Chapter 6
FORMALIZATION AND LANGUAGE

Language and formalization

Formalization of language is necessary, obviously, as a goal of artificial intelligence. It isn’t about to consider cognitive systems as languages, what would rather look like psychoanalysis declarations [NOTE 113]. It suffices to observe that language is a cognitive system actually acquired by the quasi totality of human beings. The mother tongue learning is a universal model, because anybody learns his own language. Exceptions are pathological cases (autistic persons, wild children, born deaf-mute persons before Helen Keller, etc.). Thus, this paradigm is absolutely representative. And the metalinguistic function of language, allowing it to talk about itself, fulfills a cognitive role comparable to consciousness. Linguistics (or former grammar) is itself the state of representations of language, consciousness of language learning.

The Model of Didactics of Foreign Languages

The analogy of language can be strengthened by the recourse to didactics of foreign languages. A second language learning can represent the assimilation of a linguistic system from another linguistic system. As much linguistics itself is not independent to comparativism since:

“The first [French] grammars were not composed in France for French, but out of France for foreigners.” (R. L. Wagner, La grammaire française [The French Grammar], p.2).

This analogy allows us to understand, or even to solve, the intercultural problems raised by the famous Sapir-Whorf’s hypothesis, claiming incommensurability of conceptual systems. It suffices to imagine some intermediate states (called interlingua/interlanguage), possibly grounded on intercultural primitive – of which linguistic concepts are the best examples. The linguistic problem is that any speaker can have partially different grammars, and that a good part is false (as precisely those the foreign beginners). Henri Frei, in his book, La grammaire des fautes [Grammar of Mistakes], precisely shows us the regularity of mistakes [NOTE 114]. They indicate an interlingua grounded on the system of rules already established.

More, each state of academic grammar, each state of linguistics, is itself an intermediate grammar. And every provisional grammar can distribute a norm, which can be different from the reality of one language. Unconscious agents of the circulation of these norms are naturally illiterate parents, foreigner learners (one another), or various rationalizations, among linguists too. In order to specify this gap, scientific or not, one can remind for instance that classic French grammarians used to impose to French the Latin grammar categories, Latin grammarians used to do the same with Greek grammar. The fact they could possibly modify any previous language doesn’t cancel its former reality, but does influence the posterior one with speaker’s conscious models. This makes inadmissible some intuitionist declarations: “I am suspicious of this claim just as I would be suspicious of a claim that said we should supplant our implicit theory of English grammar, the one we acquire by learning the language.” (Searle, p. 59).

Language learning, in an educated society and for an individual like Searle, supposes the passing by the practice of many grammar rules. It would not be serious to assert that an implicit theory, totally independent from the influence of this formal learning, still linger. This experimental condition is itself difficult to obtain in the scientific study a second language learning, even in the case of an illiterate person.

However, the refusal of formal can simply describe the observation of human behavior. It is indeed difficult (for beginners in a new language, or for badly educated persons) to produce a metalinguistic (formal) explanation. When a problem of communication raises, rectifications or their complements of information focus on the content, instead of focusing on the form. That seems to derive from an intuitionist  representation grounded on a negation of the necessity of linguistic medium, and from the possibility of direct communication. While it would be necessary to consider that the lack of understanding rests on the lack of reciprocal code assimilation.

The formal rule problem of AI is thus equivalent to what bother the didactics of foreign languages field, concerning the appropriateness and reciprocal advantages of formal and not formal learning techniques, i.e. grammatical (called structural) exercises and free conversations in the imitation of submersion in a language (called communicative approach). Now then, it is obvious, but also available in empirical inquiries, that this recourse to established grammatical structures is what allows learning structuration. Even if sometimes, actually, an empirically established rule conflicts with a learnt rule. The use of the experience of didactics of foreign languages is also explanatory of the false enigma of the exponential growth of complexity, which seems to punctuate the Dreyfus’ argument (pp. 91-92, 130, and 286-288). Indeed, it is perfectly normal, when working on a new area, to solve rapidly summary problems. How could it be otherwise, and how could anyone refuse to solve first what is easy! It is also normal then to come up against steps of problems. It is precisely the case of foreign language learning. Giving up in front of these difficulties is what is called fossilization then, what blocks the subject to a level where the progression becomes more difficult, and the difficulty apparently infinite. In these situations, the learner (illiterate especially) ends by slackening off at a given level.

Analysis or intuition

It can be also found, in didactics of foreign languages, or in didactics in general, the myth of the overall reading as a consequence of the holistic idea. This idea is most possibly grounded on authentic experiences or experiments. But it has also cause/induced the erring ways in the pedagogical current. To summarize the example of Chinese language in Searle, there is nothing like an overall vision of ideograms [NOTE 115], but an analysis of elementary part there are made of. Elsewhere, characters are classified, in Chinese, by counting the number of lines of simple characters, then by parts constituting a key for composed characters (Chinese holism is definitely analytic all right). In an ideographic language, primitive signs are simply more numerous than alphabetical letters. But they are not made of a different iconic nature, from a pattern recognition point of view [NOTE 116]. With practice, those used to be seen are overally  recognized, i.e. those which are already analyzed. Indeed, a sign can be learnt without having been analyzed first. But the primitives allowing its construction can be acknowledged later (exactly like etymologies in western languages). Elsewhere, it is necessary to notice that ideograms (or etymologies), preserve former classifications, and therefore cannot be used then to clarify the meaning.

Of course, some formulae are global. But it is by analysis too that one progresses in language. In situation of foreign language learning, a new word or a new sound will have to be analyzed to facilitate understanding, or its reproduction. This analysis is necessary too, because an overall perception of different languages, for subjects, isn’t equivalent to the same segmentation in other languages, according to the legitimate observational aspect of the Sapir-Whorf’s principle. For instance, almost all English vowels are diphthongs. For French, the elementary equivalent doesn’t exist. Will it be necessary to conclude that French people are driven to analysis and Anglo-Saxons to holism? Anyway, (what is called) speaking in tongues is a myth (of Anglo-Saxon religious sects!), because in practice, one learns patiently by analyzing articulatory rules for sounds, morphological or phonological composition for words, grammatical rules for sentences, uses in real life for linguistic pragmatics, etc.

