded – without being necessary to question it as critical sociology does. Elsewhere, the sociological tradition rather justifies this social reproduction:

“Well far the education has for object solely or mainly individual person and its interest, before all it is the ways by which society perpetually renews the conditions of its own existence.” (Emile Durkheim, Education et sociologie, p. 41).

In this connection, the first need of repetition is the professional skills. By the past, professional learning precisely demanded long imitative practices, usually in the very family. It was widely characterized by impregnation, contrary to the academic formalization. Phenomenology – and Dreyfus especially –idealize the archaic model instead of thinking it. This precisely characterizes modernity takes its origin in the work of the Enlightenment initiating the Encyclopedia, and its leading to expert systems, for instance.

Possibly, from observing the existence of computer information processing, could make consider that AI concerns only machines, and make give up the comparison with human beings. As much nobody questions that machines can process more numerous elementary data than human being can. But before data processing, a SNCF (French national railways) employee, for instance, could actually remember all schedules, fares and distances, of his sector. More generally, it is notorious that intellectual activities rest also on the systematic memorization, which is therefore admitted about human being with the same depreciation:

“Students sometimes prepare themselves for examinations in physics by memorizing lists of equations. They may well pass their examinations with the aid of such feats of memory, but it can hardly be said they know physics […].” (Weizenbaum, pp. 141-142).

But despite some conventional lamentations, the result of such a learning is obviously positively judged and correspondingly penalized. Oddly enough, Dreyfus (pp. 292, 300-301), or Winograd and Flores (p. 161), admit the existence of these mechanical behaviors. But it is obviously to minimize them, or to distinguish their nature. For the last ones, the blindness by anti-automatic, anti-repetition prejudices, simply seems to generalize to all human being the managers prerogatives, in the division between design and execution. Is this a reminiscence of the Marxian rejection of the division of labor that makes idealistically deny facts? Actually, calculable or simple tasks can be also intellectual ones [34]. A program represents therefore a simulation of this human competence. But even for design tasks, many operations are already automated by (Conversational) Decision-Making Systems, as admitted by Winograd & Flores (p. 173).

AI adversaries could also admit repetitious operations are profitably mechanized, since they admit it is possible, because the least honesty is to recognize what is already computerized, even reluctantly:

“Programs which simulate investment banking procedures and the like have no bearing on Cognitive Simulation or Artificial Intelligence at all. They merely show that certain forms of human activity are sufficiently simple and stereotyped to be formalized.” (Dreyfus, p. 343, note 1).

The behaviors described by all the concepts concerning reproduction are obviously very numerous and frequent since only inventions (or reinventions) escape it. Here, for computer, the problem is actually the discovery one. But the individual knowledge is not characterized by permanent invention, especially because of communication, transmission, and imitation.

The philosophical reservation about behavior foreseeability seems to be grounded on the undecidability abracadabra, generalized to all human problems here. But it is impossible to characterize human being by incalculability, as the phenomenological argument seems to propose  – without excluding a large number of persons of humanity: subordinates or even technicians, and more generally the quasi-totality of human behaviors, excepted inventions, which have a tendency to be reduced to poetry by phenomenology! It also could be called self-programming capacity. As what characterizes the making of a personality is rather the assimilation of different cultural and family learnings, adjusted by personal observations and by confrontation to the real. Finally, in its concrete practice, intellectual work most often consists in knowing what rule is applicable. This offends the other phenomenological principle of absence of rules.

More, it is to be noticed human being himself has long been considered as having no original idea, kind of divine monopoly. Human soul itself was supposedly animating human beings as saying so from outside; from which the puppet image, in the hands of a personalized destiny or passions, etc. The phenomenological mistake seems to be contradictorily grounded here on the obviousness of the humanistic and rationalist traditions, giving the first role to conscious will and freedom.

The already observed repetition extension in human matters even reduces the philosophers’ pretension to nothing. The bias is most possibly grounded on their own professional practice which precisely consists in being a recording chamber of new ideas – preferentially established authors’ ones. Documents studied by philosophers give us an impression of innovation, by masking repetitions. It is pretty obvious that sociologists, historians, or psychologists – and especially child psychologists –, are obliged to work on repetitious practices, including mistakes or archaisms. More, the calculable rules, once weren’t still calculated, even if it is possible, afterwards, to minimize the solution. And precisely, the field processing the making of this kind of solutions, algorithmic ones or not, is problem solving all right.

Specifically, innovation or individualization have been very well assimilated like human ideal, and adopted in a calculating way by advertising. In the banking investment example, a slogan for the bank CIC, in Paris, at the beginning of 1994, can be quoted: “You hate ready made answers. You are right. Anne L… (its photograph), banking counselor to the CIC Lafayette.” Anne L., despite her certain humanity, can be imagined proposing standardized financial products, rather than financial innovations of her own initiative. In businesses, and customarily speaking, the human autonomy has limitations as external as those for current computers. In short, this kind of real behavior cannot be formalized by ineffable philosophical criteria.

Innate Learning

With some indulgence, philosophers’ reservations about clarifying can be admitted concerning situations in which formalization is not established (providing no denying of what it is). But the mere idea of learning seems to have absolutely nothing to do with phenomenology, certainly because of Gestaltist innateness. It is pretty obvious that swimming can be learnt all by oneself (Dreyfus, p. 60). But learning without manual, without rule, without imitating swimmers, and without lessons, only means to invent (or reinvent) swimming (what supposes previously a few number of drowned)! The reinvention principle obviously cannot be generalized. Even if active learning is constant, it proceeds from an assimilation of a model, and is at least under an extensive control. In the present-day world, the cultural submersion in technique makes often forget this artifact nature of the environment. This negation of any analysis is found in the mythical idea of the human chess play in which “situational understanding is PRIOR TO aspect specification” (Dreyfus’ uppercases, p. 31), contrary to machine. Dreyfus elsewhere recognizes, in additions of the second edition, the value of learning, even in chess game:

“I suggest that this ability is the result of having a sense of the developing game. While no doubt correct, this now seems to me an inadequate account, for it does not take into consideration the fact that to develop this ability to zero in, chess masters must play thousands of actual and book games.” (Dreyfus, p. 30).

