Post on 18-Dec-2014
description
KNOWLEDGE REPRESENTATION:
DESTRUCTURING THE
STRUCTURED vs NON-STRUCTURED DEBATE
Jean Rohmer ESILV Paris
jean.rohmer@devinci.fr
Presented at ECAI 2012 Montpellier Workshop on AI and KM
My personal background in CS, AI and KM
Started Computer Science 45 years ago
Started AI 32 years ago
Started KM 24 years ago
Management of Bull CEDIAG team
IDELIANCE Semantic Tool (1993)
Many Military Intelligence Applications Data + Text + Semantics
Blog: "PLEXUS LOGOS CALX"
See also SLIDESHARE Jean Rohmer
Progress in KR is slow.
Mesopotamia 5500 years ago:
Mesopotamia in the 21 th Century: still Stone Age:
AI and KM: once a Love Story
In the late 80's a love story between AI and KM
Their alliances: (rings) Knowledge Representation and Inference
Importance of KRL languages, KADS modelling : Open Kads tool (1991)
Early 90's: economical crisis: the AI + KM couple almost starving
AI and KM were young, promising, but still immature
KM alone could earn some living in large corporations
The Web arrived and seduced KM
AI was left alone
<<< Tim Berners Lee paper proposing the Web was rejected at the 1991 ACM Hypertext
Conference>>>
Hypertext was very close to KM
Catastrophe
2012: Large scientific Agencies manage all their projects with EXCEL
2012: Many Engineering Schools have no real information systems
2012: ECAI program, proceedings are available just in PDF, without any tool for knowledge organization
2012: they swapped my last name and first name in SOME ECAI registration files
AI and KM are alone
AI lives with Automatic Learning Algorithms
KM flirts with wikis, blogs, social networks
The main tool for AI is SVM algorithm (sort of joke)
The main tool for KM is EXCEL + POWERPOINT (not a joke)
There is no paper on KR at ECAI 2012
Denegation: "AI is hidden everywhere"
Laurence Danlos: (NL guru):
"We failed to make machines adapt to humans; we humans have learnt how to use windows and menus"
History
In the early 80's, AI languages (LISP, PROLOG, KRL, Constraints later) were seen as the promise of a revolution in programming computers: declarative programming
1982: Alain COLMERAUER declares that PROLOG is designed to replace COBOL
European Esprit programme: 1982: KIMS project "Knowledge and Information Management
System"
Earlier: Alan Turing tried to get funds from UK Gvt to build a sort of LISP MACHINE
Earlier: Leibniz and Descartes proposed universal knowledge representation and reasoning languages.
PROJECT OF A COMPUTABLE UNIVERSAL LANGUAGE
INCLUDING UNIVERSAL ONTOLOGIES WITH « COMBINATORIAL » MECANISMS
DESCARTES :
« établir un ordre entre toutes les pensées, … de même qu'il y en a un établi entre les nombres »
« cette langue aiderait au jugement , lui représentant si distinctement les choses qu’il lui serait presque impossible de se tromper »
« je tiens que cette langue est possible … mais n ’espérez jamais la voir en usage … sauf au Paradis Terrestre … »
LEIBNIZ : « quoique cette langue dépende de la vraie philosophie, elle ne dépend pas de sa perfection »
« à mesure que la science des hommes croîtra, cette langue croîtra aussi »
« alors raisonner et calculer sera la même chose »
80's: Expert Systems with KNOWLEDGE ENGINEERS
1988 -1992: METAPEDIA project in SPAIN: a fully object-oriented encyclopaedia
1990: Idea that future Corporate Information Systems would be Knowledge Based Systems
1991: MNEMOS EUREKA European project
1991 (Bull Cediag): Corporate Intelligence = Corporate Memory + Corporate Decision + Corporate Visibility
PROLOG
In 2012 we celebrate the 40th anniversary of PROLOG
(Where is the cake ?)
