IEML – Information Economy Meta Language – is a mathematical language with the same expressive power as a natural language. It has a dictionary of 3000 concepts plus a regular grammar allowing to build recursively an infinity of sentences representing concepts and relations.
IEML has the same expressive power as a natural language. In particular, it allows the construction of complex recursive sentences that may be used for reasoning and complex causal modelling. It can explain itself (IEML is its own metalanguage), and it can translate into other natural and specialized languages.
IEML is mathematical, which means that there are functions (algorithms) for the creation and recognition of concepts networks. You don’t create the nodes and the links of IEML semantic networks one by one, but you create them using algebraic functions. In addition, IEML semantics are computable: IEML sequences of phonemes (or chains of characters) self-decode into networks of concepts and these concepts and their relations can be read in natural languages.
Properties 1 and 2 make IEML the most advanced language for symbolic AI. IEML has been designed to be used as a universal metadata system, providing semantic interoperability at the conceptual level.
An Application in Healthcare
The main idea of the GMC (global medical communication) application is to use the growing availability of medical data, the power of artificial intelligence, and the principles of evidence based medicine in order to improve healthcare collective intelligence at a global level.
The whole system will respect the data exchange standard FHIR, and the personal data will be protected by following the SOLID specifications. The artificial intelligence will be hybrid, combining a carefully crafted semantic metadata system (or ontology, or knowledge graph) in IEML and the best techniques of machine learning.
The IEML semantic metadata system will make a distinct conceptual organization for risk factors, symptoms, biomarkers and disorders/treatments in order to make the machine learning findings as relevant as possible. The conceptual breakdown and structure of the overall ontology will be designed in such a way as to accommodate the different classification systems in use.
Examples of conceptual organization of symptoms (related to psychic functions) in the field of mental health
All concepts, presented here in English, have been constructed from the words of the IEML dictionary and following IEML regular grammatical rules. In GMC, the concepts will be readable in all major languages. They will be hypertextually related to their definitions, all the way down to the concepts of the dictionary.
Table 1 below organizes the major categories of psychic functions (and related symptoms) in mental health.
Table 1: the most general psychological functions in an ontology of mental health
I used the fundamental symmetries of IEML to order the universe of discourse of mental health symptoms. The top row corresponds to the emergent qualities of the psyche in its most virtual dimension (autonomy) while the bottom row corresponds to the most actual dimension, which concretely supports the other psychic functions (memory). Between these two dimensions, the ascending and descending information is coded and decoded on a subjective (sign), emotional (being) and cognitive (thing) mode, as we see in the three intermediate rows. The three columns of the table correspond roughly to a sign / being / thing partition and the binary subdivision within each of the fifteen rectangles corresponds to a division between the most virtual aspects (for the top cell) and the most actual (for the bottom cell).
Note, in the row « subjectivation » that the subject relates to himself in the first box, to the alter-ego in the second, and to reality in the third. The « thought » represents the way the subject relates to reality, hence its affinity with the thing. On the other hand, in the third box of the « emotion » row, mood is the emotion felt and affect is the emotion expressed.
The following tables show some examples of further conceptual subdivisions. All this ontology building is based on extensive reading and conversations with experts of the field.
Table 2: A subset of Table 1, row 1, column 2 « mental self-repair »
Table 3: A zoom into « psychological self-reference » in row 2, column 1 of Table 1.
Table 4: A zoom into « intersubjective dispositions » of Table 1, row 2, column 2.
Table 5: A zoom into « thought content » of table 1, row 2, column 3
The topic of this communication is IEML, a language that I have invented, called the Information Economy Metalanguage. This language has the expressive power of a natural language and it is also a mathematical category, or an algebra.
Introduction
Like several philosophers before me (Leibniz, Peirce, Saussure, Wittgenstein, Chomsky…), I have been passionate since my early youth about the question of meaning : What is it? How does it work? And how far had we gone in the mathematization of language and in the calculation of concepts? When I say mathematization of language, I mean algebraic modelling, not statistical approximation.
In the early 1990’s, I understood that the internet would become the main communication medium. It became obvious for me that the growth of computing power would augment language and cognition, like writing, printing and electronic media’s revolutions did previously. Even before the advent of the web I knew that the future of text would be dynamic hypertext. And finally, I foresaw that a common digital memory would be able to support a new kind of large scale collective intelligence.
If all this was true, we needed to create a unifying semantic coordinate system, a digital native language that would lend itself to calculation and programming. What remained was to solve the problem of the mathematization of semantics. But what are semantics, exactly? Let’s distinguish between three aspects of semantics: first a pragmatic aspect, second a referential or logical aspect, and third a linguistic aspect. These three aspects are simultaneous and interdependant.
For the pragmatic semantics the meaning of a speech is its effect on a social situation. It depends on social games, it is about relevance, and it is a matter of sociology and game theory
For the referential semantics, the meaning is the reality or the reference that is designated by the speech. It is about truth, and it is a matter of logic.
For the linguistic semantics, the meaning of a speech emerges from the sense of its words and their grammatical organization: it is mainly about conceptualization, and it is a matter of linguistics.
Linguistic semantics are obviously the basis on which the other aspects are built, and that is why I decided to construct a language that would have a univocal, transparent and computable linguistic semantics.
First IEML has the same expressive power as a natural language. In particular, it allows the construction of complex recursive sentences that may be used for reasoning and complex causal modelling. It is able to explain itself (IEML is its own metalanguage), and it can translate any other natural and specialized language.
At the same time, its semantics are computable, which means that IEML sequences of phonemes (or chains of characters) self-decode into networks of concepts and that these concepts and their relations can be read in natural languages. It also supports algorithms for the creation and recognition of concepts networks (ontologies, knowledge graphs, domain specific languages, etc.). You don’t create the nodes and the links of semantic networks one by one but through the use of algebraic functions. We’ll see some examples later.
Currently, I am happy to say that, after more than twenty five years of research and development, the construction of the language is finished and that we have an editor. The langage includes a dictionary of three thousand words (chose « published projects ») and a completely functional and regular grammar. The editor is complete with a parser and all kinds of practical functions to create and explore conceptual networks.
Note that the explanation of IEML sentences and ontologies will come after the following section about the words of the dictionary.
The Dictionary
The dictionary of this language contains three thousand words, representing elementary concepts. The number is small to facilitate the calculations on the machine side and the cognitive manageability on the human side. The words are organized into 120 paradigms.
Here are a few examples of the paradigms topics:
climates & landscapes;
continents & regions;
oceans & seas;
sky & meteorology;
animals & plants;
countries & states;
technical functions;
anthropological functions;
types of relations;
calendar time units;
life cycles;
generative mutations;
values & ideas;
signs & semiotic functions;
data curation & critical thinking;
complex feelings;
personality types;
body parts, etc.
The three thousand words can be used as a basis for defining (recursively) all possible and imaginable concepts by means of sentences.
The general problem in building the dictionary was: how to create the concepts that will be the most useful for the creation of new concepts? In a way it is a bootstrapping problem. You can see here the six basic symbols corresponding to the most elementary concepts.
The six semantic primitives of IEML
It is important to note that these symbols correspond to three types of symmetry, unary for emptiness, binary for virtual/actual and ternary for sign/being/thing. The above figure does not represent a sentence but the symmetry structure of the primitives.
The Emptiness expresses the absence, the void, the zero, the silence, the noise (as the contrary of information)…
The virtual denotes the potential, the soul, the abstract, the immaterial or the transcendent dimension of human experience.
The actual represents the effectiveness, the body, what is concrete, tangible, material or any immanent aspect of reality.
This echoes all kinds of dualities : heaven and earth, yin and yang, abstract class and individual element, and so on
A sign is an entity or an event that means something for someone. By extension, the semantic primitive « sign » points to symbols, documents, languages, representations, concepts, and anything that is semantically close to code, message or knowledge.
A being is a subject or an interpreter. It can be a human, a group, an animal, a machine or whatever entity or process endowed with self-reference and interpretation. By extension, « being » refers to psychic interiority, the mind, the ability to conceive or interpret, intentions, emotions, people, their relationships, communities, societies and values.
A thing – when it is labelled by a sign, is often called an object or a referent. By extension, « thing » categorizes what we are talking about, objects (abstract or concrete), contextual elements. It also refers to bodies, tools, technology, material equipment, empowerment, power, and efficiency.
Sign/being/thing corresponds roughly to the semiotic triangle sign/interpreter/reference but also to all kinds of ternarities like syntax, semantics and pragmatics ; proposition, judgement and state of things ; or legislative, judicial, and executive.
In fact, these conceptual symmetries correspond to very old traditions, I did not invent them, I have only collected and compacted them.From these six symbols, I have created the 3000 words organized in 120 paradigmatic tables.
A paradigm, or a paradigmatic table, is akin to the map of a semantic field. Every IEML word belongs to one paradigm, and one paradigm only.
A paradigm of words is generated by a morphological function that combines multiplicative and additive operations. The multiplicative operation has 3 roles: substance, attribute, mode, and it is non-commutative and recursive. A two-dimensional table has two variable multiplicative roles.
The paradigm of performative acts
In each cell of a table you have two expressions, one in IEML and one in natural language. The expression in natural language is the translation of the IEML word.
In IEML algebraic expressions, letters represent positions in symmetry systems, and punctuation marks represent recursive multiplication layers.
How to read a paradigmatic table
Let’s come back to the problem of words construction. Playing with the six primitive symbols as if they were building blocks, I created new concepts, and I continued recursively on this path. Again, the problem was to create the most general concepts, in order to allow the creation of new concepts in all possible directions of the semantic space.
How could I generate general concepts with my six first symbols, concepts able to cover all the directions of the semantic space? I began with the triad sign / being / thing.
Combining sign/being/thing
On the left you can see the function that generates this paradigm. You can think of the substance as a figure and the attribute as a background.
Note that when some role is empty, we just put E in this role. In other examples, E would have been in role substance or attribute. We cannot remove E because every role must be filled with a primitive. IEML expressions must be « sequences of primitive symbols », with completely regular syntactic rules.Nevertheless, there are rules allowing the elision of E, provided that the parser can read the syntactic structure thank to the punctuation marks.
On the first row, we have sign in substance. When signs interact with signs, we get interpretation, reasonning, imagination, thoughts: reflection. When signs interact with beings, express beings, and help beings to communicate, it is language. And when the signs are engraved into things, when they are reified in a way or in another, they become memory.
On the second row, we have being in substance. When beings gather by the use of common symbols, (rituals, totems, flags, laws, institutions, music, contracts, languages…) they make society. When beings interact with themselves and with other beings, there is pleasure and pain, joy and suffering, and the whole range of emotions. When the objectivity of things is imbued with the symbolic organization, the values and the work of beings, it becomes a livable world.
On the third row, we have thing in substance with the usual variation sign /being/thing in attribute. When the objectivity of things is registered in a propositional sign, it is the truth. When the materiality of things comes to support the being, it is the life. Finally, the interaction of things, their respective positions, their envelopments, their connections form the space.
These nine very elementary concepts represent different points of departure, all equally valid, for the description of human experience. By using the same kind of reasoning, I created eighteen other lower-case letters.
Construction of the 25 lower case letters of IEML
As long sequences of the 6 Upper Case primitive symbols would have been difficult to read and write for humans, some often used sequences of three upper case letters are simplified into twenty-five lower case letters. There are ten vowels and fifteen consonnants to help human reading and understanding.
You can see, on the left of the slide, the function that generates the twenty-five lower case letters. The four colors on the table represent four symmetry systems inherent to the lower case letters. We have already studied the blue section.
I will just comment on the yellow section, with the letters y o e u a i. The first row displays three virtual actions: know (related to sign), want (related to being) and can (related to thing). The second row shows three actual actions : to say or communicate (related to signs), to commit (related to being) and to do (related to thing). For comments on the other sections and more details on the 25 lowercase letters, see https://intlekt.io/25-basic-categories/
Now let’s look at two more examples of paradigms, at higher layers. In the human development paradigm the words combine two letters. The generative function is: (s+b+t+k+m+nd+f+l) × (y+o+e+u+a+i) × (E)
The nine rows correspond to our nine concepts s b t k m n d f l and these concepts are declined (like in « declension ») according to the six types of action that we have seen before in the yellow section.
The six columns corresponds to types of knowledge, types of will or orientations, types of skills, types of signs, types of social roles, and types of tools or technologies. Now let’s have a look at the fifth column (social roles).
Basic Social roles in IEML
The reorganization of a column or row into a new table is automatic in the IEML editor. We find here our nine consonnants in position of substance and the letter a, corresponding to the notion of commitment, is in position of attribute. The generative function is (s+b+t+k+m+nd+f+l) × a × E. This gives us nine basic social roles. The interpreter corresponds to “reflection”, the storyteller to “language”, the scribe to “memory”, the chief to “society”, the parent to “emotion”, the judge to “world”, the researcher to “truth”, the healer to “life” and the guardian to “space”. Of course there are many more social roles in the dictionnary. Another paradigm multiplies these nine by themselves resulting in eighty-one other social roles.
A paradigm of scientific disciplines and sub-disciplines, contains also the objects of study and the name of the specialists. It amounts to three hundred and ninety two words. The table below is just a little part of this paradigm.
An excerpt of the paradigm of scientific disciplines
In this table, corresponding to the function: (s.y.- + b.y.- + t.y.-) × (y.- + o.- + e.- + u.- + a.- + i.-) × ( s.y.-) you can see that the rows correspond respectively to philosophy, communication and history. The columns correspond respectively to « science », « politics », « economy », « communication », « sociology » and « technology ». Of course, the common themes have parallels in functional invariants. For example, everything related to history begins by “t.”
In short, every word is part of a paradigm and every paradigm is organized along very regular semantic symmetries that are reflected into syntactic symmetries.
Sentences and Hypertexts
Now, we are going to see how to make sentences and semantic networks in IEML.
Phrase example
On the slide above you have the translation in IEML of the English sentence: » In a house, a mother tells a story from a book, to her child, with love, among joyful laughter, before she falls asleep. » The IEML words are represented by their counterpart in English and it is the structure of the IEML sentence that is underlined here.
Like words, sentences are generated in paradigms by generative functions. And we are going to see an example of a sentence paradigm later. As for the morphological function, the syntagmatic function combines multiplicative and additive operations. But it is not exactly the same operations as in the morphological function.