The above fossilization is precisely a consequence of a breaking down incapacity, especially characterizing the low-educated learner. We can recognize here the gaps produced by the absence of abstract manipulation (according to Piaget’s teaching). What is determining, here, is precisely the influence of established overall forms of mother tongue, considered as the only possible sifting, as the only possible language. Elsewhere, this point is found too, under another form, in some aspects of purism, for the best educated.

Finally, foreign languages learning just has to be observed all right, even if this competence remains imperfect. Elsewhere, this imperfection essentially lies, besides mental versatility lack, in the fact that the same ephemeral experiences of the news, or daily items, are not shared abroad. This difference precisely blurs when emigrating, because once the basis established, it’s enough to ask an explanation if needed. Beyond, it is about technical and specialized areas which are ignored by non-experts. But they are precisely already universal or encoded. Then, it would be absurd to ask more than what is the natives’ fate.

Perception and conceptualization

Artificial intelligence at least has the merit to remind that formalization of language remains the goal of logic (or the ‘art of thinking,’ as 16th century Port-Royal philosophers used to said). But this project has to be grounded on a weak idea of science avoiding dangers of logicism, and those of common sense. The first one is incomplete concerning ordinary language, and the second one obviously cannot be idealized without sinking into demagoguery. The project of formalization of natural language allows solving the false opposition signaled by genetic psychology:

“An English Rusellian said some day, in order to prove the uselessness of research about children psychology, that a logician care about true ideas while a psychologist take pleasure in describing false ones.” (Piaget, La psychologie de l’intelligence, p. 28).

AI adversaries refuse this formalization possibility, which Dreyfus even considers as what he calls a metaphysical assumption. By this term, he satirizes on partisans of AI, whose clumsiness can possibly be explained by their nonspecialists of philosophy situation, or more generally by the lack of interdisciplinarity:

“These last metaphysicians are staking everything on man’s ability to formalize his behavior; to bypass brain and body, and arrive, all the more surely, at the essence of rationality.” (Dreyfus, p. 78).

The excess of formalism of modern science can be admitted, as it often finds platonic tones by distancing from the world as perceived by senses. But usually, the metaphysical assumption is rather about claiming that human being or intelligence are beyond physics. We already have seen that this dualism is sometimes repeated by phenomenology, which split human world from material world. In Dreyfus’ continuation, the Winograd and Flores duet specifies this anti-theoretical thought by developing the idea of commitment and social interaction, against the rationalist tradition of knowledge. They also see in the lack of commitment the cause of the computers impotence in understanding natural language (Winograd & Flores, p. 12). They reinterpret computers themselves, in their jargonizing rephrasing schema, as having no existence of their own:

“Computers do not exist, in the sense of things possessing objective features and functions, outside of language. They are created in the conversations human beings engage in when they cope with and anticipate breakdown. […] Computers are not only designed in language but are themselves equipment for language.” (Winograd & Flores, pp. 78-79).

With a lot of indulgence, and following this private language, this characterization, on the contrary, could be generalized to language itself. Language is a tool of interaction to express relationships of (human) beings with the world. But this is not sufficient, because for these authors, “we create our world through language“ (Winograd & Flores, p. 11). One can think here of a kind of verbalist magic. The Searle’s influence is obviously confusing acts of language with petitio principii. Why refusing formalization when claiming that “’language as action’ and ‘language as interpretation’ can be reconciled” (Winograd & Flores, p. 54)? Because it’s enough to admit language as representation too.

Finally, the human being’s ability to find abstract principles is not a metaphysician’s assumption, but a fact. Even if reason is grounded on body, it proceeds by abstraction of principles, from observable human being’s behavior. And a robot would find abstract principles too about its own behavior by considering its interaction with its environment. A program would proceed by simulation of this interaction.

Every conceptualization is a simplification, i.e. an elimination of irrelevant details. The reductio ad absurdum is available in a Borges’ short story, “Of Exactitude in Science,” in A Universal History of Infamy, in which a king ask his geographers a map more and more precise, until it confuses with the real territory. Therefore, the romantic cliché is surprising from a scientist quoting Horkheimer, who regrets that “concepts have been reduced to summaries of the characteristics that several specimens have in common” (Weizenbaum, p. 249). As much the recognition of this elementary process of any abstraction is acknowledged by the philosopher. However, his irony places him in a dualist framework, which doesn’t imagine that the human world belongs to the same procedures than natural one:

“As Galileo discovered that one could find a pure formalism for describing physical motion by ignoring secondary qualities and teleological considerations, so, one might suppose, a Galileo of human behavior might succeed in reducing all semantic considerations (appeal to meanings) to the techniques of syntactic (formal) manipulations.” (Dreyfus, pp. 68-69, my italics).

In accordance with the traditional philosophical pretension, it can be admitted that, sometimes, Dreyfus states very well the problem by abandoning this time the richness of the cultural and material context. But we already know that this kind of rhetorical process confuses a question with a negation of the adverse standpoint. In general, elsewhere, the answer isn’t even listened.

Conversely, the immanence of meaning, most possibly intuitionist, goes until a kind of semantics quietism, in the tradition of simplicity. This cognitive ontology is here generalized to the nonhuman world, phenomenological contradiction equivalent to some mechanistic excess: “It should seem no more mysterious, in principle, that this hunk of matter, […] should be conscious than it seems mysterious that this other hunk of matter, […] should be alive.” (Searle, p. 23). No one else than a philosopher uses this rhetorical strategy of the absence of astonishment, to use at once the opposite strategy: “Our second problem was, how can atoms in the void have intentionality?” (Searle, p. 24). Thus, this parapsychological syndrome more or less animistic thinks to be scientific by associating to the academic corpus. It is the tradition of parasciences soliciting scientificity.

