Outline of Principes des systèmes intelligents (Principles of intelligent systems - 1990a)

• Words are specialized units, meaning emerges automatically to the speaker of a tongue when they are combined in a clause
• There is certainly more to thinking than words. But aren’t words a sufficient condition to thinking ?

• An intelligent system negotiates its knowledge with a user

2. What are the components of an intelligent system ?

• An intelligent system can be envisaged as a function: a user’s speech is input, the system’s responses is output

3. Does an intelligent system resemble a human being ?

• A computer is not intelligent, intelligence is assigned to a computer by its user

4. Does an intelligent system function like a human being ?

• For the past 25 years it is the increasingly intelligent behavior of the computer which has provided fruitful hypotheses about human intelligence
• Folk psychology is no source of inspiration for Artificial Intelligence

• An intelligent system possesses some knowledge
• An intelligent system transmits some knowledge
• An intelligent system learns and aims at learning
• An intelligent system negotiates the contents of its knowledge
• An intelligent system has its own history, i.e. personality

• The temptations of brute force

• Constraining output through a system of rules

• Human beings do not follow rules, they perform « educated » improvisation with self-correction
• Educated improvisation is channeled rather then strictly constrained

• An intelligent system stores prefabricated units of various sizes, ranging from individual words to entire pieces of discourse such as songs or poems

• Memory storage displays a structure which allows recollection, reasoning and speech generation

2. A historical perspective

• For Hobbes, thought is shaped by recollection ; for Locke thought is shaped by the structure of memory storage

3. Associationism

• From Aristotle to Freud

4. Thought and ideas

• The concept that thought is more than words and images combined is foreign to Chinese culture

5. Folk psychology

• The notion that there is a « theory » underlying folk psychology is flawed
• Folk psychology has not changed since antiquity
• Psychoanalysis is more distant from folk psychology than is cognitive science

6. Thought language

• The notion of a « thought language » entails an infinite regress

7. Words as signifiers

• Let us mimic thought with words only and examine whether there is something more needed to express « ideas »

1. Icons

• There is out of necessity an unconscious dimension to thought processes: the automatic hallucination of images and words taking place in thinking

2. Reasoning

• This unconscious dimension is opaque to introspection

• All empirical evidence suggests a network type of memory storage
• The memory network is most likely to be the template for speech generation

• Material associative chaining – such as "humility" / "humidity" – is likely to operate like a physical « resonance »

2. Semantic associative chaining

2.1. Synonymy

• Synonymy raises the issue of a confusion between linguistic and representational concerns

2.2. Linguistic and physical models

• Language’s first purpose is the pure operation of speech rather than the transmission of information about the world as it stands

2.3. Logic

• Logic has been abstracted from language, it is not the hard core of language as often assumed
• Much of the essence of thinking got lost in the abstraction process which led to the creation of logic as a separate rules system

2.4. Attribution

• Essential and accidental attributes behave differently when properties are inherited

2.5. « Autobiographical » associative chaining

• There is no memory trace which is not « autobiographical » ; but some memory traces are culturally shared

2.6. The « simple » connection

• The « simple » connection dominates thought processes in Far-Eastern cultures : notions are linked through symmetrical – reversible – connections
• We have become historically unable not to see inclusion and attribution in all thought processes ; we do so even in those belonging to entirely different cultures, « primitive mentality » phenomena are often nothing more

• The classical « semantic network » of knowledge representation cannot be implemented in an Artificial Intelligence system

2. Memory networks

• Words are unlikely to be stored as individual memory traces  it is the associative chaining of any two words which is likely to constitute a separately stored unit
• Indeed, the information « apple » is different when a link between Adam and Eve and when a link between pear and plum

3. The dual of a graph

• A dual representation allows for the distribution of a word over a set of associative chainings
• The P-Graph was designed to such effect