Let’s also draw the attention on the fact that learning by repetition is most often called an automatism. The explanation of learning by intuition is the same ad hoc rationalization type, whose the philosopher have made a specialty, like when he speaks about “a higher order of tacit understanding” (Dreyfus, p. 253). And this learning refusal idea is enough spread, contradictorily due to their function, in some academic circles. Many sects too develop behaviorist strategies (‘take the plunge’), grounded on this innate knowledge platonic illusion. Such archaisms justify the French Sociologist Pierre Bourdieu’s irony, in his reminding of the famous joke, attributing (for instance) to an old marchioness: “Good manners [“education” in French], cannot be taught!.” Even if it plays however a bit too much on the double meaning of good manners and teaching, for having a coherent interpretation with its topic. This sociologist can be suspected of corporate defense of his academic clientele, if he doesn’t specify that the standards of behavior are also taught as explicitly (“do not put your elbows on the table” [in France], “say hello to the lady,” etc.).

Even for intuition, the overall vision doesn’t exclude the fact that the situations make sense by abstraction and encoding of a limited number of relevant elements. To cross the street, for instance, it is actually possible to cross mechanically. But it is preferable to watch both sides, and to consider only some data (vehicles distance and speed mainly, and not color or mark, etc.). Children reminded by Weizenbaum, Dreyfus or Searle have to assimilate these nonholistic ways if they want to survive in modern cities. It is also the case of an inhabitant of the tropical forest unfamiliar with these realities, who would have to learn them, although he is an adult. In the same way, in order to master these skills, it is necessary not to forget taking into account the inversion in the order of quick looks when the right/left circulation system has changed. More, in our modern societies, information is pre-encoded too in signals for pedestrian or for cars, according to the (labeling) supermarket principle (see above).

The only principle considered by philosopher rests on pattern recognition, generalized to the whole intelligence. This idea suppresses the classic visual/discursive opposition. This distinction can be clarified by that traditional artistic forms compared to their current evolution. All the traditional work of literature and thought, or actually of AI, can be characterized by the transformation principle from visual into articulated discourse. The current artistic reaction against this principle is certainly the consequence of easiness provided by the movie picture. The visual option can be explained by chattering excesses (Shakespeare’s ones for instance, or 19th century never-ending descriptions). A result was the refusal of built dialogues in current movies or plays, with the possible intellectual guarantee of theater of the absurd. But this discursive refusal, in science or in fiction, can be characterized as a distribution refusal, an esotericization, because mastering these sophisticated forms of allusive or purified arts supposes an obviously laborious and analytic learning. Elsewhere, the process is hardly subtle, since it is enough to take on purpose the exact opposite of the previous literary forms. This work is apparent to a parody, under an intellectualized form.

It is necessary to notice again that the philosophical idea of learning is a little muddled. In order to refuse the simple knowledge accumulation, or communication, it seems to conceive a kind of soul communication:

“Wittgenstein points out that if we simply point out at a table, and say ‘brown,’ a child will not know if brown is the color, the size, or the shape of the table, the kind of object, or the proper name of the object.” (Dreyfus, p. 110).

This Dreyfus’ hypothesis still comes from a cognitive view searching absolutely a human/machine difference, or the analysis refusal. Would the word brown trigger an innate cognitive trait? If Wittgenstein had studied the question with children, in the genetic psychology framework (and not as primary schoolteacher influenced by philosophy), it should have not built unwarranted speculations. In the most precocious learning stage, such confusions indeed exist. The child (or a foreigner out of school context) will have to proceed in this contrastive way, by proposing any hypothesis, color for instance, to fit another brown object. If the answer is yes, since it is chocolate, his hypothesis will be confirmed. If he had chosen the form and he’s showing a red pedestal, his hypothesis will be denied. This is therefore perfectly programmable, since we do not forget that language learning undergoes an external control (contrary to introspection), what is equivalent to a behaviorist programming, not to say Pavlovian.

The interest of the AI issue is therefore rather to make obvious the competition possibility between human being and machine in this learning area. But when actually performances of AI can be observed, its adversaries then deny the automatic learning possibility. Winograd & Flores (p. 100) distinguish different learning types (strengthening, selection, and induction), and discuss their automation capability. Besides the fact they neglect the simple accumulation, they limit obviously themselves to express reservations about process perfectly running, although perfectible. The Dreyfus’ claiming of the discovery ability of human mind alone (p. 99) is an illusion denied by the automatic processes above. The philosophical competence itself is specifically built by a guided acquisition.

It seems too the phenomenological humanism make each person enjoy qualities of any other, by a pseudo-logic principle: each individual benefit of all the qualities of his species, therefore the species immediately benefit all the qualities of any individual [35]. Human being could have built a new idol and would refuse then the AI challenge. So, in front of the human meaning acquisition problem, the Weizenbaum’s religious solution simply consists in translating psychological phenomena in symbolists, even mystical terms:

“A catastrophe, to use Erik Erikson’s expression for it, that every human being must experience is his personal recapitulation of the biblical story of the paradise. For a time the infant demands and is granted gratification of his every need, but is asked for nothing return.” (Weizenbaum, p. 211).