Reductionism and Constructivism

The process is getting to be known. The prejudice against formalization disputes the passage from perception to representation by a neuronal reductionism. The refusal of formalization by the phenomenological/Gestaltist approach even generalizes perception to cognition. That happens first by the refusal of breaking down or calculus, to which “time factors and field interactions” are preferred (Dreyfus, p. 162), oddly enough perceived as incalculable. The guarantee of Gestalt theory claims a distinction of the human and computer recognition methods (Dreyfus, p. 123).

That human and program information processing are different is acceptable, but that doesn’t exclude a classification. Dreyfus rapidly switch from the fact this classification could be unconscious, according to Selfridge and Neisser (p. 122), to an ontological difference, most possibly on Searlian subjectivists criteria. But how any information collection can be exclude in taking into account of a dog crossing before a car (Winograd & Flores, p. 146). To say that: “We do not act as a result of consideration, but as a way of being” (W. & F., idem) doesn’t explain this way of being. The research of a rational decision has precisely for purpose the possibility of correction of an inadequate or not optimal way of being.

More, all the question is to know if a calculation cannot model a perception, or if the Gestaltist ideas of overall perception doesn’t concern forms already learned, because it took long, during childhood, to learn this usual forms. In the same way, despite the resistance to perception as calculation, the current cognitive psychology allows to assimilate perception to a construction from an interpretation and a composition of signals and elementary schemes. This explicitly contradict the dogmatic claims against the construction process (Dreyfus, pp. 269-270).

This perception process only elementarily concerns cultural patterns acquisition, because the existence of more abstract categories is, indisputably, revealed by intercultural confrontations, displacement, or differences of knowledge. Elsewhere, the acquired knowledge does not exclude analysis possibilities, contrary to a supposed dehumanization. Simply, a specialist is able to put in word an impression a nonprofessional only feel. And this verbalization is an analysis. More generally, it can simply be reminded that language is an artifact, what means it is rather necessary to break down it to understand it, and not to absorb the voice of God.

However, a resistance to an excessive constructivism, which sometimes reigns in human sciences, can be understood. The Kantian notion of cognitive frameworks, as intermediaries between us and a real inaccessible as such, often ends into an extremist identitarian relativism to justify intercultural conflicts of representations or values. If this principle is pushed a little, to say that individual builds his perceptions can drive to say there is no perception at all, or according to the antique principle, that looking is active (i. e. working like a projector and not like a camera). Then we can conclude that the world doesn’t exist, that everything is an illusion (on the immaterialist mode inspired by Berkeley or oriental philosophies), or we get an avatar of voluntarism (always by retraining of many formerly Marxist research worker) [NOTE 117].

As phenomenology alternately wavers between negation and apology of constructivism, one can hardly decide anything is on this point. We can wonder if constructivism simply wouldn’t about abstraction and classifications, in the positivistic tradition, which is disputed by AI adversaries, because philosophy constantly rediscovers what it first denies. This is precisely the definition of reductio ad absurdum!

More seriously, it can be classically admitted a weak version of constructivism, as the subject’s cognitive activity means that order is what arises, not what is seen. Elsewhere, the ancient peoples who saw the same things only perceived chaos where we admit a physical law regularity. Conceptual frames in question constitute then the cognitive knowledge, as its purpose and its cumulative means.

We can also admit that assumptions analysis, corresponding to this constructivism, allows the inductive research of principles generating discourses or actions. Then a disadvantage often lies in the tendency to the construction of a simplistic, closed and deductive, system. This approach represents from this point of view the traditional philosophy (or theology) summary considering real as deduced from a limited number of principles, or semantic primitives in the AI case. This traditional influence, as Dreyfus said, actually induces an internalist or analytic idea, associating holism (phenomena are only characteristics of the whole) and logicism (only defining the means of this deduction) whose structuralism is the outcome. Popper, by continuing to refuse induction, situates too in an internalist framework. But he judiciously added a control of conjectures by experiment, and a research of systematic refutation [NOTE 118].

Conversely, phenomenological or empiricist standpoint are coherent with the idea of an opening to the world. This opening is the only meaning of the incompleteness of our theories. We can understand thus the idea of a practice previous theory. However, an opening to the world is relative, because it happens to be an epistemological closure in term of knowledge stages. For instance, a classification of animals can be disturbed by a new discovery. But by definition, the knowledge heap must make this upheaval less probable, contrary to the (leftist) epistemological myth of permanent revolution. Scientific revolutions are not always so radical it’s made out to, as Piaget reminds:

“Any exact science, despite all the so-called crises and all the reshaping it is proud of, in order to show its vitality, still is a body of concepts maintaining the detail of its relation, and even tiding it up, when adding a new fact or principle. […] We can find the same phenomenon […] in every balanced man’s thought.” (Piaget, La psychologie de l’intelligence, p. 47).

Let’s notice that epistemological closure (and not realist one) is as necessary for action. An always-limited knowledge is elsewhere what contradicts holism by allowing action without absolute knowledge. The economic analysis precisely considers this bounded rationality with Herbert Simon. His model, such ideal it can be, doesn’t claim (by definition) a total knowledge, what makes incomprehensible Winograd & Flores’ reservations or pseudo-methodological questionings which would demand “for what observer this space of alternatives exists” (Winograd & Flores, p. 146), because the fact that some alternatives are not conscious (the case of observers could formalize a representation of action of others) precisely indicate bounded rationality. This questioning rests on the holistic prejudice. In the real world, a partial knowledge is always good enough, because it allows action, and gives a competitive advantage.

Indeed, for some other economists, action is the product of belief or ideology, and not of rational reasoning. But it concerns a judgement obviously external – rational – about beliefs. In general, speaking about beliefs is already grounded on a more exact knowledge or more formalized, possessed by these economists. More, the ideological rationalization itself is equivalent to a reasoning, only with false or limited premises, often over-generalized. The protagonist’s micro-economical interest is obviously distinct of macro-economical balances [NOTE 119].

Consciousness of formalisms

The refusal of formalism and rules is elsewhere finally contradictory with the necessity of consciousness into action. The phenomenology discourse insists on this intentional criterion, which implies the protagonist’s consciousness. This standpoint is most possibly originally explained by the concern of restoring some autonomy to the subject after the deterministic domination in human sciences. However, the intentionalist criterion challenges unconscious phenomena because they are not conscious (Dreyfus, pp. 143-144). This kind of petitio principii is the phenomenologists’ favorite pseudo-analytic mode of argument.