The additive operations correspond to junctions (like: and, or, but, because, and so on…)
The multiplicative operation has nine roles: the root corresponds to the verb (or to the main noun when it is a noun phrase), the initiator corresponds to the subject of traditional grammar, the interactant corresponds to the object of traditional grammar, there is also a recipient role and five complement roles : causality, time, place, intention and manner.
In an IEML sentence, the grammar distinguishes four kinds of parts:
The concepts are identified by a hash sign. They can be words or nested sentences.
The inflections, are identified by a tilde. They precise the meaning of concepts. For exemple, at the root role, the verb « tell » is precised by an indicative mood and a present tense and the nouns in other roles are precised by a gender, a number, or an article.
The auxiliaries identified by a star, determine the particular cases of the complements. For exemple, in this sentence, the time role is precised by the auxiliary « before » and the place role by the auxiliary « in ».
Finally, the junctions are identified by an ampersand. You can see an example of « and » here at the manner role, on the last line of the sentence.
To give you an idea of the fine nuance you can achieve in IEML, there are eighty inflections, one hundred and thirty one auxiliaries and twenty nine junctions.
References in IEML
IEML can explicitly handle proper nouns and references that are not general categories. Everything that is inside angled brackets is not a general category but a reference. This is the way to handle proper nouns, numbers and data in general. Of course, it is possible to have IEML expressions in reference and therefore the language is self-referential and can be its own metalanguage.
Example of link
Link sentences are used to explain and connect words and sentences. As you can see on the slide above, semantic links are just sentences with arguments. You can have links with one, two, three, or four arguments that will connect respectively one, two, three or four conceptual nodes (words or sentences). The link represented on the slide says that «A is the contrary of B».
Example of function
Of course a link like « A is the contrary of B » can be used in a wide variety of cases. The actualization of the link is performed by a function which determines the domain of the arguments and the syntactic conditions for the creation of the link. In this case, the function will create links like « to the right » is the contrary of « to the left ».
Two remarks here:
First, the semantic relations are created by a syntactic function.
Second, because of the regular and symmetric structure of the paradigms, a function actualizes several links. So you don’t have to create the links one by one.
Like words, sentences can be organized in paradigms. The example depicted on the slide below comes from an ontology of mental health in IEML.
Example of sentence generating a paradigm
You have one constant role « symptoms related to perception » and two variable roles corresponding to the perception problems and to the senses that are affected. As you can see, the variables are between braces. And below you see the resulting table:
Paradigmatic table of perception problems
Now let’s recap! IEML has the same power as a natural language. It can handle…
narration and causal modelling,
dialogue, reasoning and translation,
indexation, reference and self-explanation.
It has the same power as a natural language *and* at the same time, it is also a mathematical category. This means that is organizes a morphism or a systematic correspondance between an algebra and a graph structure
The algebra is about the linearity of texts: all IEML expressions are punctuated sequences of the six primitives. We have probably a non-commutative ring (the demonstration have been made for the layer of words – see here chapter 5 – it is still a conjecture for the layer of sentences).
The graph structure is about the concept network. At the grammatical level you have syntactic trees of nested sentences which cross paradigmatic matrixes. This makes a rather interesting graph, a kind of rhizome. And on top of that, link sentences combined with syntactic conditions produce more semantic relations between concepts.
Of course mathematization does not mean necessarily quantification. It can be a formalization of qualitative structures. In particular, abstract algebra can handle all kinds of symmetry systems and not only in the realm of numbers and geometrical figures.
The secret of the computability of IEML semantics lies in its coding principle. Semantic symmetries (the signified) are coded by syntactic symmetries (the signifier). And paradigmatic matrixes are created by functions with constants and variables.
If I were not held back by my modesty, I would say that Chomsky mathematized the syntagmatic trees and that – thanks to the coding system I have just explained – I have added the mathematization of paradigmatic matrices.
Conclusion
Which new perspectives would IEML bring if it was adopted as a Semantic Protocol?
First, general semantic interoperability. Semantic interoperability means that – coded in IEML – the meaning will be computable and easily sharable. Semantic interoperability is not about formats (like RDF, for example) but about architectures of concepts, ontologies and data models that would be connected across different domains, because nodes and links can be brought back automatically to the same dictionary according to the same grammar. Semantic interoperability means essentially an augmented collective intelligence.
For neuronal AI, if the tokens taken into account by the models were variables of a semantic algebra instead of phonetic chains of characters in natural languages, the machine learning would be more effective, and the results would be more transparent and explainable. My intention is to pursue the research direction of « semantic Machine learning ». Labelling / annotating data with good ontologies helps *generalization* in machine learning!
For symbolic IA, we would have concepts and their relations generated by semantic functions. Even more importanly, the mode of definition of concepts would change radically. Instead of having concepts that are defined separately from each other by means of unique identifiers (the URIs) on the model of referential semantics, we would have concepts defined by other concepts of the same language, like in a dictionary.
We know that there are problems of accumulation, sharing and recombination of knowledge between AI systems / models. A semantic protocol based on IEML will lead to logical de-compartmentalization, neuro-symbolic integration, accumulation and fluid recombination of knowledge.
The blockchain domain is important because it means the automation of value allocation. Today, smart contracts are written in many different programming languages bringing problems of interoperability between machines and readability for non-programming humans. With a semantic protocol based on IEML,smart contracts would be readable by humans and executable by machines.
The metaverse is about an immersive, interactive, social and playful user experience. Today, it includes mainly simulations, reproductions or augmentations of a physical 4D universe. With a semantic protocol based on IEML, the Metaverse could contain new sensory-motor simulations of the world of ideas, memory and knowledge.
A scientific revolution has already started with the digitization of archives, the abundance of data produced by human activities, the increased computer power availability, and data sharing within transdisciplinary teams. The name of this revolution is of course «digital humanities». But the field is still plagued by theoretical and disciplinary fragmentation, and weak mathematical modelling. With a semantic protocol based on IEML, the world of meaning and value would be unified and made computable. (Again, this does not mean reduction to quantity, or any kind of reductionism, for that matter). It would foster the emergence of an inexhaustible and complex semantic cosmos allowing for every point of view and every interpretation system to express itself. It would also lead to a better exploitation of a common memory, bringing a more reflexive knowledge to human communities.
Language allows a dynamic coordination between networks of concepts within the members of a community, from the smallest scale like a family, or a team, to the largest political or economic units. It also enables storytelling, dialogue, questioning and reasoning. Language supports not only communication but also thought as well as the conceptual organization of memory, complementary to its emotional and sensorimotor structure.
But how does language work? On the receiving end, we hear a sequence of sounds that we translate into a network of concepts, bringing meaning to a statement. On the transmitting side, from a network of concepts that we have in mind – a meaning to be conveyed – we generate a sequence of sounds. Language is the interface between sound sequences and concept networks.
Instead of phoneme chains (sounds) there can also be sequence of ideograms, letters, or gestures like in the case of sign language. What remains constant among all languages and writing systems is this quasi-automatic interfacing between a sequence of sensible images (sound, visual, tactile), and a graph of abstract concepts (general categories). And relations between concepts are also considered concepts themselves.
This reciprocal translation between a sequence of images (the signifier) and networks of concepts (the signified) suggests that a mathematical category could model language by organizing a correspondence between an algebra and a graph structure. The algebra would regulate reading and writing operations on texts, while the graph structure would organize operations on nodes and oriented links. To each text would correspond a network of concepts, and vice versa. Operations on texts reflect operations on conceptual graphs in a dynamic way.
Once we have a regular language to encode strings of signifiers, we could transform a sequence of symbols into syntagmatic trees (syntax is the order of syntagms) and vice versa. However, if its syntagmatic tree – its internal grammatical structure – is indispensable to the understanding of the meaning of a sentence, it is not sufficient. For each linguistic expression lies at the intersection of a syntagmatic axis and a paradigmatic axis. A syntagmatic tree represents the internal semantic network of a sentence, the paradigmatic axis represents its external semantic network and in particular its relations with sentences having the same structure, but from which it is distinct. To understand the phrase « I choose the vegetarian menu », it is of course necessary to recognize that the verb is « to choose », the subject « I » and the object « the vegetarian menu » and to know moreover that « vegetarian » qualifies « menu ». But one must also recognize that vegetarian is opposed to meaty and to vegan, therefore remembering that there are semantic opposition systems in the language. Establishing semantic relations between concepts presupposes that we recognize the syntagmatic trees internal to sentences, but also the concepts and their components which belong to paradigmatic matrices external to the sentence, specific to a language or held by certain practical domains.
Because natural languages are ambiguous and irregular, I have designed a mathematical language (IEML) translatable into natural languages, a computable language which can encode algebraically not only syntagmatic trees, but also the paradigmatic matrices where words and concepts take their meaning. Every sentence in the IEML metalanguage is located precisely at the intersection of a syntagmatic tree and paradigmatic matrices.
Based on the regular syntagmatic-paradigmatic grid of IEML, we can now generate and recognize semantic relations between concepts in a functional way: knowledge graphs, ontologies, and data models. On the AI side, the encoding of labels, or data categorization, in this algebraic language that is IEML would certainly facilitate machine learning.
Technically, IEML is a lightweight and decentralized project. It consists of an IEML natural language dictionary (in French and English so far), an open-source parser supporting computable functions on language expressions with a platform for collaborative editing and sharing of concepts and ontologies. Beyond AI, my vision for IEML is to foster the semantic interoperability of digital memories as well as a synergy between personal cognitive empowerment and the transparency and reflexivity of collective intelligence. The development, maintenance and use of a semantic protocol based on IEML would require ongoing research and training efforts.
The digital turn in humanities and social sciences
“About memories” by Hiroko Kono (2011)
This post reports on my presentation at the conference Humanistica in Montréal, the 20th of May 2022.
Humanities scholars build databases for analysis, mining, and sharing. Indexingonline documents is crucial for authors, publishers and readers. Today, there is a multiplicity of semantic metadata systems and ontologiesto classify data and documents. These systems (often inherited from the printing era) vary widely according to languages, disciplines, traditions and theories. In this context, the IEML metalanguage proposes a programmable modeling and indexing tool, capable of ensuring semantic interoperability, without standardizing the different points of view.
In IEML, each concept has a unique linguistic representation. Expressed in IEML rather than in natural language, concepts become more explicit and transparent. Moreover, IEML is a univocal language, without lexical or syntactic ambiguities. Concepts from different ontologies are composed of words of the same dictionary according to a regular grammar.
It is therefore possible to collaboratively exchange models and sub-modelsbetween researchers speaking different languages and coming from different disciplines. In short, IEML solves the problem of semantic interoperability.
A platform for collaborative design and maintenance of semantic graphsis in sight (ontologies, indexing systems, labels for machine learning, etc.).
A new semantic tool
IEML is not a format – of data or metadata – but a language that has : • a compact dictionary with 3000 words (accessible in English and French) • a fully regular grammar • all integrated into an editor-parser
IEML has the same semantic qualities and strengths as a natural language. Thus, IEML can translate all natural languages and can serve as a hub between natural languages. Its semantics is computable because it is a functionof its syntax(which is regular). Intended for the construction of semantic graphs, its sentences can take two forms: nodes or links. IEML has instructions for programming semantic graphs such as: hypertexts, ontologies and data models.
The words
Using the grammar and the words of the dictionary, the IEML editor allows to recursively generate as many concepts as needed. IEML words are built regularly from 6 primitive symbols.
The 6 IEML primitive symbols
The six primitives, like the other letters of IEML are words and denote concepts when used alone. But when used inside another word, they represent places in symmetry systems: symmetry 1 for E, symmetry 2 for U/A, symmetry 3 for S/B/T. For more details on IEML primitives, see: https://intlekt.io/semantic-primitives/
The generative operation for the words has 3 roles: substance ☓ attribute ☓ mode By combining this generative operation (☓) with an additive (+) operation, it becomes possible to create various word paradigms, that are always organized in tables. Below, the 25 lowercase letters are united in a paradigmatic table that multiplies U+A+S+B+T in substance by U+A+S+B+T in attribute, with an always empty mode.
The 25 lower case letters in IEML
In the image above, the colors indicate four systems of symmetries (4, 6, 6, 9) whose letters occupy specific positions. Learn more about the 25 lowercase letters: https://intlekt.io/25-basic-categories/
IEML words paradigms (and sentences paradigms as well) are systems of semantic symmetries represented by systems of syntactic symmetries. For example, the paradigm below organizes spatial relationships. The first two rows organize spatial relationships along the vertical (first row) and horizontal (second row) axes. The three rows at the bottom of the table organize entry and exit, lateralization, and pathways.
Paradigm of spatial relationships. Click on the image to enlarge 😉
Let us remember that the IEML dictionary is above all a toolbox for building new categories by means of phrases.
The sentences
The nine roles of the sentence – in green in the example – along with *auxiliaries, ~inflections and & junctions allow for the expression of narrativesand causal explanations.
Example of an IEML phrase
Evocation
In IEML, semantic relations are not created one by one « by hand » but areprogrammed. The instruction for creating semantic relations below is broken down into two parts. In the example below, the part that starts with @link states the link sentence with the two variables $A and $B: « Word A means the opposite of word B ». The numbers 0, 1 and 8 are shortcuts for the sentence roles: root, initiator and manner. The part that begins with @function states the domain (in this case the spatial relation paradigm above) that is concerned with relation creation. It states the necessary conditions for the creation of relations in terms of syntactic addresses and content. The function for creating relations uses only two types of « equations » connected by « AND » and « OR »: syntax address A == syntax address B syntax address A == content c
Example of a declarative instruction for creating semantic relations
Note that the function shown above creates 30 semantic relations at once!
Indexing, proper names, reference and self-reference.
IEML explicitly handles proper names and references that are not general categories. The example below gives three examples: a name, a number and a hyperlink. To learn more about the treatment of proper names in IEML see: https://intlekt.io/proper-names-in-ieml/
Examples of references in IEML
IEML can also refer to its own expressions: links, definitions, comments, etc. The example below is taken from the spatial relations paradigm and the relation « word A means the opposite of word B » discussed above.
Example of self-reference in IEML
Several ontologies are currently under development. Do not hesitate to contact us if you are interested in IEML!
The goal of this essay is to present an overview of the limitations of contemporary AI (artificial intelligence) and to propose an approach to overcome them with a computable semantic metalanguage. AI has made considerable progress since the time of Claude Shannon, Alan Turing, and John von Neumann. Nevertheless, many obstacles have appeared on the road paved by these pioneers. Today, symbolic AI is rooted in conceptual modeling and automatic reasoning, while neural AI excels in automatic categorization; however, both the symbolic and neural approaches encounter many problems. The combination of the two branches of AI, while desirable, fails to resolve either the compartmentalization of modeling, or the challenges of exchanging and accumulating knowledge.
