CRANIS Theory
cranio-, crani-, cran-
(Greek > Medieval Latin [c.700-c.1500]: head, skull)
This page is part of the obsolete CRANIS project (2010).
CRANIS is a theory (described in this article), reflected in a model and an implementation.
Performance of natural intelligences
All vertebrate animals (including homo sapiens) have developed complex and efficient neural structures. These structures perform I/O tasks: they handle sensory input signal and motor output signal. They perform non-conscious tasks. They perform semi-conscious and conscious tasks. The last part - conscious and semi-conscious - is a system able to solve various complex problems with an impressive speed and with a very low error rate.
In the human species this intra-individual structure has been extended with a specific inter-individual communication structure: the language. The language is said to be the specific attributes of our species, and indeed it helped the homo sapiens to completely dominate the planet biotope.
But is the language an efficient, completed structure ? The language involves messages. Messages are encoded and emitted by a source brain. They are received and decoded by a destination brain (often many or none). An efficient messaging system should:
- exchange signal in a fast way
- generate few errors
- provide efficient feedback
If the language is analyzed as a messaging system tool, it appears to be certainly far from efficient. The data transfer rate is hugely slower than the intra-individual brain message processing. A lot of errors appear during message transmission as a result of human languages features (polysemy, synonym, syntactical and semantic aberration and complexity). Besides that the attention of the destination brain is difficult to monitor, and the available feedback is extremely poor.
The language components are words. But the brain structures works on concepts, NOT on words. A human brain, or an animal brains is able to distinct and recognize category concepts, instance concepts, quantitative concepts, and various relations between them, and all of this is handled WITHOUT words.
The conclusion of these rather obvious points are summarized in this table:
| Intra-individual brain process | Inter-individual communication | |
| Support | Consciousness | Language |
| Components | Concepts | Words |
| Errors? | Few errors | Numerous errors |
| Speed | Very fast | Very slow |
The point to stress concerns words versus concepts. Concepts as basic intelligence components are MUCH better than words.
But obviously words are widely used in human shared activities, whereas concepts are generally hidden by words.
This is disappointing, but the reason is simple: a concept in itself has NO name. These things that human beings and animals intensively manipulate in their brains are impossible to represent or access through an access method. In computer science, objects are accessible through numeric indexing methods or through key indexing methods. Words are accessible objects. Concepts are inaccessible objects.
| Concepts | Words | |
| part of | Consciousness | Language |
| Efficient | Yes | No (various form of ambiguities) |
| Accessible | No | Yes |
| Speed | Very fast | Very slow |
Concepts in AI research and in linguistic research
As a result of the inaccessibility of concepts, most or all research in AI and in linguistic area work on an intricated mix of words and concepts.
A remarkable example is WordNet. This is one of the most complete network of concepts. To describe concepts, it uses the approximation of synsets (sets of synonyms). Various links are defined between synsets. But the data structure and the user interfaces provided with wordnet completely mix words and concepts, and generally non-English linguistic expert consider that WordNet is tightly linked to the English language. As a result, the research community has produced various wordnet for various languages.
The translation solutions obviously deal with words. But the internal processing generally deals with words, more than concepts. As a result, the translation is defined for any pair of language, and on a planet with more than 1000 spoken languages, a universal translation system should provide 1 000 000 pairs to allow communication between human being having identical brains and sharing identical concepts.
The subtle and complex projects involved in the Web semantic world, and certainly the ontologies, show an attempt towards concepts, rules, abstraction, but all of them show an imperfect layering between concepts and words, either in the model or in the processing.
The CRANIS view of natural and artificial intelligences states that
- Concepts are efficient and fast
- Concepts show a low level of ambiguity
- Concepts are difficult to handle as a result of their unnamability, but not impossible to handle. The natural intelligence show that it is possible.
- Concepts are networked through relations. Actually a concept is ONLY defined by the set of relations in which it is involved.
- Concepts are excitable objects
- Concepts are organized in a dedicated structure called consciousness.
Used of words (instead of concepts) is this article for instance is an interesting and paradoxical observation. It demonstrates again the unnamability and inaccessibility on concepts in the communication process. But the processes occurring in the brain of the destination brain (your brain, reader) certainly activates concepts and relations between concepts, instead of words or relations between words.
Consciousness
The consciousness is the key component of a CRANIS system, be it natural or artificial.
The following statements are based of various scientific material, but also on introspective analysis.