There are obviously unconscious processes, and some are more or less clarifiable, like cultural frameworks. The best example for our topic is grammar and its rules. This notion of rule can indifferently express unconscious or aware factors to explain behaviors. It is not contradictory with the fact that partisans of AI admit this lack of consciousness, following the classic philosophical tradition from Plato to Kant: “Feigenbaum conscientiously notes that the experts themselves are not aware of using rules” (Dreyfus, p. 28), because in order to realize expert systems, the knowledge engineer’s work, on the Socrates’ model to whom he borrows maieutics, precisely consisting in discovering rules applied by experts. The fact that expert systems actually work is a sufficient proof. It is necessary to notice too that all the second part of the Dreyfus’ book uses the method that refers to the (Kantian) bringing to light or the (Marxist) critique of implicit assumptions, corresponding to the idea of rules, generating reasoning from these assumptions.

One can elsewhere be surprise by the simultaneous refusal of the using of conscious rules in both practical examples provided by both Dreyfus and Searle:

“This bicycle example is taken from Michael Polanyi’s Personal Knowledge (p. 49): ‘From my interrogations of physicists, engineers, and bicycle manufacturers, I have come to the conclusion that the principle by which the cyclist keeps his balance is not generally known. The rule observed by the cyclist is this. When he starts falling to the right he turns the handlebars to the right, so that the course of the bicycle is deflected along a curve towards the right. This results in a centrifugal force pushing the cyclist to the left […]. But does this tell us exactly how to ride a bicycle? No. You obviously cannot adjust the curvature of your bicycle’s path in proportion to the ratio of your unbalance over the square of your speed’.” (Dreyfus, p. 330, note 2).

“When my children went to the Oakland Driving School, they were taught a rule for parking cars. The rule was: Manoeuvre your car toward the kerb with the steering wheel in the extreme right position until your front wheels are even with the rear wheels of the car in front of you. Then, turn the steering wheel all the way to the extreme left position. […] I was interested to learn this rule because it is not a rule that I follow. In fact, I don’t follow a rule at all when I park a car. I just look at the kerb and try to get as close to the kerb as I can without bashing into the cars in front of and behind me. But notice, it might turn out that my behaviour viewed from outside, viewed externally, is identical with the behaviour of the person who is following the rule. Still, it would not be true to say of me that I was following the rule.” (Searle, p. 47).

The question, on the contrary, is precisely the translation of “to the ratio of your unbalance over the square of your speed” in perceptive terms, in AI adversary’ fringe consciousness (looking a lot like a homunculus, and which would at least deserve the quotation marks they generally overuse). In the first example, the initial physicist’s explanations looked good enough and did not present an insurmountable difficulty for common sense. The (pedantic and ironic) use of mathematics terms is only explained to quantify the state of balance. More, it seems to be forgotten here that the refusal of science amounts obviously to deny it as a form of consciousness.

If, in the first example, it seems only to be a matter of common sense explanations, in the second one, it seems to be about manner of speaking, as “I don’t follow a rule at all when I park a car. I just look at the kerb and try to get as close to the kerb as I can without bashing into the cars in front of and behind me” precisely clarifies the rule used (possibly insufficiently)! The problem of the mentalist hypothesis seems to lie in the difference between rule, and formal learning in the one hand and nonformal, conscious or not in the other hand.

A subject can very well have, like in this case, a provisional more or less exact and more or less incomplete internal rule. It could elsewhere exist several methods to do so. But that doesn’t exclude that a rule can be more precise or more encompassing than another. Like in such a case, a rule is a formalization of behavior without reference to the subject’s (mentalist or intentionalist) rationalizations. This is what makes us assume: “it might turn out that my behavior viewed from outside, viewed externally.” Because, contrary to what Searle thinks, this observation criterion is a sufficient condition (an observed behavior is a phenomenon too). Indeed, observable variations (success/failure) of this parking behavior can characterize too the times when an incomplete formalization has failed, because it was not optimal. No doubt that other types of rationalizations will go with mentalists models then: culpability: ‘I ought to pay more attention’; accusation: ‘It is your fault, you have amused me’ (Searle to his wife or children); bad luck: ‘Blast, missed again,’ etc. Unsuccessful behaviors often imply an intense psychologizing mentalist activity.

The Polanyi’s example, like the Searle’s one, possess also the disadvantage to concern a usual competence, which would be simulated, in the linguistic case, by the learning of one’s own mother tongue. The apparently trivial result looks like the lesson of the master of grammar, in the Molière’s play, Le bourgeois gentilhomme [The bourgeois gentleman, whose humor is grounded on the sycophancy of aristocrats. According to the trivialization principle, what is already known appears ridiculously easy, even innate. But this impression would have been different if it had concerned the learning of a foreign language, or a new competence. The pedagogy of the master of grammar would have been perfectly judicious, as he made articulate the sounds like in any phonetics lesson.

In the bicycle or driving-school examples, bringing rules to light allows pedagogy, obviously intended to those who do not (yet) realize a performance, and replaces the only selection of those who succeed more or less at the first attempt. And reciprocally, we can understand it is not necessary to bother those who succeed. In this kind of competence, we can imagine that the cognitivist (AI) model get rid of the judgement of value which would stem from a failure, most possibly judged as a moral failure of this famous fringe consciousness [NOTE 120]. This failure can be a failure to explain of the non-formalistic educator, because even identification of handicaps is precisely equivalent to the possibility of existence of these cognitive processes.

More generally, we can rather observe a lack of understanding of the principle of formalization into rules, because it is possible to assert that formalization is equivalent to understanding, as Dreyfus (pp. 202-203) exactly put it out (but contradictorily). And this incompetence appears in language itself, or in pseudo-philosophical distinctions:

“Polanyi calls attention to the difference between strict rules and maxims: Maxims are rules, the correct application of which is part of the art which they govern. […] Maxims cannot be understood, still less applied by anyone not already possessing a good practical knowledge of the art.” (Dreyfus, p. 32).