Innate human intelligence solves these problems with language. Therefore, I propose that AI adopts a computable and univocal model of the human language, the Information Economy Metalanguage (IEML), a semantic code of my own invention. IEML has the expressive power of a natural language and the syntax of a regular language. Its semantics are unambiguous and computable because they are an explicit function of its syntax. A neuro-semantic architecture based on IEML combines the strengths of neural AI and classical symbolic AI while enabling integration of knowledge through an interoperable computing of semantics. This can open new avenues for Artificial Intelligence to create a synergy between the democratization of data control and the enhancement of collective intelligence.
Bibliographical references and links for further reading are located at the end of this essay.
Introduction
Let’s first examine how the term « artificial intelligence » (AI) is used in society at large, for example in journalism and advertising. Historical observation reveals the tendency to classify the “advanced” applications into artificial intelligence in the eras in which they first emerge; however, years later, these same applications are often reattributed to everyday computing. For example, visual character recognition, originally known to be AI, is now considered to be commonplace and is often integrated into software programs without fanfare. A machine capable of playing chess was celebrated as a technical achievement in the 1970s, but today you can easily download a free chess program onto your smartphone without a hint of astonishment from anyone. Moreover, depending on whether AI is trendy (as it is today) or discredited (as it was in the 1990s and 2000s), marketing strategies will either emphasize the term AI or replace it with others. For example, the « expert systems » of the 1980s become the innocuous « business rules » in the 2000s. This is how identical techniques or concepts change name according to the fashion, making the perception of the domain and its evolution particularly opaque.
Let’s now leave the vocabulary of journalism or marketing to investigate academic discipline. Since the 1950s, the designated branch of computer science that is concerned with modeling and simulating human intelligence is called Artificial Intelligence.
Computer modeling of human intelligence is a worthy scientific goal that has had, and will continue to have, considerable theoretical and practical benefits. Nevertheless, most researchers in the field do not believe that autonomous intelligent machines will soon be built, notwithstanding the early, enthusiastic predictions about AI’s capability declared in the early years that were later contradicted by the facts. Much of the research in this field – and most of its practical applications –is aimed at augmenting human cognition rather than mechanically reproducing it. This is in contrast to the program of research focused on the construction of an autonomous general Artificial Intelligence.
I have defended the idea of Artificial Intelligence in the service of collective intelligence and human development in my book, La Sphère Sémantique. Let’s continue this line of thought here in this essay.
From a technical perspective, AI is split into two main branches: statistical and symbolic. A statistical AI algorithm « learns » from the supplied data. It therefore simulates (imperfectly, as we will see below) the inductive dimension of human reasoning. In contrast, symbolic AI does not learn from data, rather it depends on the logical formalization of a domain of knowledge as designed by engineers. In principle, compared to statistical AI, it therefore demands a greater amount of human intellectual work. A symbolic AI algorithm applies the rules it was given to the supplied data. Hence, it simulates more of the deductive dimension of human reasoning. I will successively review these two main branches of AI, with a particular focus on underlining their limitations.
AI and its limitations
Neural AI
The statistical branch of AI involves the training of algorithms from massive accumulations of data to enable the recognition of visual, audio, linguistic and other form of information. This is called machine learning. When we talk about AI in 2022, we generally designate this type of technical and scientific research program. As we have observed, statistical AI uses human labor sparingly in comparison to symbolic AI. Instead of having to write a pattern recognition program, it suffices to provide a set of training data for the machine learning algorithm. If, for example, statistical AI is given millions of images of ducks with labels specifying that the image represents a duck, it learns to recognize a duck and, upon completion of its training, will be able to affix the label « duck » on an uncategorized image of that bird. Nobody explained to the machine how to recognize a duck: it is enough to simply give it examples. Machine translation works on the same principle: a statistical AI is given millions of texts in language A, accompanied by their translation into language B. Trained with these examples, the system learns to translate a text from language A into language B. This is how machine translation algorithms like DeepL or Google Translate work. To take an example from another field, the statistical AI used to drive « autonomous vehicles » also works by matching two sets of data: images of the road are matched with actions such as accelerating, braking, turning, etc. In short, statistical AI establishes a connection (mapping) between a set of data and a set of labels (in the case of pattern recognition) or between two sets of data (in the case of translation or autonomous vehicles). Statistical AI therefore excels in categorization, pattern recognition and matching between perceptual and motor data.
In its most advanced version, statistical AI is rooted in neural network models that roughly simulate the way the brain learns. These models are called « deep learning » because they are based on the overlapping of multiple layers of formal neurons. Neural networks are the most complex and advanced sub-field of statistical AI. This neural type of artificial intelligence has been around since the origin of computer science, as illustrated by the research of McCulloch in the 1940s and 50s, Frank Rosenblatt and Marvin Minsky in the 1950s, and von Foerster in the 1960s and 70s. Significant work in this area was also done in the 1980s, especially involving David Rumelhart and Geoffrey Hinton, among others, but all this research had little practical success until the 2010s.
Besides certain scientific refinements of the models, two factors independent of theoretical advances explain the growing use of neural networks: the availability of enormous amounts of data and the increase in computing power. Starting in the second decade of the 21st century, organizations are engaged in digital transformation and a growing share of the world’s population is using the Web. All this generates gargantuan data flows. The information thus produced is processed by large digital platforms in data centers (the « cloud ») that concentrate unprecedented computing power. At the beginning of the 21st century, neural networks were implemented by processors originally designed for computer graphics, but nowadays, the data centers owned by Big Tech already use processors specifically designed for neural learning. Thus, interesting but impractical theoretical 20th century models have suddenly become quite relevant in the 21st century, to the point of supporting a new industry.
Artificial intelligence or synthesis and mobilization of data?
The year 2022 has seen the triumph of what is called generative AI. Models trained on huge masses of data are able to produce texts (ChatGPT) or images (DALLE2, Stable Diffusion, etc.) of amazing quality. It is possible to control the output by means of « prompts » and, increasingly, by a dialogue allowing the user to send feedback to these systems. It is clear that these are remarkable advances in the field of software and the applications are already numerous, especially in the field of augmented creativity and pattern recognition. But the name « artificial intelligence » that is given to these systems can be misleading. It is enough to interact a little with ChatGPT to realize that the model does not really understand what it is told, nor the texts it produces: errors of logic or elementary arithmetic, factual errors, ethical monstrosities and other « hallucinations » (which Twitter users amuse themselves with) underline the limits of the purely statistical approach. I agree that these new tools, by synthesizing and mobilizing the content of huge databases, are capable of increasing or replacing human work. They do announce a new wave of automation with economic and cultural consequences that are still difficult to predict. But we are far from an autonomous intelligence capable of a true knowledge or understanding of the world. And it seems that more data, more parameters (number of internal connections to the model), and more computing power are not able to remedy the root of the problem.
The main problems rest in the quality of training data, the lack of causal modeling, the inexplicable nature of some of the results, the absence of generalization, the reportedly inscrutable meaning of the data, and finally, the difficulties in accumulating and integrating knowledge.
Quality of Training Data
A Google engineer is quoted as saying jokingly, « Every time we fire a linguist, our machine translation performance improves. » But while statistical AI is known to have little need for human labor, the risks of bias and error pointed out by increasingly concerned users are driving the need for better selection of training data including more careful labeling. Yet, this requires time and human expertise, precisely the factors that one is hoping to eliminate.
Absence of an Explicit Causal Hypotheses
All statistics courses start with a warning about the confusion between correlation and causation. A correlation between A and B does not prove that A is the cause of B. It may be a coincidence, or B may be the cause of A, or even a factor C not considered by the data collection is the real cause of A and B, not to mention all the complex systemic relationships imaginable involving A and B. Yet, machine learning is based on matching datasets through correlations. The notion of causality is foreign to statistical AI, as it is with many techniques used to analyze massive data collections, even though causal assumptions are often implicit in the choice of datasets and their categorization. In short, contemporary neural/statistical AI is not capable of distinguishing cause from effect. So far, when using AI to assist with decision making and more generally for orientation in practical domains, explicit causal models are indispensable, because for actions to be effective they must intervene on the causes.
In an integral scientific approach, statistical measurements and causal hypotheses work in unison and reciprocate control. But to only consider statistical correlations would create a dangerous cognitive blind spot. As for the widespread practice of keeping one’s causal theories implicit, it prevents relativizing them, comparing them with other theories, generalizing, sharing, criticizing, and improving them.
Inexplicable Results
The functioning of neural networks is opaque. Millions of operations incrementally transform the strength of connections of neural assemblies which themselves are made of hundreds of layers.
Since the results of these operations cannot be explained or justified conceptually in a way that humans can understand, it is difficult to trust these models. This lack of explanation becomes worrisome when machines make financial, legal, medical, or autonomous vehicle driving decisions, not to mention military applications. To overcome this obstacle, and in parallel with the development of a more ethical artificial intelligence, more and more researchers are exploring the new research field of « explainable AI ».
The Lack of Generalization
At first glance, statistical AI presents itself as a form of inductive reasoning, i.e., as an ability to infer general rules from a multitude of cases. Yet contemporary machine learning systems fail to generalize beyond the limits of the training data with which they have been provided. Not only are we – humans – able to generalize from a few examples, whereas it takes millions of cases to train machines, but we can abstract and conceptualize what we have learned while machine learning fails to extrapolate, let alone, conceptualize. Statistical AI remains at the level of purely reflex learning, its generalization narrowly circumscribed to the supplied examples with which it is provided.
Inaccessible Meaning
While performance in Machine translation and automatic writing (as illustrated by the GPT3 program) is advancing, machines still fail to understand the meaning of the texts they translate or write. Their neural networks resemble the brain of a mechanical parrot only capable of imitating linguistic performance without understanding an iota of the content of the texts it is translating. In a nutshell, contemporary Artificial Intelligence can learn to translate texts but is unable to learn anything from these translations.
The Problem of Accumulation and Integration of Knowledge in Statistical AI
Bereft of concepts, statistical AI has difficulty in accumulating knowledge. A fortiori, the integration of knowledge from various fields of expertise seems out of reach. This situation does not favor the exchange of knowledge between machines. Therefore, it is often necessary to start from scratch for each new project. Nevertheless, we should point out the existence of natural language processing models such as BERT, which are pre-trained on general data, and which then have the possibility of specializing. A form of capitalization is possible in a limited fashion. But it remains impossible to integrate all the objective knowledge accumulated over the centuries by humanity into a neuro-mimetic system.
Symbolic AI and its Limits
During the last seventy years, the symbolic branch of AI has successively corresponded to what has been known as: semantic networks, rule-based systems, knowledge bases, expert systems, semantic web and, more recently, knowledge graphs. Since its origins in the 1940s-50s, a good part of computer science de facto belongs to symbolic AI.
Symbolic AI encodes human knowledge explicitly in the form of networks of relations between categories and logical rules that enable automatic reasoning. Its results are, therefore, more easily explained than those of statistical AI.
Symbolic AI works well in the closed microworlds of games or laboratories, but quickly becomes overwhelmed in open environments that do not follow a small number of strict rules. Most symbolic AI programs used in real-world work environments solve problems only in a narrowly limited domain, whether it’s medical diagnosis, machine troubleshooting, investment advice, etc. An « expert system, » in fact, functions as a medium for the encapsulation and distribution of a particular know-how which can be distributed wherever it is needed. The practical skill then becomes available even in the absence of human expertise.
At the end of the 1980s, after a series of ill-considered promises followed by disappointments began what has been called the « winter » of artificial intelligence (all trends combined). However, the same processes continue to be applied to solve the same types of problems indiscriminately We have only abandoned the general research program in which these methods were embedded. Thus, at the beginning of the 21st century, the business rules of enterprise software and the ontologies of the Semantic Web have succeeded the expert systems of the 1980s. Despite the name changes, it is easy to recognize in these new specialties the old processes of symbolic AI.
In the early 2000s, the Semantic Web has been aimed at exploiting all the information available on the Web. To make the data readable by computers, different domains of knowledge or practice are organized into coherent models. These are the « ontologies », which can only reproduce the logical compartmentalization of previous decades, even though computers are now much more interconnected.
Unfortunately, we find in symbolic AI the same difficulties in the integration and accumulation of knowledge as in statistical AI. This compartmentalization is in opposition with the original project of Artificial Intelligence as a scientific discipline, which wants to model human intelligence in general, and which normally tends towards an accumulation and integration of knowledge that can be mobilized by machines.
Despite the compartmentalization of its models, symbolic AI is, however, slightly better off than statistical AI in terms of accumulation and exchange of data. A growing number of companies, starting with the Big Tech companies, are organizing their databases by using a knowledge graph which is constantly being improved and augmented.
Moreover, Wikidata offers a good example of an open knowledge graph through which the information that is gradually accumulating can be read just as well by machines as it can be by humans. Nevertheless, each of these knowledge graphs is organized according to the – always particular – purposes of its authors and cannot be easily reused to other ends. Neither statistical AI nor symbolic AI possess the properties of fluid recombination that we should rightly expect from the modules of an Artificial Intelligence at the service of collective intelligence.
Symbolic AI is a Voracious Consumer of Human Intellectual Work
There have been many attempts to contain all human knowledge in a single ontology to allow better interoperability, but then the vibrancy, complexity, evolution, and multiple perspectives of human knowledge are erased. On a practical level, universal ontologies – or even those that claim to formalize all the categories, relations, and logical rules of a vast domain – quickly become huge, cumbersome, and difficult to understand and maintain for the human who must deal with them. One of the main bottlenecks of symbolic AI is the quantity and high quality of human work required to model a domain of knowledge, however narrowly circumscribed. Indeed, not only is it necessary to read the literature, but it is also necessary to interview and listen at length to several experts in the domain to be modeled. Acquired through experience, the knowledge of these experts is most often expressed through stories, examples, and descriptions of typical situations. It is then necessary to transform empirical, oral knowledge into a coherent logical model whose rules must be executable by a computer. Eventually, the reasoning of the experts will be automated, but the « knowledge engineering » work from which the modeling proceeds cannot be.
Problem Position: What is the Main Obstacle to (Further) AI Development?