4. Belief and knowledge

• Every memory network has got a center and a periphery reflecting the history of its growth
• « Belief cores » are central to a memory network : the starting point of a memory network is out of necessity made out of « belief cores »
• « Knowledge elements » are revisable, « belief cores » are not

• Dreams are free walks within a memory network reflecting an idiosyncratic life history
• The relevance of what one says reflects the affect value it has for us

• A machine knows no affect but it can be programmed so as to mimic affect
• A memory network would be a static and dead template if it were not set into motion by an affect dynamics
• It is an affect dynamics which generates meaning from a memory network template
• Psychoanalysis is the science of affect dynamics
• Neurosis is an ailment deriving from a loss of accessibility to the entirety of the memory network
• A collection of memory traces connected by the affect, which constitutes a "preoccupation" or which has even become inaccessible, was called by Freud – after Jung’s usage – a « complex »
• A worry is a pit in our affect landscape : every thought which develops is captured by it, falls into it
• An intention is like a worry, but it is time-stamped
• Affect values draw gradients in the affect landscape, speech is thus goal-driven (a "potential" pit)
• Speaking leads to temporary relaxation of the affect values attached to the words uttered
• Psychosis may result from the inaccessibility of words which act as obligatory bridges between various parts of the memory network

• Meaning is attached to speech units of various sizes : an isolated word is pretty much "stranded" on its own

2. Reference

• Very few contexts allow a word to have a reference in the empirical world

3. Associative chaining

• The meaning of a word is the set of associative chainings it belongs to (similar to Wittgenstein’s « meaning as usage »)

4. Ambiguity

• Ambiguity only exists when one examines the whole set of associative chainings a word belongs to simultaneously ; in practice, the other « half » of the associative chaining always relieves the ambiguity

• An intention is a self-generated worry
• The French philosopher Merleau-Ponty drew the attention on the fact that one hears oneself speaking : to some extent one is made aware of what one says at the same time as others do

2. Intention as a post hoc reconstruction

• Is thought anything more than a meaningful reconstruction operated on the basis of what one has heard oneself say ?
• We paraphrase as thoughts the words we have uttered

3. The compulsive expression of a worry

• Is speech anything more than a compulsive expression of a worry ?

4. Conclusion

• Is there any discontinuity between what Freud labeled primary and secondary processes ? In other words, what is the function of consciousness ?
• The interpretation of the psychoanalyst consists in letting the patient’s « worry » grow within himself, until it can be named by him

• Two intelligent systems learn when they communicate – as they have out of necessity a distinct history
• Learning amounts to colonizing new cortical regions through the disconnection of meaningless links

• The definition of a word spreads out from the end-tails of the associative chainings it belongs to
• Contrast sets reveal the alternatives offered to a word’s associative chainings

• Different types of aphasia show that syntax and semantics can get impaired separately, suggesting the existence of two distinct linguistic mechanisms
• There is a clear distinction in our type of languages between contents words and framework words ; this is linked with a clear distinction between a semantic component of meaning and a syntactic component of meaning

2. The evidence from the Chinese

• There is no difference in ancient Chinese between contents and framework words : all words were supposed to be nouns (or adjectives) and meaning was supposed to emerge from their « pile-up »

3. Categoremes and syncategoremes

• Medieval scholasticism drew a clear distinction between contents and framework words and tried to elaborate a consistent linguistic theory about categoremes – which may act as subject and predicate in a clause – and syncategoremes - which may not

4. Syntax and semantics

• The distinction between syntax and semantics is likely to refer to a topological distinction (since the book was published I developed the hypothesis that syntax should be regarded as the critical points of a multi-dimensional semantic object when projected in the one-dimensional space of speech)

• Implementation of intelligent systems has revealed that formal logic is unable to reproduce most types of common reasoning