In short, it is about rules which are not rules, but precepts; which aren’t applied like a technique, but an art; which aren’t learned, but understood, etc. This is an illustration of the thought of castes. And of course, the backward-looking discourse of philosophy trots us out the Rousseauist trick of the non-westerners cultures idealization:

“The Japanese traditionally prefer to settle disputes, even those which relief at law is statutorily available, by what Westerners would see as informal means. […] The United State Supreme Court […] in the case of Diaz v. Gonzales, which was originally tried in Puerto Rico, refused to set aside the judgement of the court of original jurisdiction…” (Weizenbaum, pp. 225-226).

This double self-contradiction is exemplary: on the one hand, the American judgement is judicial, as the local one is; on the other hand, for a non-American person, all that rather reveals the traditionalist and community (informal or jurisprudential) characteristic of American justice! If someone is absolutely keen on distinguishing degrees in formalisms, he could say that law is less formal than mathematics, European law more formal than American law, American law more formal than local law, or writing more formal than oral [NOTE 121]. But this schema is obviously false, because inside each system, rules are applied mechanically, and counter-arguments as mechanically implemented (by more or less good lawyers). If written or oral law can precisely amount to a social relationship automation, this automation doesn’t exclude precisely, for our topic, creativity. It simply applies through principles newly characterizing or newly categorizing these situations. It doesn’t exist informal formalization.

Obviously, the negation of abstraction still has to be compared to the epistemological knowledge. Thus, we can observe, in the quotation below, that both first points of the Bachelardian analysis do span holistic, vitalist and simplist biases of philosophy:

“In its individual training, a scientific mind would necessarily pass by the three next stages, far more precise and particular than Comtian forms. The practical state in which the mind plays with the first images of a phenomenon and grounds on a philosophical literature glorifying Nature, oddly enough singing both unity of nature and its rich diversity. The practical/abstract state in which the mind adds to physical experience geometrical diagrams and rests on a philosophy of the simplicity. Mind still is in a paradoxical situation: it is all the more assured of its abstraction that this abstraction is more clearly represented by a sensitive intuition. The abstract state in which the mind undertakes information voluntarily freed from the intuition of the real space, voluntarily detached from the immediate experience and even controversial with the first reality, always impure, always shapeless.” (Gaston Bachelard, La formation de l’esprit scientifique [The Making of the Scientific Mind], p. 8).

But the third Bachelardian topic of the epistemological break has traditionally chosen anti-empiricism (following the Kantian tradition). It would rather be necessary to validate (as Piaget does) the already present formal activity to each stage, because we can observe that epistemological formalism, speaking about errors of the past, tends to a kind of a posteriori dogmatic pedantry. These preferences could also be only a question of style:

“An empiricist person has a tendency to project on laws and theories the being a fact at their origin. The only truly solid phrases are those that state the most directly facts of experience. […] A rationalist person, on the contrary, projects on laws and facts the intelligibility and the necessity which confers their insertion in a theoretical system.” (Robert Blanché, L’induction scientifique et les lois naturelles [Scientific Induction and Natural Laws], p. 161).

Such an observation allows qualifying normative trends of French epistemology, because this more or less literary epistemology of human sciences is under the dictatorship of the constructivist norm, whose best wording is very well known:

“For a scientific mind, every knowledge is an answer to a question. Without a question, there isn’t scientific knowledge. Nothing goes without saying. Nothing is given, everything is built.” (Gaston Bachelard, La formation de l’esprit scientifique [The Making of the Scientific Mind], p. 14).

This epistemological leitmotiv raises the problem to know if there was no scientific knowledge before its appearing. In the opposite case, even if that means any scientific knowledge can be expressed under the form of an answer, therefore it would only be a question of presentation. It would be enough to adopt a problematical tone rather than an affirmative one. This stylistic norm certainly explains the questioning fashion, with the generalization of expressions (in French), even verbal mannerisms (or Englishism here) like: “that questions me,” “PUF books question the world” (advertising slogan), etc. [NOTE 122]. This invading questioning fashion or taste can produce this kind of scholar overcoding: “a problem is what questions.” Elsewhere it is fun to observe that the distribution of this norm is also most possibly the origin of a televised game – oddly enough called Jeopardy –, consisting in finding the question to an already known answer! And the value of this questioning norm has certainly no more interest, as for the description of the scientific practice [NOTE 123].

The confusion between question and problem can origin from the academic mathematical usage, in which problems are exercises. Of course, sociolinguistics oblige, there is no way to refuse a new meaning, especially being grounded on a kind of etymology. But it is allowed to prefer, by reference for the common use, the term question for what has an answer (or a human agent), and to preserve the term problem for the case of which one doesn’t know the solution.

The confusion could have come from a professorial bias, because in the academic framework, a pupil has to discover a solution he did not know, while, for the teacher, it was only a question. To be known in advance for an answer is the minimum of lesson preparation for a teacher: it’s the only way to guarantee an objective evaluation – not relative to the class’ level. This relativism can only be used to check that a question corresponds to the level. Even when it concerns a common research (professor/students, groups of students), the attempting to propose (to demand) an answer while pretending not to know it, while knowing it, is an imposture reducing unknown to the known, to the prejudice. This is a frequent didactic bias in front of a not expected answer, during a fake research.

It could be asserted then that philosophy, by not making this distinction, simply consists in transforming a question whose answer is known into a problem, even into an enigma. And in transforming too a true problem into a question, or into an overcoded old chestnut, even into a delirium about unknown. In this case, knowledge is allegedly determined by unknown, of which, obviously, a preferential speaker (poet, philosopher…) will be the spokesperson, from which the retraining of avant-gardes when they prides themselves on transcendence. This phenomenon came to a sociologist, despite a guilty indulgence:

“But it would be naive to reduce the dissertation [French essay of philosophy] to the simple production of an already learnt answer. When the teaching profession claims the ideology of ‘personal thought,’ it can easily put forward the place this exercise grant to the individual initiative. Indeed, subjects of dissertation are expressly design to avoid the mechanical use of textbook question. The wording aims to scramble correspondences between the peculiar processing and the philosophical question, ideas and doctrines repertory […]. The dissertation subject is this cryptic question which is only known as having, certainly, at least one listed meaning. The ‘personal’ characteristic of philosophical ‘thought’ demands a singular and inescapable question, which could not being encoded like history subjects (‘Japan in the Meiji era’) or natural sciences ones (‘structure of the cell’). The ambiguity is the mark of an all the more legitimate intellectually field that it holds a kind of institutional monopoly of abysmal questions.” (Louis Pinto, “L’école des philosophes” [“The School of Philosophers…”], Actes de la recherche en sciences sociales [Research Proceedings in Social Sciences], n° 47/48, p. 25).