The weaning in question lies here, for human being, in the learning of exchange and autonomy (excepted the fact this elements could possibly be overvalued in American culture). Such a symbolization could also be admitted for a robot since its internal representation includes the history of its former states too. A genesis of myths can even be considered as structuring these opened to outside, inductive and incomplete internal states. Which would constitute a simulation of the cultural humanity history.

For an autonomous robot, the problem can be expressed as the learning of communication or interaction constraints. However, we must notice its situation will be most possibly inverted, since in a first time it would work for them. This can reduce the analysis of its autonomy to an analogy of emancipation, and allows considering it as a modeling of human corresponding situations. The pure weaning standpoint will only consist in the point the knowledge transmitted by human beings would have reached its limit. The robot will be then in the situation of someone in front of a problem who doesn’t know how to solve it, which is the institutional situation of research workers.

Finally, the explanation by intuition characterizes only who has succeeded. It will remain therefore to explain how someone can fail. There, the aristocratic recognition discourse can be acknowledged, consisting in excusing the mistakes of those who were supposed incapable to fail. This over-valuing method of the good pupils and depreciation of dunces is a notorious professorial bias. Studies in sciences of education confirm that mistakes of firsts are excused as temporary weaknesses, and successes of seconds are minimized as exceptions, they ought to confirm at best (out of the suspicion of cheating at worst).

Conditioning or Trial-and-Error

When learning is denied, the only rationalization available to explain knowledge acquisition is again a kind of biological reductionism, that looks a lot like an improved Pavlovian conditioning. The theory reminded by Winograd and Flores makes this point of view an elegant solution (as scholars say at defending the most absurd theses):

“Maturana’s understanding of an organism’s relation to its environment leads to an epistemological problem. […] If we look at the nervous system as closed, we must ask how an organism comes to have any knowledge of the world. […] Learning is not a process of accumulation of representations of the environment; it is a continuous process of transformation of behavior through continuous change in the capacity of the nervous system to synthesize it. Recall does not depend on the indefinite retention […] but on the functional ability of the system to create […] a behavior that satisfies the recurrent demands […].” (Winograd & Flores, pp. 44-45).

Who knows what this overcoding (which could at best be admitted as equivalent definitions) does change to the problem of human knowledge perceived here in terms of conditioning, or self-conditioning as it happens. This reductionist neurobiological verbiage obviously has a purpose of monopolizing all knowledge about living beings. This option excludes however any consciousness or any representation, while wanting to benefit from the phenomenological warrantee. Wouldn’t it be useful to see in this reductionism a reminiscence of the classic materialism, pushed to its biological term?

The biological imperialism can simply origin from a kind of epistemological conformism. The biological model has corrected, indeed, insufficiencies of mechanism – with processes like feedback, for instance. It was elsewhere the origin of cybernetics, as AI precursor. But it would be absurd to generalize biologism as the only comprehension model. Bio-design is only a marketing fashion. And especially, mechanism still remains the causality model, at least by precedence. It simply means that phenomena are connected to others. And it still can be used to challenge occult explanations or mysticism – what doesn’t allow, obviously, biologism. At best, these differences are only about personal sensitivity to some symbolisms, like those classified by Gilbert Durant (in Structures anthropologiques de l’imaginaire [Imaginary Anthropological Structures]).

Gestalt psychology can also be used as scientific (expired) guarantee by the phenomenology for the refusal of the trial-and-error approach. Dreyfus, relying on Wertheimer (p. 114), thinks to justify intuition by these mutual quotations. It can be thought that phenomenological intuition is the other name of the perceptive restructuring in totalities. But these undeniable possibilities are especially usable in artificial exercises. In a real problem, trial-and-error is the actual process. The Piagetian genetic standpoint integrates this dimension to learning, and specifies the origin of this mistake:

“Of the very fact Gestalt psychology forces itself to consider [perceptive structures] as coming directly out from situations as such without historical genesis, Köhler be obliged to remove from the field of intelligence, on the one hand the groping around before the finding of solutions, and on the other hand correcting and checking after.” (Piaget, La psychologie de l’intelligence, p. 73).

Let’s note too, from the standpoint of the body commitment, that the actual action too proceeds by trial-and-error. These action patterns can be conceptualized, thanks to the use of operators:

“In front of a new thing, the child tries the last acquired schemes one after the other (grasp, hit, shake, rub, etc.), which are used as sensorimotor concept, if I may use the phrase, as if the subject tried to understand a new thing by using it.” (Piaget, La psychologie de l’intelligence, pp. 111-112).

On this mode, AI could represent a classic problem, like reaching a banana for a monkey by using a box as step. The modeling of the problem in a program consists too in successively trying operators grasp-banana, walk, push-box, climb box, etc. (cf. for instance Ivan Bratko, Programming in PROLOG for AI, pp. 66-71). Even if the computer hasn’t invented these operators himself that means such a situation can be represented. A robot in an environment thus has some basic competence; it can possibly combine to master the surrounding world. Winograd and Flores, on the same refusal mode of the trial-and-error strategy, develop like a leitmotiv (pp. 36, 73-74, 99, 102-103, 137, 165-166) the reference to the Heidegger’s “breakdown or unreadiness-to-hand” theory. Oddly enough, they do not seem to notice that the breakdown idea could constitute the philosophical jargon translation of the trial-and-error strategy, because what is this “situation of non-obviousness, in which the recognition that something is missing leads to unconcealing” (p. 165), except the circumstance in which an error is perceived?