Personal History
1984: “Alexander Method” (Foundation of Datalog / Deductive Databases)
For me, illuminated by Prolog , “Everything was logic predicates”
1990: Expert Systems were very successful
1990: Expert Systems demand much more intellectual energy than available
1993: Start developing IDELIANCE: a personal semantic networks manager for "everybody"
fr.slideshare.net/Jean_Rohmer/ideliance-semantic-network-2000
IDELIANCE: Personal Memory + “Intelligence Amplifier “
Mid 90's: sadness that AI languages disappear from education
2003: Semantic Networks is a too complex formalism for people; 99% reject it
2003: Idea of LITTERATUS CALCULUS: use plain natural language to represent knowledge
LITTERATUS CALCULUS: express anything with "inferons": minimal and autonomous sentences in natural language
2001 +: Strong critique of Semantic Web à la W3C
Structured vs Unstructured
Unstructured is in fact HYPER-structured
Structured is in fact HYPO-structured
Natural Language is HYPER-structured
Natural language structures are so complex that we do not know how our brain master them
So-called structured information (databases, RDF triples) are trivial structures
to match computer limitations
All the problem of KR is that we are not able to write programs which understand natural language
Semantic Networks is a good compromise between man and machine
Semantic Networks were used already in the 16th Century to represent complex information
Semantic Networks are readable by humans if small enough (Not billions of triples, leave it to NoSQL! )
Semantic Networks is a 2D representation
2D representation avoids the usage of variables as in formal logic
IDELIANCE Semantic Network editor: experience since 1993
Used by many NON CS professionals in large corporation
99% of people are reluctant to write themselves semantic networks
Use semantic networks with a Subject Verb Complement (SVC) paradigm
Let people use natural language to name S, V, C (never RDF, "Resources", URI ...)
Let people write "SVC on SVC" using a 4th ID field (NOT contexts, named graphs ...) (SVCI
format):
Please users, not standardization committees
Negative effects of the Web and Semantic Web on KR
Is Semantic Web a bad Joke ?
SW 2001: "Machines understand and help Humans" (Scientific American Paper)
SW 2006: "A machine-to-machine Web of data"
SW 2011: Linked Data: "Humans help Machines"
SW 2016: ????
An endless loop / ping-pong of failures between manual and automatic, structured and unstructured
Notion of URI is just a physical address scheme without any natural support
The Web reinforces the notion of -long- document
RDF has no "human face"
RDF is at best low level engineering and exchange format
Structured data publishing -dbpedia, Google- do not follow SW standards
Ontologies are too simplistic at RDF level
Ontologies are too complex at DL level
What was difficult to solve in the 90's with powerful KR languages on limited problems
cannot be solved in the 2010's with just Java and RDF at the Web scale
What we have to do is to install a good KR on the Internet, rethinking all the KM issues
The best -only- KR available is natural language
Natural Language does not imply "Document"
Natural Language does not mean "non -structured"
Représentation 1
A good KRL should be enjoyed by people
People should write, query, compute themselves with their KRL
Example of personal objective: take my reading notes directly in a KRL
Parabola of the ship inside the bottle: Knowledge must be cut into articulated small parts
Example of personal objective:
Summarize "Cours de Linguistique Générale" of Ferdinand de Saussure with my KRL
Tools are important! Never say "This is just a tool".
Intelligence is just a tool ... ????
Natural Language is just a tool ... ???
Many people say "Computer is just a tool" AND "Computers will change everything" …
Theory
Theory of the two black holes
Man-machine compromise schema
A good KR should be targeted at killing applications (App-Killer and not Killer App!)
Applications hide all knowledge: they presents users with a closed, limited, repressive view of the world
Replace applications by the way people will interact and compute with knowledge
A good KR should be targeted at killing the Document paradigm
Document paradigm is a concept imposed by the technology of "volumen' and "codex" more
than 2000 years ago
A good KR should aim at revolutionizing the Web (what else ?)
Representation 2
People should enjoy using themselves directly KR
People should write KR instead of writing documents
Computations on KR done directly by users should replace applications exactly as EXCEL does with numeric data
KR should be the backbone of "Semantic EXCEL" and "Semantic PowerPoint"
Collective KM fails if it is not grounded in personal KM, through a personal, intensive effort
to write, read, retrieve, combine, compute knowledge with a good KR
We must invent new ways of browsing, editing, computing on knowledge.
Examples of new computations: "In between", "novelty detection", "how to", "what looks like" , "online graph mining"...
How to proceed towards a good KR ?
Issue: what else do we have than KR progress to improve information systems ?
We must abandon the paradigm of PRO-GRAMMING
PRO-GRAMMING means “WRITTEN BY ADVANCE”
We most practice IM-PRO-GRAMMING
IM-PRO-GRAMMING means IM-PRO-VE
IM-PRO-GRAMMING means IM-PRO-VISE
IM-PRO-GRAMMING needs the appropriate KR paradigm
LITTERATUS CALCULUS
The only thing you put in a computer is sentences in natural language
INFERON: minimal and autonomous sentence
Every information is INFERON
There are no entities
Entities emerge from sentences
Instead of “Sentences are built from entities”
Many tools to manage inferons: editing, browsing, query, inference, ...
My personal KB has today 70 000 inferons
A first version of a Litteratus Calculus tool is being implemented (since 2003 …)
Current work: how to install INFERONS on the Internet ?