But the fact of the always present pre-grading, with the consumption of abstracts and others, actually reduced hermeneutics to textbook question. Philosophy can be used to produce dissertations so much that some computerist can have a little fun in programming something like a philosophical text generator (“Philosophie,” Soft and Micro, n° 5, February 1985).

Human Formalization and Clarifying

Arguing against formalism, the phenomenological reductionism (Dreyfus, p. 269) is a negation of the authentic human reality (and simplicity) of these classifying formalisms. As we have already seen, encyclopedias, categories, labels, etc. do exist to organize the human existence and action. Elsewhere, such a categorizations is made by Dreyfus himself, when he distinguishes four types of intelligent activities as limitations to traditional computerization (Dreyfus, p. 292). It is obvious that a distinction of this kind can through light on some problems. And it precisely represents a knowledge to integrate into a problem solving strategy.

The human action is actually organized according to implicit or explicit segmentations or conceptualizations. All distinct behavior is a formalization too (as segmentation of the world by action). This breaking down, even apparently nonformal, is equivalent to an algorithm, or to the heuristic, when there is too many options.

The interest of a breaking down is precisely to allow a discursive activity. Besides, it is the explanatory principle – of reduction to elementary principles – which characterizes any scientific discourse. Analysis refusal is simply an effect of incompetence. Even in symbolism, a myth has to possess an explanatory value. And a mythological system would on principle have to be coherent. The great religions can also be considered as coordination efforts of fragmentary explanation. Finally, a symbolist doesn’t have be able, by rights, to say no matter what, even if it does, actually.

Such a breaking down is conversely a characteristic of empiricism/nominalism, considering all concepts as a structured totality of elementary characteristics. And many works of neo-idealistic philosophers simply consist in trying to mask this basic banality in their notion of semantics (as reminded by U. Eco). This breaking down ascends to antic Greece, and well sums up all the western thought, as a specific cultural experience – which is worth any other else by rights, but is actually much better [NOTE 124].

Syntactic calculation

The classic question of philosophy, “Whether or not every aspect of human thought is reducible to a logical formalism?” (Weizenbaum, p. 12), takes an immediate pragmatic form with data processing and AI. Weizenbaum is shocked by this possibility. But the other AI adversaries’ philosophical competence of historical reference management allows them to remind some steps of this problematics of calculability, at least the explicit steps, because philosophy in its whole does proceed from this reality:

“’When a man reasons, he does nothing else but conceive a sum total from addition of parcels,’ he wrote, ‘for REASON… is nothing but reckoning…’ [NOTE 125].” (Dreyfus, p. 69).

“Like Hobbes, Boole supposed that reasoning was calculating, and he set out to ‘investigate the fundamental laws of those operations of the mind by which reasoning is performed, to give expression to them in the symbolic language of a Calculus’. [NOTE 126].” (Dreyfus, p. 70).

To deny that thought is a reckoning could possibly to be comprehensible, although disputable on the principle of a simultaneous acceptance of common sense. But reasoning (if it is distinct of thought) is a reckoning all right, which is precisely characterized by the definition given by Hobbes and Boole, and by the brilliant Leibniz’s wording, reminded by the same erudition:

“’If someone would doubt my results,’ Leibniz said, ‘I would say to him: “Let’s calculate, Sir, and thus by taking pen and ink, we should settle the question’’ [NOTE 127]‘.” (Dreyfus, p. 69).

That reasoning is a reckoning actually belongs to the ability of explanation, to others, or to himself. There is a formal identity between problems of formal or reckoning (or cybernetic, mechanical, AI, etc.) knowledge and those of human or common (or phenomenological, literary, etc.) knowledge. And this identity can reduce the second to the first, which precisely represents a formal representation of it. From this point of view, the natural language isn’t less formal than a logical or mathematical language [NOTE 128], which simply represents a non-ambiguous, but incomplete equivalent.

The dehumanization argument is only rhetorical, because the philosophical tradition demand a perfect deduction as ideal (the judicial practice too). It is, in the literary canonical example, what opposes Dr. Watson to Sherlock Holmes. In the same way in science, the purpose of any reasoning is a perfectly reproducible performing. But obviously, it is possible to consider that these moments of automaticity do not characterize the fact that “humans, if they are machines at all, are vastly general-purpose machines” (Weizenbaum, p. 183). Then the contradiction is double: concerning computers, because universality is precisely the novelty of computing. Turing’s universal machines are merely opposable to machines dedicated to a particular task (like an elevator program). The computer, as universal machine, is precisely the model of universality of mind, of its versatility. The fact to load various programs and to be able to modify them is equivalent to the fact that a human being can play tennis, drive a car, be calculating, etc. In both cases, this means these skills can be distinguished, and be independently acquired, without conditioning the quality of human being. The only reservation to universality rather concerns human being: each individual is often specialized, in areas he can pride itself to exceed the computer (and other individuals, elsewhere, like grand masters in chess!). While each individual computer is potentially immediately universal (subject to load the good programs, or to combine several ones simultaneously).

This possibility of intellectual performance automation can again be demonstrated in the previously reminded situation (cf. here, Holistic Delirium). It isn’t about satirizing about the Dreyfus’ prediction [NOTE 129], but about wondering about his ability of understanding of the class of a problem. The computer’s way to solve this problem of analogy, consisting in finding the next figure of IQ tests, is precisely the way human beings solve these kind of problems. This is quite normal since these tests have been designed for this type of resolution.