Towards a Neuro-symbolic Artificial Intelligence
It’s now time to take a step back. The two branches of AI – neural and symbolic – have existed since the middle of the 20th century and they correspond to two cognitive styles that are equally present in humans. On the one hand, we have pattern recognition, which corresponds to reflex sensorimotor modules, whether these are learned or of genetic origin. On the other hand, we have explicit and reflective conceptual knowledge, often organized in causal models and which can be an object of reasoning.
Since these two cognitive styles work together in human cognition, there is no theoretical reason not to attempt to make them cooperate in Artificial Intelligence systems. The benefits are obvious and each of the two subsystems can remedy problems encountered by the other. In a mixed AI, the symbolic component overcomes the difficulties of conceptualization, generalization, causal modeling, and transparency of the neural component. Symmetrically, the neural component brings the capabilities of pattern recognition and learning from examples that are lacking in symbolic AI.
Both important AI researchers and many knowledgeable observers of the discipline are gravitating in the direction of a hybrid AI. For example, Dieter Ernst recently advocated an “integration between neural networks, which excel at perceptual classification and symbolic systems, which in turn excel at abstraction and inference”. [1]
Following in the footsteps of Gary Marcus, AI researchers, Luis Lamb and Arthur d’Avila Garcez, recently published a paper in favor of a neuro-symbolic AI in which representations acquired by neural means would be interpreted and processed by symbolic means. It seems that we have found a solution to the problem of the blockage in AI development: it would be beneficial to intelligently couple the symbolic and statistical branches rather than keeping them separate as two competing research programs. Besides, don’t we see the Big Tech companies, which highlight machine learning and neural AI in their public relations efforts, discreetly developing knowledge graphs internally to organize their digital memory, and to make sense of the results of their neural networks? But before we declare the issue settled, let’s think a little more about the givens of the problem.
Animal Cognition and Human Cognition
For each of the two branches of AI, we have listed the obstacles that stand in the way of a less fragmented, more useful, and more transparent Artificial Intelligence. Yet, we found the same drawback on both sides: logical compartmentalization, and the difficulties of accumulation and integration. Bringing together the neural and the symbolic will not help us to overcome this obstacle, since neither of them can do so. Yet, actual human societies can transform tacit perceptions and experiential skills into shareable knowledge. By dint of extensive dialogue, a specialist in one field will eventually make himself understood by a specialist in another field and may even teach him something. How can this kind of cognitive performance be reproduced in machine societies? What factor plays the integrative role of natural language in Artificial Intelligence systems?
Many people think that since the brain is the organic receptacle of intelligence, neural models are the key to its simulation. But what kind of intelligence are we talking about? Let’s not forget that all animals have a brain, and it is not the intelligence of, for example, the gnat or the whale that AI wants to simulate, but that of the human being. And if we are « more intelligent » than other animals (at least from our point of view) it is not because of the size of our brain. Elephants have bigger brains than humans in absolute terms, and the ratio of brain size to body size is greater in mice than in humans. It is mainly our linguistic capacity, predominantly processed in the Broca and Wernicke areas of the brain (unique to the human species), that distinguishes our intelligence from that of other higher vertebrates. However, these language processing modules are not functionally separate from the rest of the brain; on the contrary, they inform all our cognitive processes, including our technical and social skills. Our perceptions, actions, emotions, and communications are linguistically coded, and our memory is largely organized by a system of coordinated semantics provided by language.
That’s fine, one might say. Isn’t simulating human symbolic processing abilities, including the linguistic faculty, precisely what symbolic AI is supposed to do? But then, why is it that AI is compartmentalized into distinct ontologies, yet it struggles to ensure the semantic interoperability of its systems, and it has much difficulty in accumulating and exchanging knowledge? Simply because, despite its name of « symbolic, » AI still does not have a computable model of language. Since Chomsky’s work, we know how to calculate the syntactic dimension of languages, but their semantic dimension remains beyond the reach of computer science. To understand this situation, it is necessary to recall some elements of semantics.
Semantics in Linguistics
From the viewpoint of the scientific study of language, the semantics of a word or a sentence can be broken down into two parts which are combined in practice, yet conceptually distinct: linguistic semantics and referential semantics. Linguistic semantics deals with the relationship between words, while referential semantics is concerned with the relationship between words and things.
Linguistic Semantics (word-word). A linguistic symbol (word or sentence) generally has two aspects: the signifier, which is a visual or acoustic image, and the signified, which is a concept or a general category. For example, the signifier « tree » has the following meaning: « a woody plant of variable size, whose trunk grows branches starting at a specific height ». Given that the relationship between signifier and signified is established by a language, the signified of a word or a sentence is defined as a node of relationships with other signifieds. In a classical dictionary, each word is situated in relation to other associated words (the thesaurus), and its meaning is explained by sentences (the definition) that use other words that are themselves explained by other sentences, and so on, in a circular fashion. Linguistic semantics are fundamental to a classical dictionary. Verbs and common nouns (e.g., tree, animal, organ, eat) represent categories that are themselves connected by a dense network of semantic relations such as: « is a part of, » « is a type of, » « belongs to the same context as, » « is the cause of, » « is prior to, » etc. We think and communicate in the human way because our collective and personal memories are organized in general categories connected by semantic relations.
Referential Semantics (word-thing). In contrast to linguistic semantics, referential semantics bridges the gap between a linguistic symbol (signifier and signified) and a referent (an actual individual). When I say that « oaks are trees, » I am specifying the conventional meaning of the word « oak » by placing it in a species-to-genus relationship with the word « tree »; therefore, I am strictly bringing linguistic semantics into play. But, if I say that « That tree in the yard is an oak, » then I am pointing to a real situation, and my proposition is either true or false. This second statement obviously brings linguistic semantics into play since I must first know the meaning of the words and English grammar to understand it. But, in addition to the linguistic dimension, referential semantics are also involved since the statement refers to a particular object in a concrete situation. Some words, such as proper nouns, have no signified; their signifier refers directly to a referent. For example, the signifier « Alexander the Great » refers to a historical figure and the signifier « Tokyo » refers to a city. In contrast to a classical dictionary which defines concepts or categories, an encyclopedic dictionary contains descriptions of real or fictitious individuals with proper nouns such as deities, novel heroes, historical figures and events, geographical objects, monuments, works of the mind, etc. Its main function is to list and describe objects external to the system of a language. It therefore records referential semantics.
Nota bene: A category is a class of individuals, an abstraction. There can be categories of entities, processes, qualities, quantities, relations, etc. The words « category » and « concept » are treated here as synonyms.
Semantics in AI
In computer science, the real references, or individuals (the realities we talk about) become the data while the general categories become the headings, fields or metadata used to classify and retrieve data. For example, in a company’s database, « employee name », « address » and « salary » are categories or metadata while « Tremblay », « 33 Boulevard René Lévesque » and « 65 K $ / year » are data. In this technical domain, referential semantics corresponds to the relationship between data and metadata and linguistic semantics, to the relationship between metadata or organizing categories, which are generally represented by words, or short linguistic expressions.
Insofar as the purpose of computer science is to increase human intelligence, one of its tasks must help us make sense of the flood of digital data and to extract as much usable knowledge as possible from them. To that end, we must correctly categorize data – that is, implement word-things semantics – and organize the categories according to relevant relations which allow us to extract all the actionable knowledge from the data – which corresponds to word-word semantics.
When discussing the subject of semantics in computer science, we must remember that computers do not spontaneously see a word or a sentence as a concept in a certain relation to other concepts in a language, but only as a sequence of letters, or « string of characters ». Therefore, the relationships between categories that seem obvious to humans and that are part of linguistic semantics, must be added – mostly by hand – to a database if a program is to take them into account.
Let’s now examine the extent to which symbolic AI models semantics. If we consider the ontologies of the « Semantic Web » (the standard in symbolic AI), we discover that the meaning of words do not depend on the self-explanatory circularity of language (as in a classical dictionary), but that words points to URIs (Uniform Resource Identifiers) in the manner of referential semantics (as in an encyclopedic dictionary).
Instead of relying on concepts (or categories) that are already given in a language and that appear from the start as nodes of relations with other concepts, the scaffolding of the Semantic Web relies on concepts that are defined separately from each other by means of unique identifiers. The circulation of meaning in a network of signified is discounted in favor of a direct relationship between signifier and referent, as if all words were proper nouns. In the absence of a linguistic semantics based on a common grammar and dictionary, ontologies thus remain compartmentalized. In summary, contemporary symbolic AI does not have access to the full cognitive and communicative power of language because it does not have a language, only a rigid referential semantics.
So why doesn’t AI use natural languages – with their inherent linguistic semantics – to represent knowledge? The answer is well known: natural languages are ambiguous. A word can have several meanings, a meaning can be expressed by several words, sentences have multiple possible interpretations, grammar is elastic, etc. Since computers are not embodied beings imbued with common sense, as we are, they are unable to correctly disambiguate statements in natural language. For its human speakers, a natural language provides a dictionary, which is a net of predefined general categories that are mutually explanatory. This common semantic network enables the description and communication of multiple concrete situations as well as different domains of knowledge. However, because of their irregularities, AI cannot use natural languages to communicate or to teach machines directly. This is why AI remains fragmented today into micro-domains of practices and knowledge, each with its own particular semantics.
The automation of linguistic semantics could open new horizons of communication and reasoning for Artificial Intelligence. To deal with linguistic semantics, AI needs a standardized and univocal language, a code specially designed for machine use and which humans could easily understand and manipulate. This language would finally allow models to connect and knowledge to accumulate. In short, the main obstacle to the development of AI is the lack of a common computable language. This is precisely the problem solved by IEML, a metalanguage which can express meaning, like natural languages, and whose semantics are unambiguous and computable, like a mathematical language. The use of IEML will make AI less costly in terms of human labor, more adept at dealing with meaning and causality, and most importantly, capable of accumulating and exchanging knowledge.
Without language, we would have no access to enquiry, dialogue, or narrative. Language is simultaneously a medium of personal intelligence – it is difficult to think without inner dialogue – and of collective intelligence. Much of society’s knowledge has been accumulated and passed on in linguistic form. Given the role of speech in human intelligence, it is surprising that we have hoped to achieve general artificial intelligence without a computable model of language and its semantics. The good news is that we finally have one.
IEML: A Solution Based on a Semantic Code
The Information Economy Metalanguage
Many advances in computer science come from the invention of a relevant coding system which renders the coded object (number, image, sound, etc.) easily computable by a machine. For example, binary coding for numbers and pixel or vector coding for images. Therefore, we have been working on the design of a code that makes linguistic semantics computable. This artificial language, IEML (Information Economy MetaLanguage) has a regular grammar and a compact dictionary of three thousand words. More complex categories can be constructed by combining words into sentences according to a small set of grammatical rules. These complex categories can in turn be used to define others, and so on, recursively. To sum up, any type of category can be built from a small set of words.
On a linguistic level, IEML has the same expressive capacity as a natural language, and can be translated in any other language. It is also a univocal language: each word of the dictionary has only one meaning (unlike in natural languages) and a concept has only one expression, making its linguistic semantics computable. It is important to note that IEML is not a universal ontology but is indeed a language that can express any ontology, or classification.
On a mathematical level, IEML is a regular language in the sense established by Chomsky: it is an algebra. It is therefore amenable to all sorts of automatic processing and transformations.
On a computer science level, as we’ll see in more detail below, this metalanguage provides a programming language specialized for the design of knowledge graphs and data models.
The IEML Editor
The Information Economy MetaLanguage is defined by its grammar and three thousand word dictionary, which can be found on the website intlekt.io. This metalanguage comes equipped with a digital tool to facilitate its writing, reading and use: the IEML editor.
The IEML editor is used to produce and explore data models. This notion of « model » encompasses semantic networks, semantic metadata systems, ontologies, knowledge graphs, and labeling systems for categorizing training data. The editor contains a programming language to automate the creation of nodes (categories) and links (semantic relationships between categories). This programming language is declarative, which means that it does not ask the user to organize a flow of conditional instructions, but only to describe the desired results.
1. With the IEML editor, the human modeler can draft the categories that will serve as containers (or memory boxes) for different types of data. As said above, if some categories cannot be found in the 3,000 word IEML dictionary, the modeler can create more of them by combining words to make sentences, bringing a lot of refinement to the categorization.
2. From the categories, the modeler then programs the semantic relations (“is a part of”, “is a cause of”, etc.) that will connect the categorized data. Linking between nodes is automated based on the grammatical roles of the categories. The mathematical properties of relations (reflexivity, symmetry, transitivity) are then specified.
3. Once the data has been categorized, the program automatically weaves a network of semantic relations, finally giving the data even more meaning. Data mining, hypertextual exploration and visualization of relationships by tables and graphs will allow end users to explore the modeled content.
Advantages
Several fundamental features distinguish the IEML editor from contemporary data modeling tools: categories and relationships are programmable, and the resulting models are interoperable and transparent.
Categories and relationships are programmable. The regular structure of IEML allows categories to be generated and relationships to be woven functionally or automatically, instead of creating them one by one. This property saves the modeler considerable time. The time saved by automating the creation of categories and relationships more than makes up for the time spent coding categories in IEML, especially since once created, new categories and relationships can be exchanged between users.
The models are interoperable. All models are based on the same three thousand words dictionary and set grammar rules. The models are, therefore, interoperable, meaning that they can easily merge or exchange categories and sub-models. Each model is still customized to a particular context, but models can now compare, interconnect, and integrate.
The models are transparent. Although coded in IEML, models written with the IEML editor are readable in natural language. Since the categories and relations are labeled with words, or with more elaborate sentences in natural languages (and without semantic ambiguity), the models are clearer to both modelers and end-users, therefore aligning with contemporary principles of ethics and transparency.
The user does not need to be a computer scientist or be familiar with the IEML language to learn how to use it successfully; the learning curve is short. Only the grammar (simple and regular) needs to be mastered. The IEML editor could be used in schools and therefore paving the way for a democratization of data literacy.
The IEML Neuro-Semantic Architecture
Figure 1: A neuro-semantic architecture for AI
We’ll now propose an AI system architecture based on IEML. This architecture (schematized in Figure 1) is clearly a particular instance of a neuro-symbolic architecture, but it is called neuro-semantic to emphasize that it solves the problem of semantic computation and semantic interoperability between systems.
