« Accueil » : différence entre les versions
Aller à la navigation
Aller à la recherche
Contenu ajouté Contenu supprimé
Aucun résumé des modifications |
Aucun résumé des modifications |
||
| (4 versions intermédiaires par le même utilisateur non affichées) | |||
| Ligne 7 : | Ligne 7 : | ||
The three levels are readable as separated parts (sets of articles) in this wiki. |
The three levels are readable as separated parts (sets of articles) in this wiki. |
||
An [[abstract]] provides a business oriented summary. |
|||
Assothink is a project of [mailto:philippe.gonze@matscape.com Philippe Gonze]. |
Assothink is a project of [mailto:philippe.gonze@matscape.com Philippe Gonze]. |
||
== Abstract == |
|||
The Assothink project relates to artificial intelligence. |
|||
It includes both a general theory of associative intelligence and a set of software intended ti implement the theory. |
|||
The model is layered. |
|||
The first layer is a concept universe. While most or all AI theories use ''named'' concepts, the Assothink model is built on ''unnamed'' concepts: this makes it langage-independant. |
|||
The second layer, called passive jelly, adds to the concepts links between them. Links may be qualified (when the link nature may be defined itself as a concept) or fuzzy (when the link is associative, i.e. general and fuzzy,). Associative links have a permeability value, indicating how strongly concepts are linked. |
|||
The third layer, called active jelly, includes excitation levels to all conceptual nodes. Excitation levels are scalar values evolving in time (a transposition of the conscience content for natural intlligence). The evolution of excitation levels is determined internally and externally. The internal incidence is guided by the network of links between active (excited) concepts, the permeability playing the central role in this diffusion/resonance process. The external incidence is guided by concepts excited by any kind of external source, which is also evolving in time. The global excitation level is ''conservative :'' this means that excitation levels are transferred between concept nodes, but never created or destroyed; the sum of all excitations in a system is constant. The speed of the active jelly is the critical component for implementtions of the model. It determines how fast excitation levels of the concept nodes are updated, propagated and adapted when external signal affect the system. |
|||
The fourth level adds language anchors to the concepts present in the universe. Language anchors are not critical for the perfomance of the system, but they are absolutely mandatory to allow communication betwen the system and the exetranl world. This is required when excitation signal rae introduced. It is also required to understand and actually describe the content of a system, i.e. the currentyl excited nodes. All human language are poorly designed, unpractical and very slow, but this complex part of the project is required to allow commication between |
|||
The Assothink model may be implemented through hardware or through software. |
|||
The hardware should be a new kind of computer (an associative computer, intended to ''think'' instead of to compute). The hardware implementation should possibly include very numerous (~100000) basic processors, one per concept, and should mainly be focused on the propagation of signals between these concept-node-processor basic entities. This engine would possibly be an analogic (instead of digital) device. This approach is rather heavy, and would probably require years of hardware-oriented reserach, so it is NOT the first targeted implmentation of the Assothink model. |
|||
The software solution is more realistic and feasible. Its performance might not be as impressive as the performance expected from a dedicated device (nor as the perfoamnce of natural brains). But it is a much easier and smaller project, achievable by a set of hig hlevel programmers, model expert, linguistics expert. The language would be iether C (for performance) or preferably a OO langage (for maintenability). A networked solution should include thsi components, preferably on seprated machines: (a) active jelly engines (b) communication point, to send excitation signals, to watch engine behavious, to organize learning processes. A staff of 3 to 10 good profiles should reach interesting software results with 2 or 3 years. |
|||
Learning would be the main challenge for an Assothink based entity (here we call it Alex). The learning process should allow to create concepts and links; to create and tune permeabilities, so that the propagation produces interesting ''resonance'' effects. To organize input and output for Alex. Exaclty like a human baby (or a young animal), Alex should progressivelyy fill, enrich and tune the content of its jelly, to make it fast, exhaustive, reliable. Practically, Alex should be fed by ontologies (numerous available), language-base system (like wordner and babelnet), knowledge sets (like google ffreebase, becoming knowledge graph), encyclopedia (wikipedia, wiktionary, wikidata), and possibly later by human-readable texts (providing Alex progressively builds a semantic analyzer able to decode human sentences into concept articulations). |
|||
The potential application fields are as numerous and various as the field of our natural-intelligence-driven activities. They include |
|||