In order to deny calculability, mentalism postulates a holographic idea of semantics. It would be necessary then that signified (meaning) hides in the smallest possible fragments of signifiers (signs). We can recognize here the symbolist, or Freudian, ambition. This demand is even about quantitative elements:

“In adding 6 and 3, [the calculator] doesn’t know that the numeral ‘6’ stands for the number six, and that the numeral ‘3’ stands for the number three, and that the plus sign stands for the operation of addition. And that’s for the very simple reason that it doesn’t know anything.” (Searle, p. 48).

Nevertheless, this is a fact that the machine knows how to – has the competence to – do calculations. When a human being adds two numbers, he applies syntactic rules too. It is even a good example of a situation in which one can apply rules without understanding them. In these operations, only mistakes can show that one hasn’t understood something (or was distracted). Leibniz again, in order to show us how we apply syntactic operations, stated that, in the case of operations upon big numbers, we precisely apply rules without picturing a realistic representation of them. He thus proves the idea of a “blind thought” (cf. Eco, La quête de la langue parfaite [The Perfect Language Quest], pp. 319-320), precisely guided by formalisms, justifying his philosophical language project.

Actually, thought perceived as a black box, like for operations tables, is simply constituted by the acquisition of automatisms. These automatisms can be analyzed to be transposed in computers (by one means or another, since binary calculation is different from decimal calculation).

Finally, if we get rid of the philosophical all‑or‑nothing assumption framework, a minimal standpoint of a calculating thought, by principle of a weak science, can first consist in calculating what is calculable. This more realistic approach really means a possibility of breaking down in steps and shows a cumulative possibility. We can recognize here the AI partisans’ pragmatism, against the trying to be modest academic norm. Weizenbaum elsewhere shows a strange indulgence about generative capacities of a universal grammar prefiguring the general project of AI, in contrast with his contempt for the Simon and Newell’s claimed ambitions about the AI competence “is going to increase rapidly until – in the visible future – the range of problems they can handle will be coextensive with the range to which the human mind has been applied.” (Weizenbaum, p. 138). Contrary to his irony, this correspondence is actually already here for algorithmic operations, which are human mind’s aptitudes. And if machines’ aptitudes can be developed, it is also because they aren’t perceived as absolute – although there is no reason to suppose they are limited by the human mind’s ones. Indeed, common sense knowledge also include, today, simple arithmetical knowledge, at least. They aren’t either innate knowledge, especially due to the late adoption of the decimal system or of the Arabic numeration, only allowing the use of these very tables.

That reminds us too the calculator history. It make us understand that calculation in the binary system, or the machine possibilities of calculation, aren’t any more artificial, or mechanics, than a suddenly humanized decimal calculation! In front of the Pascal’s calculators or Leibniz’s one, it could be fruitful to remind the feelings of the time to estimate all the way done:

“That one could give such an ability to a machine, in the sense we mean it, seemed a little supernatural. So wrote the Pascal’s sister: […] ‘the mind look like as if it had been captured by the machine’ because ‘it was possible to do all kind of calculations without any mistake, such extraordinary without pencils, but, even more, without even knowing arithmetic’. This Pascal’s sister thought is possibly one of the first about the problem being called artificial intelligence today.” (René Moreau, Ainsi naquit l’informatique [The Computer Comes of Age], p. 13).

That seems a good enough answer to the questioning about the intelligence of a pocket calculator, a clock, an accountancy or word processing program (Winograd & Flores, pp. 11, 93, 176). We precisely can see that the arithmetical ability of the machine is grounded on the automation of calculation rules, while it was formerly believed it was only a human mind competence. It is the only we can relativize the idea of intelligence at each step instead of reserving it only to Giovanni Pico della Mirandola or to polytechnic skills, while elsewhere despising interdisciplinarity. Can we see phenomenological illusions as reminiscences of anti-novelty devilizing? As we have forgotten too those concerning the same calculators:

“ […] And so the inquisition sued the Leibniz’s mother as for having copulated with the devil because, otherwise, how would have she got a son interested in such realizations.” (René Moreau, Ainsi naquit l’informatique [The Computer Comes of Age], p. 13).

Calculating Relativism

The AI opponent’s critics can validly concern the mere observation that few peoples do reason correctly, or even that few peoples use reasoning before acting. The purpose of practical intelligence rather seems to be an acquisition of automatisms. The simplest calculating breaking down is often challenged in common life, in which reckonings  even do not use academic knowledge. It is necessary to say they are sometimes forgotten, as for reading without daily practice.

A perfect example of non-application of rules, nevertheless learned or available, is notoriously that mathematical knowledge. This oblivion of mathematical being elsewhere allows to easily welcoming the reckoning depreciation, as origin of anti-IA demagoguery. On this principle of competence loss, I was personally told about a contesting of work hours sum, grounded on the relative difficulty, the non-obviousness, of sixty-basis calculation. It was about a total of 3 h  + 3 h 30 + 3 h 45 which doesn’t yield to 10 h 20: (3,45 + 3,30 + 3,45), but to 11 h, (9 h + 120 minutes). Here, the interest is that the calculation method is available, on principle, but is in conflict with the intuitive, pregnant, habit of the decimal basis calculation. In the relativism debate framework, it is elsewhere interesting to observe that, during the discussion, it was remarked that the correct (sexagesimal) calculation was a funny calculating way. This hinted it could be a foreign habit – due to the employee’s membership of a strongly exotic culture.

Purely cognitively speaking, this episode illustrates too the difficulties to explain, precisely without breaking down. The employee had first made the same mistake rapidly corrected, what had aroused some distrust, and an explicit preference for the previous, more intuitive, calculation. But a problem is the incapacity to use a neutral strategy to solve a problem too, especially when an interest (even limited to 40 minutes wages) is at stake.