We must distinguish various types of training data (text, image, sound, etc.) from which various types of neural networks result. Sensory neural networks that have been trained by examples of data categorized in IEML will input information in the system. The data categorized by the sensory neurons is transmitted to the semantic knowledge base. If inconsistencies, errors, or biases are detected, the training data or their conceptualization must of course be revised. Hence, the system must include a dialogue loop between the data annotators who train the neural networks and the engineers who manage the knowledge base.
At the output, motor neural networks transform categorized data into data that control actions, such as text writing, image synthesis, voice output, instructions sent to effectors (robots), etc. These motor neurons are trained with examples that match the categorized data of IEML to motor data. Again, training data and neural networks must be distinguished according to their types.
The knowledge base is organized by a semantic network; therefore, it is preferably supported by a graph database. In terms of interface, this knowledge base is presented as a hypertextual encyclopedia. It also allows the programming of simulations and various dashboards for monitoring and intelligence.
The IEML editor mentioned in the previous section can also be used for tasks other than modeling. In effect, it allows varied read-write operations conditioned by the presence of semantic contents located at certain grammatical roles. When they are coded in IEML, concepts become variables of an algebra, which is obviously not the case when they are expressed in natural language. Therefore, semantic transformations can be programmed and computed. This semantic programming opens the way not only to the classical logical reasoning to which symbolic AI inference engines have accustomed us for decades, but also to other forms of automatic reasoning. Since in IEML semantics is a functional image of syntax, it becomes possible to automate analogical reasoning such as « A is to B what C is to D ». Other semantic operations can also be programmed, such as: selection and search; substitution, insertion, or deletion; extraction of relevant semantic subnetworks; summarization or expansion; inversion, allusion, attenuation, or amplification; extraction or projection of narrative structures, and so on.
Various Applications
Some applications of our IEML neuro-semantic AI architecture are evident: data integration; decision support based on causal models; knowledge management; comprehension and summarization of text; controlled generation of text (unlike GPT3 type systems where text creation is not controlled); chatbots and robotics. We will now briefly comment on two application examples: text comprehension and controlled generation of text.
Regarding controlled text generation, let’s imagine telemetry data, accounting information, medical exams, knowledge test results, etc. as input. As output we can program narrative texts in natural language synthesizing the content of the input data streams: medical diagnoses, school reports, advices, etc.
About text comprehension, let’s first assume the automatic categorization of the content of a document entered into the system. In a second step, the semantic model extracted from the text is written into the system’s memory and integrated with the knowledge that the system has already acquired. In short, Artificial Intelligence systems could accumulate knowledge from the automatic reading of documents. Assuming IEML is adopted, Artificial Intelligence systems would become capable not only of accumulating knowledge, but of integrating it into coherent models and exchanging it. This is obviously a long-term prospect that will require coordinated efforts.
Conclusion: A Humanistic Future for AI
Even if the neuro-semantic architecture proposed above does not entirely dislodge the obstacles in the path of general Artificial Intelligence, it will usher AI in the creation of applications capable of processing the meaning of texts or situations. It also allows us to envisage a market for data labeled in IEML which would stimulate the already booming development of machine learning. It would also support a collaborative public memory that would be particularly useful in the fields of scientific research, education, and health.
Today, the multiplicity of languages, classification systems, disciplinary viewpoints and practical contexts compartmentalizes our digital memory. Yet the communication of models, the critical comparison of viewpoints, and the accumulation of knowledge are essential to human symbolic cognition, a cognition that is indissolubly personal and collective. Artificial intelligence will only be able to sustainably increase human cognition if it is interoperable, cumulative, integrable, exchangeable and distributed. This means that we will not make significant progress in Artificial Intelligence without concurrently striving for a collective intelligence capable of self-reflection and of coordinating itself into a global memory. The adoption of a computable language which functions as a universal system of semantic coordinates – a language that is easy to read and write – would open new avenues for collective human intelligence, including an immersive multimedia interaction in the world of ideas. In this sense, the IEML user community could be the start of a new era of collective intelligence.
Contemporary AI, the majority of which is statistical, tends to create situations where data thinks in place of humans, unaware. In contrast, by adopting IEML, we propose to develop an AI that helps humans to take intellectual control of data in order to extract shareable meaning, in a sustainable manner. IEML allows us to rethink the purpose and operation of AI from a humanistic point of view, a point of view for which meaning, memory and personal consciousness must be treated with the utmost seriousness.
NOTES AND COMMENTED REFERENCES
On the Origins of AI The term « Artificial Intelligence » was first used in 1956 at a conference at Dartmouth College in Hanover, New Hampshire. Conference participants included computer scientist and cognitive scientist researcher, Marvin Minsky, (Turing Award 1969) and the inventor of the LISP programming language, John McCarthy (Turing Award 1971).
On Cognitive Augmentation
Cognitive augmentation (rather than imitation of human intelligence) was the primary focus of many computer science and Web pioneers. See for example:
– Bush, Vannevar. « As We May Think. » Atlantic Monthly, July 1945.
– Licklider, Joseph. « Man-Computer Symbiosis. » IRE Transactions on Human Factors in Electronics, 1, 1960, 4-11.
– Engelbart, Douglas. Augmenting Human Intellect. Technical Report. Stanford, CA: Stanford Research Institute, 1962.
– Berners-Lee, Tim. Weaving the Web. San Francisco: Harper, 1999.
On the History of Neural AI
Many people recognize Geoffrey Hinton, Yann Le Cun and Yoshua Benjio as the founders of contemporary neural AI. But neural AI began as early as the 1940s in the 20th century. A brief bibliography is provided below.
– The first theoretical paper on neural IA was published in 1943: McCulloch, Warren, and Walter Pitts. “A Logical Calculus of Ideas Immanent in Nervous Activity.” Bulletin of Mathematical Biophysics, 5, 1943: 115-133. Warren McCulloch published several papers on this themes that were collected in Embodiments of Mind. Cambridge, MA: MIT Press, 1965. I wrote a paper on his work: Lévy, Pierre. “L’Œuvre de Warren McCulloch.” Cahiers du CREA, 7, Paris, 1986, p. 211-255.
– Frank Rosenblatt is the inventor of the Perceptron, which can be considered as the first machine learning system based on a neuro-mimetic network. See his book Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms, published in 1962 by Spartan Books.
– Marvin Minsky’s 1954 Ph.D. dissertation was entitled, « Theory of neural-analog reinforcement systems and its application to the brain-model problem. » Minsky would criticize Frank Rosenblatt’s perceptron in his 1969 book Perceptrons (MIT Press) written with Seymour Papert and would later continue the symbolic AI research program. Also by Minsky, The Society of Mind (Simon and Schuster, 1986) summarizes well his approach of human cognition as emerging from the interaction of multiple cognitive modules with varied functions.
– Heinz von Foerster was the secretary of the Macy Conferences (1941-1960) on cybernetics and information theory. He was director of the Biological Computer Laboratory at the University of Illinois (1958-1975). His main articles are collected in Observing Systems: Selected Papers of Heinz von Foerster. Seaside, CA: Intersystems Publications, 1981. I studied closely the research done in this lab. See : Lévy, Pierre “Analyse de contenu des travaux du Biological Computer Laboratory (BCL).” In Cahiers du CREA, 8, Paris, 1986, p. 155-191.
– In the eighties of the XX° century, let’s notice the publication of the landmark book of McClelland, James L., David E. Rumelhart and the PDP research group. Parallel Distributed Processing: Explorations in the Microstructure of Cognition. 2 vols. Cambridge, MA: MIT Press, 1986.
– This same year 1986, Rumelhart, David E.; Hinton, Geoffrey E.; Williams, Ronald J. published an important paper: « Learning representations by back-propagating errors » in Nature 323 (6088): 533-536, 9 October 1986. Hinton was later recognized for his pioneering work with a Turing Award along with Yann LeCun and Joshua Benjio in 2018. One of their most cited common paper is: Y LeCun, Y Bengio, G Hinton “Deep learning” Nature, 521, 436–444, 2015.
The Critique of Statistical AI
Concerning the critique of statistical IA, this text resumes some of the arguments put forward by researchers like Judea Pearl, Gary Marcus and Stephen Wolfram.
– Judea Pearl received the Turing Award in 2011 for his work on causality modeling in AI. He and Dana Mackenzie wrote The Book of Why, The new science of cause and effect, Basic books, 2019.
– Gary Marcus wrote in 2018 a seminal article: « Deep learning, a critical appraisal » https://arxiv.org/pdf/1801.00631.pdf?u (Accessed on August 8, 2021). See also Gary Marcus’ book, written with Ernest Davis, Rebooting AI: Building Artificial Intelligence We Can Trust, Vintage, 2019.
– In addition to Judea Pearl’s work on the importance of causal modeling in AI, let’s remember philosopher Karl Popper’s theses on the limits of inductive reasoning and statistics. See, in particular, Karl Popper, Objective Knowledge: An Evolutionary Approach. Oxford: Clarendon Press, 1972.
– The recent report by the Center for Research on Foundation Models (CRFM) at the Stanford Institute for Human-Centered Artificial Intelligence (HAI), is entitled: On the Opportunities and Risks of Foundation Models. It begins with this sentence: « AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. » https://arxiv.org/abs/2108.07258
– Integrating existing knowledge into AI systems is one of the main goals of Stephen Wolfram’s « Wolfram Language ». See https://www.wolfram.com/language/principles/ Accessed on August 16, 2021.
– Neurosymbolic AI: The 3rd Wave, Artur d’Avila Garcez and Luis C. Lamb, December, 2020 (https://arxiv.org/pdf/2012.05876.pdf) Accessed on August 8, 2021.
– On the neuro-symbolic fusion, see also the recent report from Stanford University « 100 Year Study on AI » which identifies the neuro-symbolic hypothesis as one of the keys to advancing the discipline. https://ai100.stanford.edu/ Accessed on September 20, 2021.
On Semantic Interoperability
– All semantic metadata editors claim to be interoperable, but it is generally an interoperability of file formats, the latter being effectively ensured by Semantic Web standards (XML, RDF, OWL, etc.). But in this text, I am talking about the interoperability of semantic models themselves (we are talking here about architectures of concepts: categories and their relations). It is important to distinguish semantic interoperability from format interoperability. Models written in IEML can be exported in standard semantic metadata formats such as RDF, JSON-LD or Graph QL. On the notion of semantic interoperability, see: https://intlekt.io/2021/04/05/outline-of-a-business-model-for-a-change-in-civilization/ Accessed on January 10, 2022.
On Chomsky and Syntax
One of the first researchers to have undertaken a mathematization of languge is Noam Chomsky. See his Syntaxic Structures. The Hague and Paris: Mouton, 1957 and the paper he wrote with Marcel-Paul Schützenberger. « The Algebraic Theory of Context-Free Languages. » In Computer Programming and Formal Languages. Ed. P. Braffort and D. Hirschberg. Amsterdam: North Holland, 1963. p. 118-161. For a more philosophical approach, see Chomsky, Noam. New Horizons in the Study of Language and Mind. Cambridge, UK: Cambridge UP, 2000. To understand how IEML continues several trends of the XX° century linguistic research see my article on “The linguistic roots of IEML”: https://intlekt.io/the-linguistic-roots-of-ieml/ Accessed on January 10, 2022.
On Proper Names
I adopt here the position of Saul Kripke who is followed by most philosophers and grammarians. See, by Saul Kripke, Naming and Necessity, Oxford, Blackwell, 1980. See my recent blog entry on this subject: https://intlekt.io/proper-names-in-ieml/ Accessed on January 10, 2022.
Pierre Lévy on IEML
– « Toward a Self-referential Collective Intelligence: Some Philosophical Background of the IEML Research Program. » Computational Collective Intelligence, Semantic Web, Social Networks and Multiagent Systems, ed. Ngoc Than Nguyen, Ryszard Kowalczyk and Chen Shyi-Ming, First International Conference, ICCCI, Wroclaw, Poland, Oct. 2009, proceedings, Berlin-Heidelberg-New York: Springer, 2009, pp. 22-35.
– « The IEML Research Program: From Social Computing to Reflexive Collective Intelligence. » In Information Sciences, Special issue on Collective Intelligence, ed. Epaminondas Kapetanios and Georgia Koutrika, vol. 180, no. 1, Amsterdam: Elsevier, 2 Jan. 2010, pp. 71-94.
– The philosophical and scientific considerations that led me to the invention of IEML have been amply described in La Sphère sémantique. Computation, cognition, économie de l’information. Hermes-Lavoisier, Paris / London 2011 (400 p.). English translation: The Semantic Sphere. Computation, Cognition and Information Economy. Wiley, 2011. This book contains an extensive bibliography.
– The general principles of IEML are summarized in: https://intlekt.io/ieml/ Accessed on January 10, 2022.
– L’intelligence collective, pour une anthropologie du cyberespace, La Découverte, Paris, 1994. English translation by Robert Bonono: Collective Intelligence, Perseus Books, Cambridge MA, 1997.
– Les systèmes à base de connaissance comme médias de transmission de l’expertise » (knowledge-based systems as media for transmission/transfer of expertise), in Intellectica (Paris), special issue on « Expertise and cognitive sciences », ed. Violaine Prince. 1991. p. 187 to 219.
– I have analyzed in detail the work of knowledge engineering in several cases in my book, De la programmation considérée comme un des beaux-arts, La Découverte, Paris, 1992.
Based on the Information Economy MetaLanguage (IEML), semantic computing brings together several perspectives: an improvement of artificial intelligence, a solution to the problem of semantic interoperability, an algebraic model of semantic linguistics: all this at the service of human collective intelligence.
Art: Emma Kunz
Neuro-Symbolic Artificial Intelligence
Every animal has a nervous system. Neuronal computing, which is statistical in essence, is the common basis of all animal intelligence. Machine learning, and in particular deep learning, is a partial automation and externalization of this neural computing. By contrast, symbolic or logical computing distinguishes human intelligence from other forms of animal intelligence. Language and complex semantic representations are the most obvious manifestations of symbolic computing, which is of course supported by neural networks. After what has been labeled automatic reasoning, expert systems and semantic web, knowledge graphs are today the name of an automation and externalization of natural symbolic computing.
The point that we want to make here is that a progress in human intelligence – which is what we are looking for – does not necessarily come from an augmentation of neuronal computing power. It may be achieved by the invention and use of new symbolic systems. For example, compared to anterior irregular numbering systems like the roman one, the invention of the position numbering system with a zero improved markedly arithmetical calculations. This example suggests that, considering two identical neural networks, one of them may have a much more efficient processing than the other just because of an improved data labelling.