*creation of a synthetic language (more suited than human languages for effective communication) |
|||
*improved translation system (concept-based) |
|||
*theoretical design of new hardware device (associative computers) |
|||
*non numeric computation (similar to the computation performed by human brains) |
|||
*efficient diagnostic systems (medicine and others) |
|||
*software programming performed by associative articial intelligence (this probably involves the creation of some new computer language) |
|||
*support device for natural intelligence (old persons losing memory - mentally handicapped persons) |
|||
*intelligent robots |
|||
Globally the Assothink project is a typical ''emergence''-oriented project. Most scientific intiative are reductionnist, while the most spectacular progress in the world and often in science are actually the result of unpredicatble emergence effects. This means that most potential applications of Assothink are typically unpredictable too. |
|||
As of january 2017, the Assothink projects includes 3 years of modelling, design and programming. This includes already several trail-and-erros, many redesigns. It may be used a solid base for further and wider developments. |
|||
== The assothink theory: overview == |
== The assothink theory: overview == |
||
| Ligne 75 : | Ligne 38 : | ||
The currently worked architecture (2012, 2013, 2014) is the [[A3_architecture|A3 architecture]]. |
The currently worked architecture (2012, 2013, 2014) is the [[A3_architecture|A3 architecture]]. |
||
== Assothink online == |
== Assothink online == |
||
You can directly access the online web application (basic architecture) on [http://www.assothink.com www.assothink.com] (note: |
You can directly access the online web application (basic architecture) on [http://www.assothink.com www.assothink.com] (note: currentlydisabled). |
||
Online '''user guides''' exist for the [[Passive_Jelly_Browser|Passive Jelly Browser]] and for the [[Active_Jelly_Browser|Active Jelly Browser]]. |
Online '''user guides''' exist for the [[Passive_Jelly_Browser|Passive Jelly Browser]] and for the [[Active_Jelly_Browser|Active Jelly Browser]]. |
||
| Ligne 96 : | Ligne 60 : | ||
|
|
||
|
|||
| ⚫ | |||
| ⚫ | |||
The project hibernates since 2014, the main reason being its global size (human resources, staffing, time horizon, financial resources). |
The project hibernates since 2014, the main reason being its global size (human resources, staffing, time horizon, financial resources). |
||
Dernière version du 26 janvier 2017 à 19:29
The Assothink approach involves three complementary levels.
- A general theory of associative intelligence
- The Assothink model
- A set of hardware and software architectures implementing the model
The three levels are readable as separated parts (sets of articles) in this wiki.
An abstract provides a business oriented summary.
Assothink is a project of Philippe Gonze.
The assothink theory: overview
The Assothink theory is both descriptive and elaborative. It is descriptive when attempting to describe parts of natural intelligence processes. It is elaborative when trying to set up forms of artificial intelligence.
Unlike other approaches (ontologies, dictionaries, thesaurus), the Assothink approach tends to manipulate concepts free from any kind of language anchoring. At the basic level, concepts are intrinsically unnamable objects.
The key hope in the assothink theory is emergence. This clearly distinguishes Assothink from reductionnist initiatives (see "reductionnism vs emergence"),.
A more detailed description of the theory is developped in the theory pages.
More recent document describe the theory using various simple but instructive diagrams:
The assothink model
A detailed description of the model is developped in the model pages.
Assothink architecture
The standard computers are poorly adapted to the need on an associative parallel software, like Assothink.
So several architectures are considered, more or less realistic, more or less expensive.
The currently worked architecture (2012, 2013, 2014) is the A3 architecture.
Assothink online
You can directly access the online web application (basic architecture) on www.assothink.com (note: currentlydisabled).
Online user guides exist for the Passive Jelly Browser and for the Active Jelly Browser.
Various technical notes related to the software are described in the software pages.
Additional internal notes
The current and planned development are in the Roadmap.
Originally the project name was Alex. Both names refer to the same project, but Assothink is used for the model and for the theory, while Alex would be used to name the emerging object with AI capacities.
And here is an overview of the Alex phasing process.
Project sizing and status (as of end of 2016)
The project hibernates since 2014, the main reason being its global size (human resources, staffing, time horizon, financial resources).
It is now considered that
- a minimal staff would include 3 programmers
- a reasonable and effcient staff would include 3 more profiles (including probably an advanced linguist).
- staffing might give possibly work to 20+ profiles (depending on the assigned goals)
- financial resources for entreprise-based project should then be at least 200 k€ per year during at leat two year
- public money is certainly available for such projects (european ? , national ? , chinese ? ...) but public money requires academic frame or enterprise frame (with mid/long term vision)