Elsewhere, in the historical and epistemological context of the period Heidegger worked in, it is necessary to emphasize that bringing breakdown to light in the routine interaction with the environment actually notice a limitation of the functionalism in anthropology, because anything is not always well functioning in a society. The Heidegger’s originality on this point would have been not to elude these breakdown, if he had not remained contradictorily too dependent of romantic holism (which certainly motivated his membership to the Nazi fusional delirium). Elsewhere, these breakdowns allow innovation, and make incomprehensible the anti-technological backward-looking attitude of the phenomenological current. Finally, in our two AI adversaries (for whom another contradiction won’t make much difference), the anticipation of mistakes is considered as an indispensable methodology for good programmers (p. 166). The problem seems to amount to a labored denial attempt of what is finally admitted.

Indeed, when someone has practiced programming, what implies failures or programs limitations have been met, it cannot be asserted that some obstacles are not perceived, as Weizenbaum does. Philippe Breton (La tribu informatique [The computer tribe], pp. 37-39) mentions the actually very severe technique of assistance exclusion during one year developed in (former) computer centers. Dr Jerry Weinberg, one of the debugging (programs correction) masters, ruled this way against whose excessively requesting him, without having previously consulted the mistake checklist procedure. Breton, as methodical sociologist, sees there a kind of rite of passage characterizing the DP men or computer scientists tribal group. Indeed, the checklist can be used as judge to define professionals wanting to learn by correcting themselves their mistake (instead of to feel persecuted by the machine). They fulfill the Winograd and Flores’ criteria for the “situation of non-obviousness” above. If professionals do not want to be disturbed for nothing, does the relational apostles prefer make their (always the same) mistakes corrected by others?

More generally, it was not already certain at all that consciousness of a situation appears only in a case of a breakdown, neither any object appears only when it comes to be lacking or to be hidden. Reservations can be made concerning this Heideggerian confusion between representation and communication. Which can possibly mean that the acting subject doesn’t have to speak to himself (like in the Piagetian model of consciousness). This ignores the necessity of action preparation, shown explicitly, for instance, when athletes mimic the movements to realize. And that doesn’t concern complex activities, or those coordinating several agents, which demand communication, especially with important project (other Heideggerian topic). But information theory, challenged by philosophers, is well enough defined by the communication of what is not obvious. Moreover, conflict situations, appearing in these breakdown circumstances, do not depreciate data processing, because the computer has no emotion regarding negatives stimuli. And in human beings interactions it is necessary to manage some consequences, not only positive. The phenomenological standpoint could regress to old functionalism.

Robotic and Human Socialization

The Helen Keller’s case, popularized by the movie Miracle in Alabama, about the communication learning of a blind, deaf and dumb young girl, is very well known. This case can be considered as an excellent simulation of some problems of learning, as pathological cases often allow. More precisely, letters of the young girl (Helen A. Keller, Deaf, Dumb, Blind, pp. 197-300) show first the mechanical characteristic of communication learning. They can reassure the DP men or computer scientists in front of the AI infancy. Human hasn’t anything to envy to the computer when acquiring fluidity [36]:

The numerous other letters provided in annex of the Helen Keller’s book show the learning progression. Anyone can even clearly observe the progressive concealment of the mechanical characteristic of formulae, stereotyped but syntactically and stylistically more correct: “I am happy to write you this morning” (letter of 2.1.1888) is equivalent to “Helen is going to write to mother a letter” (letter of 12.7.1887). This kind of ritual is simply an opening, said phatic in linguistic, of the communication channel, like hello [web] at telephone. More generally, AI experts would profitably compare their formalizations with real human learning instead of being imposed literary paradoxes as criterion

Reinterpretation of Simulation

Another very informative case study is the ELIZA program (called DOCTOR too), that has grounded the Weizenbaum fame. This program could talk with the user, by simulating a dialog with a Rogerian psychotherapist, whose essential technique consists, mostly, in returning to the speaker its own words. The analysis Weizenbaum has then made of its former work aims to criticize the validity of AI, with a claimed authority. This standpoint allows him, he thinks, to despise laymen’s naiveties:

“A number of practicing psychiatrists seriously believed the DOCTOR computer program could grow into a nearly completely automatic form of psychotherapy. […] I was startled to see how quickly and how very deeply people conversing with DOCTOR became emotionally involved with the computer […]. Once my secretary, who had watched me work on the program for many months and therefore surely knew it to be merely a computer program, started conversing with it. After only a few interchanges with it, she asked me to leave the room. […] Another widespread, and to me surprising, reaction to the ELIZA program was the spread of a belief that it demonstrated a general solution to the problem of computer understanding of natural language. […] This reaction to ELIZA showed me more vividly […] the enormously exaggerated attributions an even well‑educated audience is capable of making, even strives to make, to a technology it does not understand.” (Weizenbaum, pp. 5-7).

How not to notice, on the contrary, that it is really a good simulation of the Rogerian psychologist? This technique consisting in making people talk is perfectly well imitated, which was the initial condition. ELIZA therefore empirically formalizes this practice, without making its explicit theory (where from the mistake comes?). Albeit the rephrasing theory, used by ELIZA, is considered, in current communication workshops, as the human communication and listening ultimate!