Semantic Interoperability
In this line of thought, it is only on the basis of an adequately coded symbolic AI that we will be able to effectively exploit our new machine learning capabilities. We are standing for a neuro-symbolic perspective on AI, but we think that an improvement of the symbolic part is needed. Symbolic AI has been invented before the Internet, when the problem of semantic interoperability did not exist. Because we now have a global memory and because our communication systems are processed by algorithms, natural languages are not anymore the right tool for knowledge metadata. Natural languages are multiple, informal, ambiguous, and changing. To make things worst, cultures, trades and disciplines divide reality in different ways. Finally, the numerous metadata systems used to classify data – often inherited from the age of print – are incompatible. The reader may object that the problem of semantic interoperability is solved by the semantic web standard RDF (Resource Description Framework) and other similar standards. It is true that current standards solve the problem of interoperability at a technical – or software – level. However *semantic* interoperability is not about files standards but about categories and architectures of concepts. To learn more on this point, see the following blogpost (a must-read).
Digital computers exist for less than a century. We still live in the prehistory of automatic computing. Today we enjoy universal coordinate systems for space and time, but no coordinate semantic system. Public health, demography and economy statistics, training and education resources, talent management, job market, the internet of things, smart cities and many other sectors rely on multiple incompatible classification systems and ontologies, inside and among themselves. To take a classical example, disaster management requires an intense emergency communication between different services, departments and organizations. But currently these institutions do not share the same metadata system, even inside the same country, the same state or the same administration.
Language Intelligence
The solution proposed by INTLEKT Metadata to the problem of semantic interoperability is not a universal ontology, not even one standard ontology by domain, which would be a drastic over-simplification and impoverishment of our collective intelligence. We want to promote interoperability and semantic computability while allowing diversity to flourish.
Our solution is rather based on a techno-scientific breakthrough: the invention of a univocal and computable semantic code called IEML (the Information Economy MetaLanguage) that has beenspecially designed to solve the problem of semantic interoperability, while improving the calculability of semantics. In one word, IEML semantics are optimally computable because they are a function of its syntax. IEML is a programmable language (akin to a computable Esperanto) able to translate any ontology or semantic metadata system and to connect all of their categories. So, if their metadata speak the same metalanguage, a great diversity of classifications and ontologies, reflecting the situations and pragmatic goals of different communities, will be able to connect and exchange concepts.
IEML has a compact dictionary (3000 words) that is organized by subject-oriented paradigms and visualized as keyboards. IEML paradigms work as symmetrical, nested and interconnected « micro-ontologies ». This feature enables the weaving of semantic relations between IEML words by the means of functions. IEML grammar is completely regular and is embedded in the IEML editor. All IEML texts are produced by the same grammatical operations on the same small dictionary. In brief, a computer only needs a dictionary and a grammar to “understand” an IEML text, which is notoriously not the case for texts in natural languages. Indeed, IEML has the expressive power of a natural language and can therefore translate any language. These properties make it an ideal pivot-language.
Collective Intelligence
IEML can not only improve inter-human communication, but also make inter-machine and human-machine communication more fluid to ensure a collective mastery of the Internet of things, intelligent cities, robots, autonomous vehicles, etc. Contemporary collective intelligence works in a stigmergic way. It is a massively distributed read-write process on our digital memories. By framing our information architecture, we structure our memory, we train our algorithms, we determine our thoughts and influence our actions. Collective intelligence requires metadata intelligence. Everyone should be able to structure her digital information in her own way and, at the same time, be able to exchange it with the utmost precision through the channels of a universal semantic postal service. IEML is the semantic metadata system adapted to the new situation, able to harness our global computing environment for the benefit of human collective intelligence.
A semantic knowledge base organized by an IEML metadata system would play simultaneously three roles:
an encyclopaedia of a practical domain: a knowledge graph, including causal models à la Judea Pearl;
a training set (or several training sets) for machine learning;
a decision support system in real time, able to save the users attention span, while telling them what they need to know.
Today, the encyclopedia part is the work of the knowledge graph people and their RDF triples. The training set and its labels is the job of the machine learning specialists. The data scientists, for their part, build business intelligence tools. These activities will converge and a new kind of semantic engineers will emerge. Furthermore, communities, institutions and companies of all sorts will be able to share their semantic knowledge in a virtual public database, or to exchange their private data on a new semantic market.
When I published « Collective Intelligence » in 1994, the WWW did not exist (you won’t find the word « web » in the book) and less than one percent of the world’s population was connected to the Internet. A fortiori, social media, blogs, Google and Wikipedia were still well hidden in the realm of possibilities and only a few visionaries had glimpsed their outlines through the mists of the future. Ancestors of social media, « virtual communities » only gathered tens of thousands of people on Earth and free software, already pushed by Richard Stallman in the early 1980s, would only really take off in the late 1990s. At that time, however, my diagnosis was already established: (1) the Internet was going to become the main infrastructure of human communication and (2) networked computers were going to increase our cognitive abilities, especially our memory. The advent of digital technology in the course of Human adventure is as important as the invention of writing or the printing press. I thought so then, and everything confirms it today. People often say to me, « you foresaw the advent of collective intelligence and well, look what happened! » No, I predicted – along with a few others – that humanity would enter into a symbiotic relationship with algorithms and data. Given this prediction, which we will acknowledge is coming true, I asked the following question: what civilization project should we embrace to best harness the algorithmic medium for the benefit of human development? And my answer was: the new medium enables us, if we decide to do so, to increase human collective intelligence… rather than continuing the heavy pattern of passivity in front of fascinating media already started with television and remaining obsessed by the pursuit of artificial intelligence.
Art: Emma Kunz « Kaleïdoscopic Vision »
A quarter of a century later
Having contextualized my 1994 « call for collective intelligence », I now turn to some observations on the developments of the last quarter century. In 2021, 65% of humanity is connected to the Internet and almost ninety percent in Europe, North America and most major cities. Lately, the pandemic has forced us to use the Internet massively to work, learn, buy, communicate, etc. Scholars around the world share their databases. We consult Wikipedia every day, which is a classic example of a philanthropic collective intelligence enterprise supported by digital technology. Programmers share their code on GitHub and help each other on Stack Overflow. Without always clearly realizing it, everyone becomes an author on his or her blog, a librarian when tagging, labelling or categorizing content, a curator when gathering resources, an influencer on social media and online shopping platforms, and even an unwilling artificial intelligence trainer as our slightest online actions are taken into account by learning machines. Distributed multi-player games, crowdsourcing, data journalism and citizen journalism have become part of everyday life.
Ethologists who study social animals define stigmergic communication as an indirect coordination between agents via a common environment. For example, ants communicate primarily by leaving pheromone trails on the ground, and this is how they signal each other about paths to food. The emergence of a global stigmergic communication through digital memory is probably the biggest social change of the last twenty-five years. We like, we post, we buy, we tag, we subscribe, we watch, we listen, we read and so on… With each of these acts we transform the content and the system of internal relations in the digital memory, we train algorithms, and we modify the data landscape in which other Internet users evolve. This new form of communication by distributed reading-writing in a collective digital memory represents an anthropological mutation of great magnitude that is generally little or poorly perceived. I will come back later on to this revolution by reflecting on how to use it as a support point to increase collective intelligence.
But first I need to talk about a second major transformation, linked to the first, a political mutation that I did not foresee in 1994: the emergence of the platform state. By this expression I do not mean the use of digital platforms by governments, but the emergence of a new form of political power, which is the successor of the nation-state without suppressing it. The new power is exercised by the owners of the main data centers, who in fact control the world’s memory. One will have recognized the famous Sino-American oligarchy of Google, Apple, Facebook, Amazon, Microsoft, Baidu, Alibaba, Tencent and others. Not only are these companies the richest in the world, having long since surpassed the old industrial flagships in market capitalization, but they also exercise classic regal powers: investment in crypto-currencies that escape central banks; control and surveillance of markets; authentication of personal identities; infiltration of education systems; mapping from street to satellite view; cadastral records; management of public health (wristbands or other wearable devices, recording of conversations at doctors’ offices, and epidemiological memory in the cloud), crisscrossing the skies with networks of satellites. But above all, the data overlords have taken control of public opinion and legitimate speech: influence, surveillance, censorship… Be careful what you say, because you risk being deplatformed! Finally, the new political apparatus is all the more powerful as it relies on psychological mechanisms close to addiction. The more users become addicted to the narcissistic pleasure or excitement provided by social media and other attention-grabbers, the more data they produce and the more they feed the wealth and power of the new oligarchy.
Confronted with these new forms of enslavement, is it possible to develop an emancipatory strategy adapted to the digital age? Yes, but without harboring the illusion that we can end the dark side once and for all through some radical transformation. As Albert Camus says at the end of his 1942 essay The Myth of Sisyphus, « The struggle itself towards the heights is enough to fill a man’s heart. One must imagine Sisyphus happy. » Increasing collective intelligence is a task always to be taken up and deepened. Maximizing creative freedom and collaborative efficiency simultaneously is a matter of performance in context and does not depend on a definitive technical or political solution. However, when the cultural and technical conditions I will now evoke are met, the task will be easier and efforts will be able to converge.
The dialectic of man and machine
Communication between humans is increasingly done through machines via a distributed read-write process in a common digital memory. Two poles interact here: machines and humans. Machines are obviously deterministic, whether this determinism is logical (ordinary algorithms, the rules of so-called symbolic AI) or statistical (machine learning, neural AI). Machine learning algorithms can evolve with the streams of data that feed them, but this does not make them escape determinism. As for humans, their behaviour is only partially determined and predictable. We are conscious social animals that play multiple games, that are permeated by all emotions, that show autonomy, imagination and creativity. The great human conversation expresses an infinite number of nuances in a fundamentally open-ended process of interpretation. Of course, it is humans who produce and use machines, therefore they fully belong to the world of culture. But it remains that – from an ethical, legal or existential point of view – humans are not deterministic logico-statistical machines. On the one hand the freedom of meaning, on the other the mechanical necessity. However, it is a fact today that humans keep memories, interpret and communicate through machines. Under these conditions, the human-machine interface is barely distinguishable from the human-human interface. This new situation provokes a host of problems of which out-of-context interpretations, nuance-free translations, crude classifications and communication difficulties are only the most visible symptoms, whereas the deep-rooted evil lies in a lack of autonomy, in the absence of control over the techno-cosmos on a personal or collective scale.
Let’s now think about our interface problem. The best medium of communication between humans remains language, with the panoply of symbolic systems that surround it, including music, body expression and images. It is therefore language, oral or written, that must play the main role in the human-machine interface. But not just any language: an idiom that is adequate to the multitude of social games, to the complexity of emotions, to the expressive nuance and interpretative openness of the human side. However, on the machine side, this language must be able to support logical rules, arithmetic calculations and statistical algorithms. This is why I have spent the last twenty years designing a human language that is also a computer language: IEML. The noolithic biface turns on one side to the generosity of meaning and on the other to mathematical rigor. Such a tool will give us a grip on our technical environment (programming and controlling machines) as easily as we communicate with our fellow men. In the opposite direction, it will synthesize the streams of data that concern us into explanatory diagrams, understandable paragraphs, and even empathetic multimedia messages. Moreover, this new techno-cognitive layer will allow us to go beyond the opaque stigmergic communication that we maintain through the digital memory to reach reflexive collective intelligence.
Towards a reflexivecollective intelligence
Images, sounds, smells and places mark out human memory, as well as that of animals. But it is language that unifies, orders and reinterprets our symbolic memory at will. This is true not only on an individual scale, but also on a collective scale, through the transmission of stories and writing. Through language, humanity has gained access to reflexive intelligence. By analogy, I think that we will only reach reflexive collective intelligence by adopting a language adequate to the organization of the digital memory.
Let’s look at the conditions needed for this new form of large-scale critical thinking to take place. Since social cognition must be able to observe itself, we must model complex human systems and make these models easily navigable. Like dynamic human communities, these digital representations will be fed by heterogeneous data sources organized by disparate logics. On the other hand, we need to make conversational thought processes readable, where forms emerge, evolve and hybridize, as in the reality of our ecosystems of ideas. Moreover, since we want to optimize our decisions and coordinate our actions, we need to account for causal relationships, possibly circular and intertwined. Finally, our models must be comparable, interoperable and shareable, otherwise the images they send back to us would have no objectivity. We must therefore accomplish in the semantic dimension what has already been done for space, time and various units of measurement: establish a universal and regular coordinate system that promotes formal modeling. Only when these conditions are met will digital memory be able to serve as a mirror and multiplier for the collective human intelligence. The recording and computing power of gigantic data centers now makes this ideal attainable, and the semantic coordinate system (IEML) is already available.
In the maze of memory
Until the invention of movable type printing by Gutenberg, one of the most important parts of rhetoric was the art of memory. It was a mnemonic method known as « places and pictures ». The aspiring orator had to practice the mental representation of a large architectural space – real or imaginary – such as a palace or a temple, or even a city square, where several buildings would be arranged. The ideas to be memorized were to be represented by images placed in the places of the palatial architecture. Thus, the windows, niches, rooms and colonnades of the palace (the « places ») were populated with human figures bearing emotionally and visually striking characters, in order to be better remembered (the « images »). The semantic relations between ideas were better remembered and used if they were represented by local relations between images.
From the 16th century in the West, the fear of forgetting gave way to the anxiety of being drowned in the mass of printed information. The art of memory, adapted to the oral and manuscript eras, was followed by the art of organizing libraries. One discovers then that the conservation of information is not enough, it is necessary to classify it and to place it (both things go together before the digital era) so that one finds easily what one seeks. The plan of the imaginary palace is followed by the layout of the shelves of the library. The distinction between data (books, newspapers, maps, archives of all kinds) and metadata was established in the early 18th century. A library’s metadata system essentially consists of a catalog of stored documents and cardboard cards filed in drawers that give, for each item: its author, title, publisher, date of publication, subject, etc. Not to mention the index number that indicates the precise location of the document. This splitting of data and metadata is initially done on a library-by-library basis, each with its own organizational system and local vocabulary. A new level of abstraction and generalization was reached in the 19th century with the advent of classification systems with a universal vocation, which were applied to numerous libraries, of which Dewey’s « decimal » system is the best known example.
With the digital transformation that began at the end of the 20th century, the distinction between data and metadata has not disappeared, but it is no longer deployed in a physical space on a human scale.