From a sociopolitical standpoint, it is funny to notice that the employer’s wife, and a friend of her, have made the mistake and not the employee (a student, as it happens). The external authority of the husband, consulted by telephone, having elsewhere appeared, too, helpless to decide. Rather than a Marxist interpretation in term of reciprocal defense of (class) interest, I prefer to reason here in term of competence hierarchy, in this particular case of a status inversion [NOTE 130]. In the current social state, the hierarchy of competence can be upside down, in contradiction with the referential literary conventions (Le Bourgeois gentilhomme [The Molière’s Bourgeois Gentleman], Flaubert’s Bouvard et Pécuchet [in which nouveau riches are mocked at for their intellectual pretensions]…). This can be read as a deficiency of the social payment structure, and a resurgence of a caste society.

Another relativist example: oddly enough, the question of the thought as calculation is, actually, already solved by some local meanings. In the south of France, in popular language, the use of calculer (to reckon) is/was traditionally frequent for what in standard French is said to think [penser, réfléchir]. One can (or could, a little once) commonly meet wordings like: “Wait, I calculate,” etc. It is therefore possible to consider that the problem raises too as a simple question of cultural difference. One could still think that southerners are under-developed persons, or that the popular expression is metaphorical, and then despise this objection. Anyway, it will be necessary to abandon the simultaneous reference to cultural relativism, Sapir-Whorfian, if one doesn’t want to try to solve this very problem. As it remain to explain the cultural fact that some peoples seem to find natural to calculate, while others seem to see in thought a mysterious phenomenon, even magical. This seems to invert the under-development characterization, or to mean that the socialist or (Catholic) Christian rejection of business is too well assimilated in philosophy. Maybe would it be necessary to reinterrogate this tradition! Non-westerners peoples (or some pockets of resistance) know they calculate. Contrary to what the so-called anti-utilitarian sociologists believe, social protagonists calculate all what they can calculate. More, in philosophical terms, if the terms calculation and thinking are ostensive (descriptive), the phrase “thought is a calculation” is receivable as explanation. While other researches about the term thinking, especially in term of etymology, are useless, or ridiculous.

Lets remind that Heideggerro-Lacanians puns grounded on morphology precisely claim to a semantic value. The fact this etymological or symbolist value is often not available for native speakers themselves removes a lot of relevance to this morphological semantics (and obviously, all the more, to intentionality). It concerns simply archaic residues that language is carrying, and whose possible semantic value can be interesting, but of a very varied productivity. Elsewhere, the work of Lacanian philosophers themselves, which is like a symbolic analysis, can be considered as a calculation. And the corresponding pathology is precisely a patient excess of symbolist calculation – or a therapist’s one by dint of reproduction. One can thus simulate the generating of Lacanian or Heideggerian puns, or of simply humorous ones. The etymologism would allow returning to the literary persons, or to subjectivist psychoanalysts, the insult of being mechanistic. This simulation of the pathological or literary behavior allows us to consider cognitive normality as an acquisition of control mechanisms (and not as the etymological spontaneous combinatory).

The Ambiguity Obstacle

The linguistic cliché of the human language ambiguity has obviously become one of the favorite AI adversaries’ topics. The term ambiguity usually refers, for the automatic linguistics (working quasi exclusively on the word, or on the sentence), a simple polysemy that simply supposes, for the machine, the choice of one meaning among those available. Anyone meets the same problem for an analysis of any new word. But any difficulty is perceived in term of ontological impossibility:

“In confronting a new usage, a machine must either treat it as a clear case falling under the rules, or take a blind stab. A native speaker feels he has a third alternative. He can recognize the usage as odd, not falling under the rules, and yet he can make sense of it – give it a meaning in the context of human life in an apparently nonrulelike and yet nonarbitrary way.” (Dreyfus, p. 199).

We can see here the magical role of context [NOTE 131] (why should it be human alone?). But then, as in the previous example, human beings will do the same thing than a machine and for the same reason. What Dreyfus calls context will be simply processed as hints. A new statement will be understood only when hints are sufficient. A lack of clues would mean syntax and words doesn’t mean anything. Well, it is true that it is a poet’s idea [NOTE 132]. This is a serious indication to corroborate the suspicion of parapsychology. Elsewhere, even if something like intuition, or even thought transference, would exist, the means to establish them would be grounded on indices.

This tendency to base oneself on ambiguity, or to produce an infinity of meanings, most possibly amounts to over-interpretation. Umberto Eco, burning what he has formerly led some people to believe (whose he considers now as insane), challenges it today – it lefts some hopes in human being. Another evocation of infinity seems to be the claiming of links, constituting an immense semantic network, with all possible concepts. Does it amount to deny any analysis? It is pretty obvious that anything can be connected with anything, at least in term of general classifications, like typical classifications of living beings for instance (or by links, classics in AI or database, like: part of, or subset of, etc.). But it is obvious too the link types do not have the same characteristics. Links above are oriented and not symmetrical (all felines are mammals, but all mammals are not feline), while others will be.

Automatic Language Analysis

How automatic analysis specifically works? Contrary to philosophical constraints, accepted by some linguists, the ambiguities removing is possible by no matter what means. For the sake of what semantic theory, except of some linguistic sect one, could it be decreed that: “The central theory of meaning (semantics) deal with words and sentences in terms of their literal meaning, which is treated as not context-dependent.” (Winograd & Flores, p. 19).

The problem of ambiguities, as many others reminded here, is assumed by AI research. Each partial solution of ambiguity removing can be considered as a step toward a good modeling of the problem. It is true that the task seems difficult if we believe some canonical examples. Some, like ‘La petite brise la glace’ (The little-girl breaks the ice/The little breeze freezes her) in French, even include an ambiguity (polysemy) on each word:

la: determiner (article)/pronoun.
petite: adjective/name.
brise : name/verb.
la: determiner/pronoun.
glace: name/verb.

Each lexical ambiguity can be removed, on principle, by a formal means perfectly known. It can be syntax, semantics, valency (number of complements of some verbs)… Or the transmitter of the statement can simply be asked, as we already have seen. Usually, in linguistic, syntax is the most used means. It suffices to it in analyzing the possible sentence types, for instance, with classic Chomskyan models:

sentence = nominal group + verbal group
nominal group = determiner + name
nominal group  [NOTE 133] = determiner + name + adjective
verbal group = verb + nominal group
verbal group = pronoun + verb