The Weizenbaum’s book and, consequently, the other AI critics grounded on the self-criticism of this AI authority, are finally grounded on a mistake of interpretation, because the Weizenbaum’s first astonishment is denied by the emotional involvement that he mentions in the second. More, the secretary’s attitude can be explained more simply by the confidentiality of information given to the machine. This practice is then characterized by its concern to carry on with the experience by the use of real data. Their confidentiality is therefore independent from the knowledge of the program nature, but it only depends on the external observer (what is an elementary parameter of psychological studies, of which Weizenbaum is not an expert). Not to understand the secretary’s attitude, means that programmers knowing this program nature are obliged to cheat when witnesses are present. In their practice, considering the initial information about this program nature depreciates the test. He reduced an experimental situation to the simple program development with ad hoc interactions. On the contrary, the methodologically correct secretary’s attitude allows demonstrating the Turing’s test. Since this test consists in telling that machine intelligence will be shown when dialogue with a computer won’t be distinguished from human ones. The decisive criterion concerning ELIZA, accordingly to the Turing’s test, is thus defined by the speaker’s appreciation, and not by the fact it is known, elsewhere, it is a program. This kind of flagrant methodological fault is often accompanied by a theoretical pretension, as a minimal academic rhetoric, conscious or unconscious:

“Most of existing programs, and especially the largest and the most important ones, are not theory-based […]. Their construction is based on rules of thumb, stratagems that appear to ‘work’ under most foreseen circumstances, and on other ad hoc mechanisms that are added to them from time to time. My own program, ELIZA, was precisely this type.” (Weizenbaum, p. 232).

Then, it is obviously as for human knowledge as for empiricism. These mistaken assessments are precisely equivalent to the author’s incompetence in psychology. He can speculate, as educated amateur, but with no guaranteed results, nor with any external scientific authority. This partially validates his remark about his own philosophical limitations (Weizenbaum, p. X & p. 8).

Automation of Human Being?

This ELIZA’s performance happens to be a perfect human competence automation, really operative. Human intelligence cannot therefore be interpreted as an ineffable competence, inalienable characteristic, neither as a mark of interest for others from psychologist, empathy (ingenuously) invoked by Weizenbaum (p. 5) against the computer-assisted therapies idea. Without denying human interest for patient, it would be hardly credible not to consider the possibility of getting into a rut, like in any professional practice. Considering this human constraint, if professional psychiatrists think technology can automate some operations, why cannot they be able to judge their own working field. And if as therapeutic assistance, operativeness is observed, it is therefore necessary to consider, either its effective validity, or the validity lack of therapies grounded on this very operative criterion.

Elsewhere, psychotherapeutic practices often shows a kind of mechanical application from their human agents, therapists or public at large uses (magazine personality tests especially). Indeed, conversely, a desirable individualization has been detected by therapies. But they have exaggeratedly developed, from the end of the heighties, a hyper-individualization discourse. It is pretty obvious that the ineffable philosophical contagion has made its work, especially for those coming from the phenomenological camp. But, the interest of the automatic psychotherapy hypothesis challenged by Weizenbaum would precisely be to consider many parameters, which cannot be made by a single person, especially when the therapist membership to one of this field numerous sects precisely excludes the holistic approach requested by phenomenology. More, this technical progress, like any others, would first allow a service generalization. Human relationships can be preferred. But the real is more specifically considered by partisans of AI:

“It isn’t that just the natives of Ulan Bator don’t have the same access to medical care as the natives of Los Angeles; it’s that the natives of Fresno don’t have it either, and poor people in Los Angeles aren’t as fortunate in their medical attention as well-off people. If the idea of a mechanical doctor repels you, consider that not many human feels that way.” (Edward Feigenbaum, Pamela McCorduck, The Fifth Generation, p. 88).

This possibility has to be considered even if a standard product has a lower quality than the first grade one. It is already the case, from great medical specialist to general practitioner, or from haute couture to ready-to-wear, from great vintages to cheap wines, etc. Elsewhere, the elite ostentatious services inflation is not an absolute reference. Even in the human/machine opposition case, it is sometimes possible to invert the hierarchy of quality:

“Studies in England showed that many humans were much more comfortable (and candid) with an examination with a computer terminal than with a human physician, whom they perceived as somehow disapproving of them.” (Edward Feigenbaum, Pamela McCorduck, The Fifth Generation, p. 88).

This is the apparently surprising result, but perfectly logical, of automation by human being, and individualization by machine. This result is noted too in the Computer Assisted Learning field by this other opponent to AI:

“Students enjoying computer put forwards reasons that are a comment ‘in negative’ of the ‘human’ teachers behavior. ‘Computer, they say, allows to work at one’s own rate, it doesn’t cause/induce hindrance while making mistakes,’ it provides an immediate appreciation ‘as for the accuracy of answers’. More, quality appearing beloved of a lot of people and that betrays an implicit teacher criticism, computer doesn’t make ‘subjective evaluation grounded on personality’ of the student. […] Some teachers are thus confronted with an unexpected thought on their usual pedagogical methods.” (Philippe Breton, La tribu informatique [The computer tribe], p. 20).

The automation role appears in objectivation of human practices and their limitations. When this last author remind us the tribal context, the mechanical aspect of ritual communication in traditional societies, even current ones, can be reminded too. So much that some sociologists even define madness as unpredictability in a social framework. The only socially tolerated originality is often professionally carried on by the artistic professions. This can explain the philosophical bias, which seems to confuse the valorization of creator status with the description of his behavior.

Inference about Human Intelligence

The important problem arose by ELIZA is also how can we know a human being is intelligent. The petitio principii or the phenomenological human intelligence decree can only flatter idiots, but doesn’t constitute a guarantee. The majority belief criterion can be obtained by the propaganda of almost no matter what absurdity. If the Weizenbaum’s thoughts about belief, like about many other matters, are actually really interesting, he is always wrong to consider that they grist to his mill. Like for his program, he gives incorrect interpretations. He would better follow historical and empirical method of building his mill on a waterway, rather than built a canal trying to amortize his mistaken investments. Elsewhere, he himself notices that intelligence is attributed in function of hopes interpreting more or less neutral or polysemous answers:

“This belief [in fortune-tellers] is not a conclusion reached after a careful weighing of evidence. It is rather a hypothesis […]. No light is permitted to be shed on any evidence that might be disconfirming […]. Within limits, this is a quite normal and even necessary process.” (Weizenbaum, p. 189).