In an ordinary database, each field corresponds to a general category (the name of the field is a kind of metadata, such as « address ») while the value of the field « 8, Little Bridge street » corresponds to a piece of data. The metadata system is none other than the conceptual scheme that governs the database structuring. (In Excel tables, the columns correspond to the metadata and the content of the cells to the data). Each database obviously has its own schema, adapted to the needs of its user. Traditional databases were designed before the Internet, when computers rarely communicated.
Everything changes with the massive adoption of the Web from the end of the 20th century. In a sense, the Web is a large distributed virtual database, each item of which has an address or an “index number »: the URL (Uniform Resource Locator), which begins with http:// (Hypertext Transfer Protocol). Here again, metadata are integrated into the data, for example in the form of tags. As the Web potentially makes all memories communicate, the disparity of local metadata systems or improvised folksonomies (such as hashtags used in social media) becomes particularly glaring.
But the abstraction and unbundling of metadata experienced by libraries in the 19th century is reinvented in the digital world. The same conceptual model can be used to structure different data while promoting communication between distinct repositories. Models such as schema.org, supported by Google, or CIDOC-CRM, developed by cultural heritage conservation institutions, are good examples. The notion of semantic metadata, developed in the symbolic artificial intelligence community of the 1970s, was popularized by the Semantic Web project launched by Tim Berners-Lee in the wake of the Web’s success. This is not the place to explain the relative failure of the latter project. Let’s just point out that the rigid constraints imposed by the standard formats of the World Wide Web Consortium have discouraged its potential users. The notion of ontology is now giving way to that of Knowledge Graph, in which digital resources are accessed by means of a data model and a controlled vocabulary. In this last stage of the evolution of memory, data is no longer contained in the fixed tabular schemas of relational databases, but in the new graph databases, which are more flexible, easier to evolve, better able to represent complex models and allowing several « views ». A knowledge graph lends itself to automatic reasoning (classical symbolic AI), but also to machine learning if it is well-designed and if the data is sufficiently large.
Today, a large part of the digital memory is still in relational databases, without clear distinction between data and metadata, organized according to mutually incompatible rigid schemas, poorly optimized for the needs of heir users: knowledge management, coordination or decision support. Gathered in common repositories (datalakes, datawarehouses), these datasets are sometimes catalogued according to disparate systems of categories, leading to incoherent representations or simulations. The situation in the still minority world of knowledge graphs is certainly brighter. But many problems remain: it is still very difficult to make different languages, business domains or disciplines communicate. While the visualization of space (projection on maps), time (timelines) and quantities has become commonplace, the visualization of complex qualitative structures (such as speech acts) remains a challenge, especially since these qualitative structures are essential for the causal understanding of complex human systems.
We need to make an extra effort to transform the digital memory into a support for reflexive collective intelligence. A collective intelligence allowing all points of view, but well coordinated. To do this, we need to reproduce at an additional height the gesture of abstraction that led to the birth of metadata at the end of the 18th century. So let us split metadata into (a) models and (b) metalanguage. Models can be as numerous and rich as one wants, they will communicate – with natural languages, humans and machines – through a common metalanguage. In the library of Babel, it is time to turn on the light.
Conventional banks now offer their customers an application that enables them to carry out transactions on a smartphone. But the data from the smartphone is most often decoded and sent into the central system overnight, then processed and finally re-encoded the next day to be sent back to the smartphone application. Bottom line: it takes two days for your account to be updated after a transaction on your phone.
In contrast, with 21st century banking systems, all processing takes place in the same data center accessible via the Internet. In addition, the mobile and central applications communicate immediately because they use the same data format and categorization. As a result, accounts are updated instantly after a transaction on the smartphone. The information systems of the new banks are said to be data-centric from the start. By making the flow of information more fluid, the central challenge of the data-centric organization is to improve the customer experience or, to use another formulation, to create more value for the beneficiary of a service.
Art: Emma Kunz
The value chain
The notion of value covers a wide semantic field. It can refer to ethical values, such as justice, courage, wisdom, or the harmony of human relations. Such values obviously have no monetary counterparts. As for the goods and services that are exchanged on the market, beyond a temporary and local adjustment of supply and demand, it is very difficult to assign an essence to their value. Economic value may correspond to a necessity (such as eating), to a desire for entertainment or beauty, to the acceleration of a boring job, to a prospect of making money (lottery or speculation instrument), to an improvement in the quality of life, to the acquisition of skills, to a broader understanding that will enable one to better decide, to a competitive advantage, to a more flattering image, etc. Value is therefore not a simple function of the work invested in the production of a good or service. It depends on the subjective appreciation and comparisons of those who benefit from it, all taking place in a changing economic and cultural context. Despite the evanescent nature of its essence, which is probably due to its relationship to desire, value is at the heart of economic theory and business practice. Every organization creates value for its customers (a private company), its public (a municipal service) or its patients (a hospital) and this creation is the main justification for its existence.
It is often useful to distinguish between two distinct people, the customer, who pays for the good or service, and the consumer, who uses it. For example, a company’s IT department (the customer) buys software, but it is the employees (the consumers) who use it. In the following analysis, I will focus on the relationship between the producers and consumers of value. Each collaborator creates value for the colleagues who come after him on the chain, the good execution of their work depending on his. The value chain does not necessarily stop at the borders of a single organization. It can connect networks of companies, which may themselves be located in several countries, with each type of company contributing to the design, production of parts, assembly, transportation and sale of the product. Supply chains, which have been talked about so much since the COVID-19 pandemic, are a case of a value chain that focuses on the material activities and transportation within a particular industry. The final consumer benefits from the value created at each stage of the production of the good or service.
Increasing the productivity of organizations and industries is the basis of economic prosperity. This increase comes from innovations that create more value at lower cost. But the overall performance of a company – or of a larger value chain – depends on the performance of each activity or trade, but also on the link between these activities. This is where we come back to the theme of the data-centric organization, because activities – and even more so the links between activities – require the reception, processing and exchange of information.
From application-centric computing to data-centric computing
In the second half of the 20th century, during the first wave of computerization, each « trade » of a company had developed applications to increase its performance: the computer-aided design system, the production robots, the inventory management, the employee payroll, the company accounting, the customer database, etc. Each particular application was designed according to the cultural norms and vocabulary of its environment. Input data was formatted specifically for the application that used it, while output data was formatted for the needs of its immediate users. The result was an application-centric « siloed » computerization, with each application controlling the structure of its input and output data.
The traditional bank at the beginning of this text is a good example of this 20th century computing, whose main defect is the difficulty of communication between applications. Indeed, the conceptual breakdown and formatting of the output data of one application do not necessarily correspond to those of the input data of another application. For example, if the inventory management software does not share its data with the customer database software, it is difficult to respond quickly to an incoming order. But since the beginning of the 21st century, Internet connections have become more and more commonplace. On the hardware side, information processing is increasingly taking place in the large data centers of Amazon or Microsoft, which rent memory to their customers as easily as parking lots and computing power on demand as if it were electricity. Memory and computing power are becoming commodities that you don’t have to produce yourself. This is called cloud computing. On the software side, APIs (application programming interfaces) are data encoding/decoding interfaces that allow applications to exchange information. As a result of the changes mentioned above, application-centric computing is becoming increasingly obsolete, although it is still the de facto situation in most organizations in 2021.
In contrast to the 20th century, 21st century computing is data-centric. We need to imagine a common warehouse where different applications come to get their input data and deposit their output data. Instead of specialized data being ordered around particular applications, multiple applications, some of which are ephemeral, are ordered around a common and relatively stable digital memory. We say then that applications become interoperable. The take-off of data-centric computing can be dated back to 2002, when Jeff Bezos, the head of Amazon, asked all his developers to make their data available through an API.
From an economic point of view, data-centric computing improves the productivity of organizations because it allows different activities to share their data and coordinate more easily: the value chain becomes more fluid. Contrary to those administrations displaying indecipherable forms in bureaucratic jargon and asking users ten times to give the same information in different versions because their applications don’t communicate, large cloud companies (like Big Techs and BATX) have accustomed clients to immediate reaction times and optimized interfaces. The richest companies in the world are data-centric. So are dynamic sectors of the economy, such as the video game industry or the online distribution of movies and series. Since the benefits of data-centric computing are so obvious, why isn’t it implemented everywhere? Because there can be no data-centric computing outside a data-centric organization, and the transition to this new type of organization requires a considerable epistemological and social change. The major cloud companies date from the 21st century or the very end of the 20th century. They were born in the digital paradigm, and it is they who invented the data-centric organization. Older industries, on the other hand, are struggling to keep up.
To any activity (production, sales, etc.) corresponds a practical culture, a certain way of cutting up objects, naming their relations and sequencing operations. The computerization of an activity implies not only the creation of an application but also of a metadata system, and both are conditioned by a dated and situated practical culture. Merging an organization’s data collections requires « reconciling » the different metadata systems and, once this is done, committing to maintaining and evolving the common metadata system to accompany the needs. All this requires many discussions with experts from different spheres of activity and harmonization meetings, where bargaining over concept definitions can be tough. The reconciliation of data models is no less complex than any intercultural negotiation weighed down by power issues. Indeed, for most of the actors involved, it is not only necessary to revise their cognitive habits and ways of doing things, but also to give up a part of their local sovereignty. It will no longer be possible to organize one’s practical memory without coordinating with the other activities in the value chain, both on a semantic and technical level. From now on, data governance, for which the main person in charge is the « Chief Information Officer » or « Chief Data Officer », becomes one of the main functions of the company.
Data governance
Data governance faces two intertwined problems: semantics and politics. On a political level, it should be noted that metadata systems – that is, the categories that organize data – are always linked to the social, cultural and practical characteristics of their users. For example, in a large telecommunication company, consumption data will be organized by « lines » and not by « customers ». A customer may have several lines and the same line may be used by several customers. It is clear that customer relations would be easier if the data were classified and analyzed according to the physical or legal persons who use the company’s services. But this is not the case because the telecom company is dominated by a culture of engineers for whom the « real » data are those of the lines. This hardware-based rather than human-based approach also makes pricing as « objective » as possible and removes it from negotiation. In short, the way an institution organizes its memory reflects and reifies its identity. To reorganize its memory is to change its identity. The parallelism between metadata and social contexts makes data governance a political issue.
As for the semantic issue, it no longer concerns the subjective side of identity – whether personal or collective – but its logical side. If we want applications to be interoperable from one end of the value chain to the other, objects, relationships and processes must be named in a unique way. The difficulty here comes from the multiplicity of businesses, each with its own jargon, and the plurality of languages, particularly in international companies or sectors. When it comes to coordinating activities, synonyms (different words for the same thing) and homonyms (one word meaning several different things) become obstacles to collaboration. Homonyms, in particular, can cause serious miscalculations. For example, it happened in an airline company that the word « Asia » covered different geographical areas depending on the branch and that this semantic inconsistency caused strategic decision errors. When all operations are automated and driven by data, an ambiguous term can give false indications to managers, or even disrupt a supply chain.
The « data dictionary » or catalog is the primary tool for data governance. It is where all the data types are listed and the unique way of categorizing them. If, as is often the case, the catalog has not been unified, then « alignment tables » must be used between systems. Beyond the problems of consistency, data governance must also deal with the quality of the data. For this purpose, a « data control catalog » is used, which lists the methods for testing the quality of the data according to its nature. For example, how do you detect errors in customer names when the company operates in seventy countries? There are countries where we do not split into first and last names, other countries where numbers are acceptable in a name (in Ukraine), others where a name can have four or five consonants in a row, etc.
The transition to a data-centric organization implies a change of culture and an evolution of management. All of a sudden, words and concepts become important, not only in communication and marketing, but also in production, which is no less digitized than the other functions of the company. In addition, the cultural change calls for more openness and communication between departments, branches, services and businesses. No good management without data management, and no data management without good metadata management. We thought that interest in semantics was reserved for cultural studies departments in American universities, but now it is a condition of business productivity!
Distinguish between words and concepts
Finally, I note that the most sophisticated metadata editing and management tools on the market (Pool Party, Ab Initio, Synaptica) have no way of clearly distinguishing between « words » or « terms » in a particular natural language and « concepts » or « categories », which are more abstract and cross-linguistic notions. The same concept can be expressed by different words in different languages and the same word can correspond to several concepts, even in the same language (is the « mole » an animal, a spot on the skin, an infiltrated spy, the Avogadro’s number…?). Words are ambiguous and multiple, but recognizable by humans. The underlying formal concepts are unique and should be interpretable by machines. By proposing a unique encoding system for concepts and their relations that is independent of natural languages, IEML allows words and concepts to be distinct and articulated. This new encoding system not only advances semantics, but also has an unsuspected power to make value chains more fluid and increase collective intelligence.
P.S. I would like to thank John Horodyski, Paul-Louis Moreau, Samuel Parfouru and Michel Volle for answering my questions, thus helping to inform this post. Errors, inaccuracies, and heterodox opinions should nevertheless be attributed only to the author, Pierre Lévy.
Or how to move from a metadata language to a culture of collective intelligence…
THE METADATA ISSUE
Metadata are the data that organize the data. Data are like books in a library and metadata are like the library card index and catalog: their function is to identify the books in order to store and find them better. Metadata are less about describing things exhaustively (it is not about making maps at the same scale as the territory…) than about providing reference points from which users can find what they are looking for, with the help of algorithms. All information systems and software applications organize information through metadata.
We can distinguish between…
material metadata, such as a file’s format, creation date, author, license, etc.
semantic metadata which deals with the content of a document or a set of data (what it is about) as well as its practical dimension (what the data is used for, by whom, in what circumstances, etc.).
Art: Emma Kunz
The main focus here is on semantic metadata. A semantic metadata system can be as simple as a vocabulary. At a higher level of complexity it can be a hierarchical classification or taxonomy. At the most complex level, it is an « ontology », i.e. the modelling of a domain of knowledge or practice, which may contain several taxonomies with transverse relationships, including causal relationships and automatic reasoning capabilities.
Semantic metadata are an essential part of artificial intelligence devices:
they are used as skeletons for knowledge graphs – or knowledge bases – implemented by big techs (Google, Facebook, Amazon, Microsoft, Apple…) and more and more in large and medium-sized companies,
they are used – under the name of « labels » – to categorize the training datasets for deep learning models.