2. Logic is emergent

• That a type of behavior can be simulated as the expression of a system of rules, does not allow to infer that such behavior results in actuality from the following of the rules elicited : it may result from educated improvisation with self-correction
• Logic is a normative model which was abstracted from spontaneous associative chaining of clauses
• In the course of the abstraction process that led to the creation of logic, many elements constitutive of common speech got out of the picture

3. The syllogism

• The syllogism is a way of further exploring the definition of a term through extending one associative chaining to which the term belongs, with another one to which it does not belong, but the second term in the original associative chaining does

4. Logic is insufficient to ensure the compatibility of consecutive sentences

• Formal logic ensures the compatibility of consecutive sentences through fixing the framework words which compose it

• The issue of the role played by contents words in the sentence is subsumed under the word « truth »

• A new information is the creation of a new associative chaining with a particular affect value attached

3. Encyclopedic and descriptive knowledge

4. Potential world and actual world

• Aristotle’s distinction

5. Analytic and synthetic truth

• Kant’s distinction

6. Commonsense knowledge

7. Truth and temporality

• Is there anything like a distinction between short-term and long-term memory or is the extant evidence only suggesting that inscription (storage) of memory traces takes a certain time ?

• The notion which corresponds to our « truth » in ancient China is k’o, « appropriate »

• French psychoanalyst Jacques Lacan introduced the distinction between truth of speech applying when two people are in a dialog and truth of discourse which holds a universal realm of application

• Only few of our memory traces are likely to hold information about the world « as it stands »
• Some African languages distinguish what we know through experience and what we know from having heard about it ; we’re making nonsense of this distinction by translating the first as « knowledge » and the second as « belief » (or the other way round)
• Except in the case of conversion, what we integrate in our memory is what is compatible with what we already know (has already been stored) ; as an ancient Chinese thinker expressed it : we remember what accords with our prejudice

4. The masters of truth

• Expressions of « knowledge » vs. « belief », « truth » vs. « falsehood », etc. are polemical expressions based on fear (anxiety) and aggression (anger), and reflect the power balance between the speakers, they do not express the existence of mental states such as « knowledge » or « belief »

• According to an ancient Chinese treatise, to infer is « to use something which is the same in something he refuses to admit and in something he accepts, and show this to him »

3. Degrees of adhesion

• A conversation can be regarded as arm-wrestling between two memory networks
• « Persuasion », « conviction », are the ways one memory network imposes its structure upon another one

4. Intelligence as a disposition to negotiating

• Expressing varying degrees of adhesion to the contents of the clauses we are uttering is a way for expressing our emotions which cover a range extending from fear (anxiety) to aggression (anger)
• The expressions of knowing, believing, desiring, intending, etc. do not reflect « mental states » but preparedness to negotiate on the basis of our experiencing anxiety or anger

5. Opinion and demonstration

• When accounting for reasoning there is only a partial compatibility between the principles of logic, which are algorithmic, and the principles of adhesion, which are that of a power balance between the speakers
• So far, Artificial Intelligence has dealt with the algorithmic (logic) dimension of speech, not with the power balance dimension (adhesion)

6. Conclusion

• The notion of « truth » combines the concern for an accurate depiction of the world as it stands with that for the power balance which regulates the relationship between human beings

• Associative connectors are covered by formal logic : "all", "some", "is a…", etc.
• Compatibility modulators contain the logical connectors such as "and", "if… then…", but also those compatibility modulators which have resisted all attempts at integrating them within the domain of formal logic : "despite", "nonetheless", "though", etc.
• Contents words do also play a role for maintaining compatibility between chained clauses

2. Theoretical concepts and pre-systematic notions

• A word’s meaning gets focused in the unfolding speech ; some words remain pre-systematic notions others fortify into theoretical concepts
• The distinction between pre-systematic notions and theoretical concepts is borrowed from philosophy of science ; actually it applies as well to everyday conversations : a conversation equates with a « local theory »

• I have tried to broaden the perspective of Artificial Intelligence by integrating relevant contributions from anthropology, philosophy and psychoanalysis