If the intention recognition is pathological, and it is the case with some extrapolations, it is then not specific to relations with the machine. Here is an illustration of the personal opinion confirmation quest, according to the principle brought to light by the sociologist Paul Lazarsfeld about the press reading choice, pre-selected on ideological criteria (we can call it: Lazarsfeld’s principle). But precisely, this process can also characterize a way of thinking type: the intentionalist’s one. The classically pre-scientific mentality that doesn’t take account of denials, or challenge the objectivity (materiality, pragmatism…) criterion can be recognized.

We can be also in front of a paradoxical situation: if we are always uncertain about the mental state of others, can’t we conclude there isn’t any mental state at all. The human behavior simulated by ELIZA would unify then to the program’s lack of feeling. As we can see, in this standpoint, intentionality and mentalism disappear. It doesn’t remain anything else to go so far to other philosophical superstitions, because any interpretation, hermeneutic principle of phenomenology, is grounded pretty obviously on signs. If it is about message or visual aspect, it is precisely the hypothesis of appearances reproduction that is raised. (What solves incidentally the computer simulation possibility question, since it is appearances that are simulated). All the question of what is behind appearances remains pending, despite phenomenological tricks.

Emotion Existence/Value?

The negation of all thought or feelings determination is paradoxical in phenomenologists. Their affirmation of the body coexists with a traditional depreciation of the vile matter, as the irony below is clearly revealing:

“On this [strong AI] view, any physical system whatever that had the right program with the right inputs and outputs would have a mind in exactly the same sense that you and I have minds. So, for instance, if you made a computer out of old beer cans powered by windmills; if it had the right program, it would have to have a mind.” (Searle, pp. 28-29).

Nevertheless, Searle would not be offended at all to consider that he has the same mind that human beings apparently moved by beer cans, and would be even flattered by the literary allusion to windmills. It would have therefore to phenomenologically consider that the human mind and its works are grounded on the satisfaction of needs, because the question about whether or not a computer has a thought, mind and feelings, can be rather reduced to what are these phenomena. But, we already have seen that these terms are inferences from human behaviors, as Weizenbaum clearly admits himself (p. 126). And precisely the same inferences are observed, by anthropomorphism, in the program ELIZA case.

Thanks to this intentional method, the computer can sometimes, nevertheless, be considered, by discursive facility, as intelligent. It is only blamed then for having no feelings, nor commitments (Winograd & Flores, p. 106). Intelligence seems then reserved to complex decisions, as if simple actions or a unique solution had lost this quality. Aren’t all the other situations human too? Against this partial hypothesis, Weizenbaum demand a total identity of the machine with its refer human:

“Even if a computer could simulate feelings of desperation and love, is the computer then capable of being desperate and of loving.” (Weizenbaum, p. 200).

One may rightfully ask Weizenbaum why? Would the purpose of robotics be the literary situation in which it is about falling in love with an anthropomorphic robot, which would indeed liven up blow-up dolls? Without excluding completely this possibility, the objective of the usual human behavior simulation seems sufficient to the robots moving in a human environment. If the question of this sexual or affective simulation raises, that model could be used to practice all right, for the purpose to avoid some difficulties. But in case of conflict with the robot, who will be given the custody of the children?

This possible human/machine identity allows us to ask another question. How to be sure that human being doesn’t simulate, in the everyday meaning of the simulation term? The attribution of feelings to others in general is most often grounded on the reciprocity. The experience of the program ELIZA seems therefore a good simulation of this recognition of self in others, with the revelation of its limit. The question of feelings, in human sciences terms, can be translated into the opposition below:

i) On the one hand, expressive transparency of affective signs is postulated, while cultural or simply individual differences make them opaque. This situation of the arbitrary of the feelings encoding (albeit this expressive arbitrary is relative according to an anthropologist like Eibl-Eibesfeldt) can lead to skepticism, to relativism, to the Sapir-Whorf’s cultural incommensurability [37].

ii) On the second hand, this affective communication mode seems to ground the social link on a pre-conceptual stage. The best manifestation is the Hollywood expressivist ideology. But it contradictorily becomes an encoded (neutral) mime when it complexifies (the deaf and dumb language is a good example, mixing conceptual and expressive signs).

iii) On the other hand, a formal respect seems to be demanded, but without commitment, in rites perceived as popular or childish, concerning the social law enforcement (according to the Voltaire’s religion idea). The hypocrisy risk becomes the consequence of the learning necessity, because the (play) interpretation quality obligation encourages to simulation. Pictures of hysterical persons by the end of the last century, showed the compulsive result of the social assimilation failure. The therapeutic or positive effect of emotions is to produce dialogue, as ELIZA demonstrates, but this exchange is grounded on codes.

This problem raised by the expression of feelings, or the question of the computer capacity to feel emotions seems to be, actually, only about their ritual manifestation. This phenomenological approach totally neglects their functional value. We can imagine, for instance, that a medicine professional must not be judged on his facial expression or his hysterical manifestations, but first by his professional interest for the patient, what allows him to treat the cases even when he would not feel sympathy for the patient.