Because they structure contemporary knowledge, whose medium is digital, metadata systems represent a considerable stake at the scientific, cultural, political levels…
One of the goals of my company INTLEKT Metadata Inc. is to establish IEML (Information Economy MetaLanguage) as a standard for the expression of semantic metadata systems. What is the contemporary landscape in this area?
THE SEMANTIC METADATA LANDSCAPE TODAY
Standard Formats
The system of standard formats and « languages » proposed by the World Wide Web Consortium – W3C – (XML, RDF, OWL, SPARQL) to achieve the « Semantic Web » has been around since the late 20th century. It has not really caught on, and especially not in companies in general and big tech in particular, which use less cumbersome and less complex formats, such as « property graphs« . Moreover, manual or semi-manual categorization of data is often replaced by statistical approaches for automated indexing (NLP, deep learning…), which bypass the need to design metadata systems. The W3C system of standards deals with the *filesformats and programs* handling semantic metadata but *not the semantics itself*, i.e. the categories, concepts, properties, events, relations, etc. that are always expressed in natural languages, with all the ambiguities, multiplicities and incompatibilities this implies.
Standard models
On top of this system of standard formats, there are standard models to deal with the actual semantic content of concepts and their relationships. For example schema.org for web sites, CIDOC-CRM for the cultural domain, etc. There are standard models for many domains, from finance to medicine. The problem is, there are often several competing models for a domain and the models themselves are hypercomplex, to the point that even the specialists of a model master only a small part of it. Again, these models are expressed in natural languages, with the problems that this implies… and most often in English only.
Specific metadata systems
Taxonomies, ontologies and other metadata systems implemented in real applications for organizing data sets are mostly partial uses of standard models and standard formats. Users submit – to varying degrees of success – to these layers of standards in the hope that their data and applications will become the happy subjects of a realm of semantic interoperability. But their hopes are disappointed. The ideal of the decentralized intelligent Web of the late 1990s has given way to search engine optimization (SEO) more or less aligned with Google’s (secret!) knowledge graph. We have to admit, almost a quarter of a century after its launch, that the W3C’s Semantic Web has not kept its promises.
Problems encountered
To achieve semantic interoperability, i.e. fluid communication between knowledge bases, information system managers submit to rigid models and formats. But because of the multitude of formats, models and their disparate applications, not to mention language differences, they do not achieve the expected gain. Moreover, producing a good metadata system is expensive, because it requires a multidisciplinary team including: a project manager, one or more specialists in the domain of use, a specialist in formal modelling of the type of taxonomy or ontology (cognitive engineering) who is able to find his way through the labyrinth of standard models, and finally a computer engineer specializing in semantic metadata formats. Some people combine several of these skills, but they are rare. Finally, a recent survey shows that W3C « linked data » tools (including RDF, OWL and SPARQL) are too complex for web developers and end-users.
HOW CAN IEML SOLVE PROBLEMS IN THE WORLD OF SEMANTIC METADATA?
IEML in a nutshell
IEML – patented by INTLEKT Metadata – is neither a taxonomy, nor a universal ontology, nor a model, nor a format: it is a *language* or *meta-ontology* composed of (1) a few thousand semantic primitives organized in paradigms and (2) a fully regular grammar.
Unique features of the IEML language
IEML is « agnostic » with respect to formats, natural languages and hierarchical relationships between concepts. IEML allows to build and share any concept, concept hierarchy or relationship between concepts. Therefore IEML does not produce a flattening of expressive possibilities. However, IEML does provide semantic interoperability, i.e. the ability to merge, exchange, recombine, connect and translate almost automatically metadata systems and the knowledge bases organized by these metadata. IEML thus reconciles maximum originality, complexity or cognitive simplicity on the one hand and interoperability or communication on the other, contrary to the contemporary situation where interoperability is « paid for » with a restriction of expressive possibilities.
Unique features of the IEML editor
Another advantage: unlike the main contemporary metadata editing tools (Smart Logic Semaphore, Pool Party, Synaptica, Top Braid Composer) the IEML editor designed by INTLEKT will be intuitive (visual interface based on tables and graphs) and collaborative. It is not designed for specialists in RDF and OWL (the standard formats), like the editors mentioned above, but for application domain experts. A method accompanying the tool will help specialists to formalize their domains in IEML. The software will automatically import and export the metadata in the standard formats chosen by the user. Thus, the IEML editor will reduce the complexity and cost of creating semantic metadata systems.
Market for metadata management and edition tools
It is easy to see that, as the amount of data produced continues to grow, along with the urge to extract usable knowledge from it, there is an increasing need to create and maintain good metadata systems. The market for semantic metadata system editing and management tools is now worth $2 billion and could reach (by a very conservative estimate) $16 billion by 2026. This projection aggregates:
data from the semantic industry itself (companies that create metadata systems for their customers),
semantic annotation tools for training datasets for machine learning used in particular by data scientists,
management of their internal metadata systems by the big tech.
INTLEKT GOALS FOR THE NEXT 5-10 YEARS
The Foundation
We want IEML to become an open-source standard for semantic metadata around 2025. The IEML standard should be supported, maintained and developed by a non-profit foundation. This foundation will moderate a community dedicated to the collaborative edition of IEML metadata systems and provide a public knowledge base of IEML-categorized data. The foundation will create a socio-technical ecosystem conducive to the growth of collective intelligence.
The private company
INTLEKT will continue to maintain the collaborative editing tool and to design custom semantic knowledge bases for solvable clients. We will also implement a marketplace – or exchange system – for IEML-indexed private data that will be based on the blockchain. The IEML-indexed knowledge bases will be interoperable on the parallel planes of data analysis, automatic reasoning, and neural models training.
However, before reaching this point, INTLEKT must demonstrate the effectiveness of IEML through several real-world use cases.
INTLEKT’S MARKET IN THE 2-5 YEAR HORIZON
Interviews with numerous potential customers have enabled us to define our market for the coming years. Let’s define the relevant areas by elimination and successive approximations.
Human affairs
IEML is not relevant for modeling purely mathematical, physical or biological objects. The exact sciences already have formal languages and recognized classifications. On the other hand, IEML is relevant for objects from the humanities and social sciences or for interactions between objects from the exact sciences and objects from the humanities, such as technology, health, the environment or urban phenomena.
Non-standard fields
For the time being, we will not exhaust ourselves in translating all existing metadata models into IEML: they are very numerous, sometimes contradictory, and rarely used in full. Many users of these models are content to select a small, useful subpart of them and will not invest their time and money in a new technology without necessity. For example, many SEO (Search Engine Optimization) companies extract a useful subset from schema.org‘s classes (sponsored by Google) and Wikidata‘s entities (because they are trusted by Google) and have no need for additional semantic technologies. Other examples: the gallery, museum, library or archive sectors have to submit to rigid professional standards with limited possibilities of innovation. In short, sectors that are content to use an existing standard model are not part of our short-term market. We will not fight losing battles. In the long term, however, we envision a collaborative platform where the voluntary translation of current standard models into IEML can take place.
Let’s also eliminate the e-commerce market for the moment. This sector does use category systems to identify broad domains (real estate, cars, appliances, toys, books, etc.), but the multitude of goods and services within these broad categories is captured by automatic natural language processing or machine learning systems, rather than by refined metadata systems. We do not believe that IEML will be adopted in the near term in online commerce.
This leaves non-standard domains – which do not have ready-made models – or multi-standard domains – which must build hybrid models or crossroads – and for which statistical approaches are useful… but not sufficient. Think for example of collaborative learning, public health, smart cities, e-democracy, smart contracts documentation, massively multiplayer online games, analysis of complex corpora from several disciplines in digital humanities, etc.
Modelling and visualization of complex systems
Within the non-standard domains, we have identified the following needs that are not met by the semantic technologies in use today:
The modelling of complex human systems, where several heterogeneous « logics » meet, i.e. groups obeying various types of rules. This includes data produced by processes of deliberation, argumentation, negotiation and techno-social interaction.
The modelling of causal systems, including circular and intertwined causalities.
The modelling of dynamic systems during which the objects or actants transform. These dynamics can be of various types: evolution, ontogeny, successive hybridizations, etc.
The interactive 2D or 3D visualization and exploration of semantic structures in huge corpora, preferably in a memorable form, i.e. easy to remember.
In the coming years, INTLEKT intends to model complex dynamic systems involving human participation in a causal manner and to provide access to a memorable sensory-motor exploration of these systems.
IEML being a language, everything that can be defined, described and explained in natural language can be modelled in a formal way in IEML, thus providing a qualitative framework for quantitative measurements and calculations. It will be possible to perform automatic reasoning from rules, prediction and decision support, but the main contribution of IEML will be an increased capacity for analysis, synthesis, mutual understanding and coordination in action of user communities. Semantic interoperability should be the by-product of a cognitive gain, not a costly obligation.
THE NEXT SIX MONTHS
The IEML language already exists. Its development has been funded with one million dollars in an academic setting. We also have a prototype of the editor. We now need to move to a professional version of the editor in order to meet the market needs identified in the previous section. For this we are seeking a « seed » private investment of about 250 K US$, which will be used mainly for the development of a collaborative editing platform with the appropriate interface. Welcome to investors.
Based on the Information Economy MetaLanguage (IEML), semantic computing brings together several perspectives: an improvement of artificial intelligence, a solution to the problem of semantic interoperability, an algebraic model of semantic linguistics: all this at the service of human collective intelligence.
Art: Emma Kunz
Neuro-Symbolic Artificial Intelligence
Every animal has a nervous system. Neuronal computing, which is statistical in essence, is the common basis of all animal intelligence. Machine learning, and in particular deep learning, is a partial automation and externalization of this neural computing. By contrast, symbolic or logical computing distinguishes human intelligence from other forms of animal intelligence. Language and complex semantic representations are the most obvious manifestations of symbolic computing, which is of course supported by neural networks. After what has been labeled automatic reasoning, expert systems and semantic web, knowledge graphs are today the name of an automation and externalization of natural symbolic computing.
The point that I want to make here is that a progress in human intelligence – which is what we are looking for – does not necessarily come from an augmentation of neuronal computing power. It may be achieved by the invention and use of new symbolic systems. For example, compared to anterior irregular numbering systems like the roman one, the invention of the position numbering system with a zero improved markedly arithmetical calculations. This example suggests that, considering two identical neural networks, one of them may have a much more efficient processing than the other just because of an improved data labelling.
Semantic Interoperability
In this line of thought, it is only on the basis of an adequately coded symbolic AI that we will be able to effectively exploit our new machine learning capabilities. I am standing for a neuro-symbolic perspective on AI, but I think that we need an improvement of the symbolic part. Symbolic AI has been invented before the Internet, when the problem of semantic interoperability did not exist. Because we now have a global memory and because our communication systems are processed by algorithms, natural languages are not anymore the right tool for knowledge metadata. Natural languages are multiple, informal, ambiguous, and changing. To make things worst, cultures, trades and disciplines divide reality in different ways. Finally, the numerous metadata systems used to classify data – often inherited from the age of print – are incompatible. The reader may object that the problem of semantic interoperability is solved by the semantic web standard RDF (Resource Description Framework) and other similar standards. It is true that current standards solve the problem of interoperability at a technical – or software – level. However *semantic* interoperability is not about files standards but about categories and architectures of concepts.
Digital computers exist for less than a century. We still live in the prehistory of automatic computing. Today we enjoy universal coordinate systems for space and time, but no coordinate semantic system. Public health, demography and economy statistics, training and education resources, talent management, job market, the internet of things, smart cities and many other sectors rely on multiple incompatible classification systems and ontologies, inside and among themselves. To take a classical example, disaster management requires an intense emergency communication between different services, departments and organizations. But currently these institutions do not share the same metadata system, even inside the same country, the same state or the same administration.
Language Intelligence
The solution proposed by INTLEKT Metadata to the problem of semantic interoperability is not a universal ontology, not even one standard ontology by domain, which would be a drastic over-simplification and impoverishment of our collective intelligence. We want to promote interoperability and semantic computability while allowing diversity to flourish.
Our solution is rather based on a techno-scientific breakthrough: the invention of a univocal and computable semantic code called IEML (the Information Economy MetaLanguage) that has beenspecially designed to solve the problem of semantic interoperability, while improving the calculability of semantics. In one word, IEML semantics are optimally computable because they are a function of its syntax. IEML is a programmable language (akin to a computable Esperanto) able to translate any ontology or semantic metadata system and to connect all of their categories. So, if their metadata speak the same metalanguage, a great diversity of classifications and ontologies, reflecting the situations and pragmatic goals of different communities, will be able to connect and exchange concepts.
IEML has a compact dictionary (less than 3500 words) that is organized by subject-oriented paradigms and visualized as keyboards. IEML paradigms work as symmetrical, nested and interconnected « micro-ontologies ». This feature enables the weaving of semantic relations between IEML words by the means of functions. IEML grammar is completely regular and is embedded in the IEML editor. All IEML texts are produced by the same grammatical operations on the same small dictionary. In brief, a computer only needs a dictionary and a grammar to “understand” an IEML text, which is notoriously not the case for texts in natural languages. Indeed, IEML has the expressive power of a natural language and can therefore translate any language, which makes it an ideal pivot-language.
Collective Intelligence
IEML can not only improve inter-human communication, but also make inter-machine and human-machine communication more fluid to ensure a collective mastery of the Internet of things, intelligent cities, robots, autonomous vehicles, etc. Contemporary collective intelligence works in a stigmergic way. It is a massively distributed read-write process on our digital memories. By framing our information architecture, we structure our memory, we train our algorithms, we determine our thoughts and influence our actions. Collective intelligence requires metadata intelligence. Everyone should be able to structure her digital information in her own way and, at the same time, be able to exchange it with the utmost precision through the channels of a universal semantic postal service. IEML is the semantic metadata system adapted to the new situation, able to harness our global computing environment for the benefit of human collective intelligence.
A semantic knowledge base organized by an IEML metadata system would play simultaneously three roles:
an encyclopaedia of a practical domain: a knowledge graph, including causal models à la Judea Pearl;
a training set (or several training sets) for machine learning;
a decision support system in real time, able to save the users attention span, while telling them what they need to know.
Today, the encyclopedia part is the work of the knowledge graph people and their RDF triples. The training set and its labels is the job of the machine learning specialists. The data scientists, for their part, build business intelligence tools. These activities will converge and a new kind of semantic engineers will emerge. Furthermore, communities, institutions and companies of all sorts will be able to share their semantic knowledge in a virtual public database, or to exchange their private data on a new semantic market.