The consciousness involves a wide number of nodes called concepts.
Within a consciousness, the concepts are linked to other nodes through relations.
Relations themselves are concepts.
Concepts are organized in categories and subcategories, using a tree structure.
Categories themselves are concept.
CRANIS systems form a category.
The perception of itself by a CRANIS system is a specific unique concept (an instance of the CRANIS systems category, and a concept of 'myself')
Concepts may be class or instances.
The top categories include at least : relations, category, nouns, verbs, adjectives, cranis systems, quantity concepts,...
The concepts nodes and the links concepts together form a passive jelly. This jelly is the field where excitation signals and excitation levels are created or propagated. The excited, activated jelly is described in terms of focus and excitation.
Consciousness, Focus and Excitation
The following statements are based on introspective analysis.
The consciousness active in any of us involves excitation levels for various nodes, and signal propagation based on relations.
But the consciousness is not working as a single entity. A human brain is able to perform various semi-conscious tasks (walking, eating,...) simultaneously. Beside that, it is able to perform various conscious tasks : play a game, discuss with other players, evaluate winning chances, performing calculus tasks, manipulating recent and future game events, thinking to personal generic problems. These numerous tasks are not processed simultaneously. They are processed sequentially, and the focus of the conscious part of a brain jumps from one focused taks to another one. The player has to choose a card. He switches to a review of recently played cards. He switches to quantitative probabilistic evaluations. He switches to an attempt to decode facial signals in other players... Finally: he plays. Subtasks have produced their partial results, and the global playing task has integrated the available data into one single choice decision.
Possibly, interruption have risen during the thinking process. A language message has possibly been received, a landscape view with some faces has possibly been loaded and decoded...
These successive focus are organized as a tree. Various tasks call subtasks, and wait for their results. Interrupting tasks receive temporary attention. Most are quickly discarded.
The focus move and the task switching mechanism suggest:
- a form of serialization for conscious focuses
- a tree structure for the focus tree
- a result oriented activity for each focus
- the presence of a main root focus
Where operates the excitation level ? The excitation level is related to a focus, not the global consciousness.
Finally the CRANIS theory states that, on top of the passive jelly :
- excitation is a scalar value linked to any concept node
- excitation is dynamic within a focus: it is transmitted across concept nodes using excitation signals, and relation channels
- focus are organized as a tree, with a permanent persisting main focus (sleeping time is the exception)
- focus produce results
- focus are subject to switches and interruptions
- within a focus the networked excitation generates and uses resonance effects
Qualified and fuzzy relations
Concepts are linked through relations.
Sometimes a relation is qualified. The relation between an instance (a visible dog) and a class (the class of all dogs) is qualified: it is a instance-to-class relation. The relation between 3 (the quantity) and 4 (the next quantity) is qualified: it is an quantity-to-next-quantity relation. Any relation is a concept itself. Relation concept may also be linked by relations: quantity-to-next-quantity and quantity-to-previous-quantity are linked by a reciprocity relation.
But many relations are unqualified. In the brain of a human being, we can find numerous associated concepts without directly identifiable relation. Some of them are objective,a nd present in most brains. And many other are personal, relating from various experiences of the involved specimen. These unqualified relations are fuzzy: some of them are high, which means that they establish strong connections between two concepts. Other are low, but present, which means that they establish weak connections between two concepts. This property of links is named permeability. A strong connection is the result of a high permeability, and a weak connection is the result of a low permeability. No direct link between two concepts is the effect of no relation between them. The permeability determines the way signal are propagated across concept nodes.
Resonance
If various concepts nodes propagate signal to each other, the combined effect of these propagation generated resonance. The resonating nodes are those forming together a network of concepts are elected as a combined effect of
- external excitation of the involved nodes
- relationships between resonating concepts stronger than for any other possible combinations of concepts
In the CRANIS theory, this resonance across concepts is where intelligence raises.
Problem and solution
A problem is introduced in the form of a set of excited concepts.
The resonance processing is an extension modification of the excited concept set.
When the resonance has evolved to a stable state, or when a node of a specific target category is excited, the solution is declared available.
Simple calculus problems are solved this way.
Complex calculus problems are solved using a subtree of focus. The simplest focuses provides results to the more complex focuses.
In pattern recoginition, the candidate targets are concepts nodes. The pattern problem is achieved when the one of the target nodes is in excitation state significantly higher than all other candidates.