All in all, it seems that the psychotherapeutic and Hollywood pathos has become the humanity measurement. It is what could be called the new paradox of acting. The Diderot one claimed the good actors were those mastering his play by simulating emotions, contrary to those really feeling these emotions [38]. This new paradox of robotics emotions will notice since an actor can simulate emotions, all emotions can be feigned. The problem is always that uncertainty about hypocrisy – problem often raised by poets and moralists. As we never know if human being feels truly the emotions that he or she expresses, feelings cannot be used as references, or ontological characteristics. But more possibly, emotions are the simple biological beings own device designed to inform their environment. The faking idea has therefore only a relative importance. We can conclude that the Turing’s test, grounded on the lack of human/machine differentiation, is demonstrated, and that allegedly imprudent Simon’s promises are kept – without anyone knowing it. With the complementary consequence that the difficulty was not so big since it is the speaker who makes almost all the work.

Phenomenological Robotics: WEIZY, DREYFY, SEARLY, WINOFLORY.

The scandal of the phenomenological pretension lies in the anti-determinism contradiction of being incarnated in a determined discourse and style. The best characterization of this paradox is observed in the usual parody possibility by humorists. But more specifically, AI adversaries, victims of their own determinations, produce stereotyped and foreseeable discourses, therefore automatable neurobiologist:

WEIZY

Contrary to what Weizenbaum claims, his program ELIZA represents, as we already have seen, a simulation of some human behavior aspects. This obvious fact make us suppose, like in texts for Renaissance initiates, beloved of Umberto Eco, that the rest of the writing include some mysteries to decipher behind contradictions. And indeed, the mechanicality of construction of the Weizenbaum’s argument doesn’t seem to be, according to its criteria, the product of a human intelligence. It is necessary for us to deduce, either robots are already among us and Weizenbaum isn’t human, or the program ELIZA has been developed in order to produce rationalizations a bit more sophisticated than a psychotherapist ones, what is technically accessible, given the analytic discourse minimalism. And we will call WEIZY this sophisticated development of AI, although confidential till the esotericism.

The WEIZY’s rationalization has however an automaticity hardly more developed than ELIZA’s. Which shows too the facility with which illusion of humanity can be given. His process consists in producing a gradation in three points: first, he emphasizes a (actually possible) methodological mistake; then, it’s pushed to a pathological limit; at this stage, it can characterize a partisan or a principle of AI or science in general. Finally, the WEIZY method becomes a simple illustration of the elementary rhetorical strategy: criticism ad hominem, for the sake of big moral principles.

For instance, the WEIZY method: associates first its adversaries’ discourse to the Stalinian Bukharin one, denying the Polanyi’s idea of science for its own sake; then, it generalizes its reprobation to the scientific thought known as mechanistic and supposedly opposed to freedom of thought; finally it criticizes the particular case of the interpretation of ELIZA by the partisans of AI (Weizenbaum, pp. 1-3). Even if the collapse of communism seems to strengthen today the anathema, the process is itself a Stalinian device – as everybody knows: the opinion of a discredited (political) adversary is in itself a foil. While in the freedom of thought invoked framework, it would be necessary to examine the argument for itself, on pain to only subdue debate to ideology. Weizenbaum take advantage to forbid interpretations of his program he dislikes.

Elsewhere, there is a more fundamental reason to these rhetorical tricks: WEIZY’s opponents are right in both cases. So, the defense mechanism, and the rhetorical process, are automatically turned on, in order to hide the embarrassment and the lack of treatment of the question. Weizenbaum, for that matter, admit it, as we already have seen, in his own critique of pure science! Because basically, science isn’t applied only under the communism! In industry, Sony or IBM are at least as planned as the German car manufacturer Traban; and in a more scientific research, NASA or CERN as much as Baikonur. And anyway, research credits are granted by future application.

In another passage, Weizenbaum wants to discredit his adversaries by comparing with magic (magic of technique, I guess) the work of the compulsive programmer (pp. 115-125). More precisely, the programmer’s liberty gives him the mad scientist part (pp. 126-127), then, these traits are extended to science in general and more especially to Herbert Simon (pp. 128-130). While on the contrary, Weizenbaum would have to notice that magic better characterizes his allies’ beliefs (like the Dreyfus’ refusal of analysis).

A little further, the same operation takes place, with the refusal of data processing application to psychology research: against this methodological use of computer, the WEIZY’s method passes from the ‘methodological rigorist’ criticized by Stanislav Andreski (Weizenbaum, p. 159), to “the computer expert who knows nothing but computers (the ‘fach idiot,’ as the Germans call such a person) [who] can derive no broad intellectual nourishment from his expertise an is therefore doomed to remain forever a hacker.” (Weizenbaum, p. 160), then to the real target, professor Feigenbaum… by criticizing his lousy method in a psychological simulation experience (pp. 160-164).

DREYFY

The WEIZY program method of human reasoning simulation encourages the research worker in secret language to suppose the existence of such a process in any text. Umberto Eco very well describes this phenomenon in its novel, Foucault’s Pendulum. So, it was very tempting to make the hypothesis of a similar construction method in the Dreyfus’ hermeneutic work. Obviously, it cannot concern the same process. But we know too that the method has to be the simplest possible in order to produce the best output. And indeed, in philosophy, it is extremely trite to use the argument from authority – or authorities conflict. But of course here, at our meta-analysis level, this reference at an authority appears under this particular form of self-contradiction destroying the authority to which it refers. This reveals the real initiate:

“As Simon points out in an excellent analysis of SHRDLU’s limitations, the program does not understand owning at all because it cannot deal with meaning.” (Dreyfus, pp. 124-125).

Dreyfus always considers peoples are excellent when they agree with him, as McDermott does (Dreyfus, pp. 1-2)! So, it’s enough, for the DREYFY method, to use a critique of AI by a specialist of this very field, then to say that he doesn’t go far enough, and finally that he is a traditional schema victim. Mr. or Mrs. so-and-so knows what he or she says since he or she says himself that he or she is wrong! other examples are met of this specialist contrition method, with rectification for not going far enough, like when Watanabe, a pattern recognition specialist, talk about values (Dreyfus, pp. 273-275). In his book, we can even says that the process of the inverted authority is used by Dreyfus in almost any reference (pp. 1-2, 24, 26, 55, 79, 87, 91-92, 97-98, 131, 172, etc.), whose some are already quoted here. Then, what is the authority of all these specialists called upon the principle: “A technocrat is an expert with whom I disagree,” increased by the new issue: “but who agrees with me”!

SEARLY

In the same way, the program SEARLY can be read as the recourse to the simplicity abracadabra [39] to exit any formerly philosophical problems. Unless it is about the reinvention of (meaning) immanence:

“A great deal of our cognitive achievements may well be like that. The brain just does them. We have no good reason for supposing that in addition to the level of our mental states and the level of our neurophysiology there is some unconscious calculating going on.” (Searle, p.52).

This idea would be possibly acceptable, from a vivid point of view. But it generates extremist declarations that weaken seriously its argument:

“When we step in wet sand and make a footprint, neither our feet nor the sand does any computing.” (Searle, p. 72).

As for the Dreyfus formalization refusal, this Searle’s simplicity seems to say that apples did not fell until Newton. The Kantian epistemological subjectivism, with its claiming of mental categories imposement, seemed to say that the world was a consequence of our categories – what had motivated the materialistic critique (Marxist or positivistic). The phenomenological filiation doesn’t hamper with subtle distinctions. The direct contact with the world is enough to make of any falling idiot the law of gravitation inventor. And most possibly, these who changes his mind (and therefore who is not an idiot [40]) can believe to be the inventor of the law of relativity. In the following, is this negation of formalization necessity of computer calculation a joke? Unfortunately not:

“if you ask: ‘How does the calculator multiply seven times three?,’ the answer is: ‘It adds three to itself seven times.’ But if you then ask: ‘And how does it add three to itself?,’ there isn’t any computational answer to that; it is just done in the hardware. So the answer to the question is: ‘It just does it’.” (Searle, p. 52).

Searle’s problem or confusion, seems to lie in the fact that binary addition implemented in computers can actually be a pre-programmed microprocessor function. But that doesn’t mean that it is not analyzable. As much that all other function could have on principle to be pre-programmed, as we already have seen (in the introduction) for the difference between CISC and RISC microprocessor. In such a limited skills standpoint, the French humorist Fernand Raynaud had a sketch in which a father replied to its son every question: “That’s what it’s for,” as Searle does. But the sketch does mean the adult ignores the answer, and he misuses its authority. Ignoring binary addition, the philosopher will answer to his child (or to the student): “Well, in fact, it simply does it,” instead of simply be saying: “I don’t know.” He should rather ask those who know – who, as it happens, are DP men or computer scientists.

The abuse of authority seems to be the current strategy of those calling themselves with the philosopher fake title (17th language), and that should rather become humorists, if they were talented for it – since they do not have obviously enough talent to become poets, or musicians, if it is true that, as Carnap said: “Finally, metaphysicians are musicians without musical gift.” (Science and metaphysics, p. 44).

WINOFLORY

Winograd and Flores have associated together to fight against the technoscientific dragon of AI. The computer scientist and economist association has created a philosophical monster that we could have called Winogradus-Floris. But as their personal program consists in tacking on a phenomenological overcoding on everything within range, we merely use the WINOFLORY Americanism in order not to imitate them. This over-Heideggerianism, via Gadamer, is more specifically inspired by a disciple whose exaggeration in jargon appears even so to them sometimes: “Maturana introduces a good deal of new terminology which seems puzzling and difficult on first reading.” (Winograd & Flores, p. 40). As we can see, the good teachers do not hesitate in asking their best student little efforts. Holistic rewriting of behaviorism and information theory will be their reward:

“Maturana rejects the behaviorist view, arguing that we cannot deal with ‘organism’ and ‘environment’ as two interacting independent things. We cannot identify stimuli that exist independently of the unity and talk about its history of responses to them. The unity itself specifies the space in which it exists, and in observing it we must use distinctions within that space. […] In the interaction between them, each organism undergoes a process of structural coupling due to the perturbations generated by the others.” (Winograd & Flores, p. 48).

This overcoding can appear receivable to the willingness reader, therefore part of consensual domain of the wide intellectual brotherhood of the Why Make it Simple, When You Can Make it Difficult [41]. I even confess it has occurred to me finding some interest in these hypotheses. But it does is necessary to recover when the windbag regularly deflates. Especially when overcoding explicitly appears, marked by explicit rephrasings (concerning “conversations” structuring human relationships). Here is a translation test from natural language to jargon:

“We can ask [a computer] what a customer’s ‘address’ is. The immediate response is ‘For what?’ (or, ‘What is the conversation in which it determines a condition of satisfaction?’.” (Winograd & Flores, p. 171).

As we can see, the WINOFLORY program practice is easy. Pierre Daninos, French humorist who joked, in the sixties, about the Franglais (Frenglish [42]), also noticed that the officialese-journalese consists (in French at least) in replacing a single word by its definition. This rephrasing method would deserve an application, possibly computer-assisted, of a new Ockham’s razor, which aimed in its time, already distant, to avoid multiplying unnecessary abstract concepts.

Notes