The Semantic Web: Implications for Future Intelligent Systems

1
The Semantic Web Implications for Future
Intelligent Systems

• Lee McCluskey,
• Artform Research Group,
• Department of Computing And Mathematical
  Sciences,
• University of Huddersfield

2
Meta-Talk

• Outline
• Web / Semantic Web
• XML,RDF,Oil,DAML
• Ontologies
• Intelligent Systems
• IS and the SW
• Related Areas Distributed AI (DAI), Ontologies,
  Mark-up Languages, Standards, Knowledge
  Management, e-Commerce (B2B,B2C), OODBS, ...
• Caveat I usually give seminars on subjects which
  I am familiar as a researcher and builder of the
  technology - NB with this one I am only a
  potential user!

3
Take-home Slide

• The Semantic Web is the Vision (not a current
  reality) of having an internet with resources
  that are machine understandable or accessible to
  automated processes - machines should do much
  more than present the information visually or do
  human-consumable IR.
• Already very high level languages are being
  designed for this purpose (with XML as machine
  code)
• The first International SW conference took place
  in 2002 (ICSW02, Sardinia)
• Processes on the SW will need to do AI Planning
  and AI Planning can exploit the SW to do
  Knowledge Acquisition.
• BUT still a long long way to go before realisation

4
WWW

• - is successful largely through the use of layers
  of internationally accepted standards
  (TCP/IP,html)
• - first generation - hand written html pages
• - second generation - dynamic web - pages
  created by programs to display the results of a
  process, or the output of a query of an accessed
  database.
• --so web pages used as an interface to networked
  processes (services) as well as for general
  information display.

5
WWW

• Now much RD has been directed at writing
  programs/services that utilise HTML web info
• EG my favourite - ISIs travel assistant - a web
  service that uses other web services (weather,
  timetables, hotel) to make travel plans in
  response to a high level directive I need to be
  in X on days Y using budget Z
• BUT this is very hard because of poor web
  structures.. Eg ISIs travel assistant has to use
  a learning program to induce web page wrappers

6
Metadata and XML

• Not surprisingly, to start to give meaning to
  info on the web we must use META-DATA eg using
  tags around data to describe its content.
• In XML - eXtensible Mark-up Language - tags are
  not fixed - one can invent new tags to structure
  the information in a web page.
• XML is considered to be the basis for all
  semantic web languages - the machine code of
  the new generation web

7
RDF - the Resource Description Framework

• RDF is a convention for describing meta-data.
• Its a lightweight model in data terms - and
  one that can be encoded in XML. It is based on
• everything having a URI Universal Resource
  Identifier
• Properties - resources with a name such as
  slots in an object frame
• An RDF document is a series of Statements -
• (Resource, Property, Value)

8
RDF example

• RDF set of (Resource, Property, Value)
• "The Author of http//scom.hud.ac.uk/scomtlm/Artfo
  rm/planning.html
• is Lee McCluskey.
• IN RDF
• ltrdfDescription about http//scom.hud.ac.uk/sco
  mtlm/Artform/planning.html'gt ltAuthorgt Lee
  McCluskey lt/Authorgt
• lt/rdfDescriptiongt
• Resource, Property, Values can all have URIs

9
RDFS -

• RDF Schema RDF classes, properties of
  properties, etc - gives more structure to RDF
• This would make the Semantic Web look like one
  enormous distributed OODB ( but DBs usually
  angled towards one application - the SW (will
  be?) is ubiquitous..)

10
Vocabularies.

• RDF/RDFS allows anyone to write their own
• name-space document (a schema). This defines
  properties and classes in some application domain
  
• These form vocabularies which can be used
  globally for sharing the meaning of tags

11
Ontologies

• An Ontology is an agreed on, shared, common
  understanding of a domain written as an explicit,
  formal specification.
• (cf my seminar on Ontologies last year! Slides
  can be got via my website under seminar slides)

12
Ontologies

• First used for Knowledge-Sharing in KBS, it seems
  that many scientific areas are creating their own
  Ontologies (even Genomics!!)
• They can be as simple as a concept hierarchy or
  as complex as an axiomatic theory of sets.
• There are two kinds of ontology representation
  ontology (axiomatization of basic operations used
  in many applications) and application ontology
  which are domain specific.

13
Oil (Ontology Infrastructure Language)

• Oil is a web-integrated prototype standard for
  specifying Ontologies

Based on Web Standards XML and RDF(S)
Natural Frame-based (OO) language
OIL
Description-logic-based semantics and reasoning
Tool Support e.g. Oiled
14
Hierarchy of Languages

• DAML OiL
• RDFS
• RDF
• XML

15
DAMLOil example

• Define a "product number"'s domain and range..
• ltdamlDatatypeProperty rdfID"productNumber"gt
• ltrdfslabelgtProduct Numberlt/rdfslabelgt
• ltrdfsdomain rdfresource"Product"/gt
• ltrdfsrange rdfresource
• "http//www.w3.org/2000/10/XMLSchemanonNegativeI
  nteger"/gt
• lt/damlDatatypePropertygt
• Availability" is a sort of enumerated type..
• ltdamlClass ID"Availability"gt
• ltdamloneOf parseType"damlcollection"gt
• ltdamlThing rdfID"InStock"gt
• ltrdfslabelgtIn stocklt/rdfslabelgt
  lt/damlThinggt
• ltdamlThing rdfID"BackOrdered"gt
• ltrdfslabelgtBack orderedlt/rdfslabelgt
  lt/damlThinggt
• ltdamlThing rdfID"SpecialOrder"gt
• ltrdfslabelgtSpecial orderlt/rdfslabelgt
  lt/damlThinggt
• lt/damloneOfgt
• lt/damlClassgt

16
What about Sevices?

• The idea of the SW now extends to Services
  networked processes may be specified in terms of
  pre- and post conditions, where these conditions
  are written in a globally accepted mark-up
  language!

17
Intelligent KBS

• Two related, major problems have remain unsolved
  in symbolic intelligent systems -
• -- Knowledge Acquisition - can be prohibitive!
  New forms of reasoning / intelligent process have
  had to have new hand-crafted knowledge bases
  built for them for each application area in which
  they are need to work.
• -- General reasoning systems dont scale up
  The high profile successes tend to be in very
  narrow applications such as in Chess or Expert
  Systems

18
The Solution in the Semantic Web

• The SW may solve BOTH these problems
• -- The SW infrastructure of Ontologies and
  structured web information could ease the
  problems of knowledge acquisition
• -- It could provide a global, dynamic,
  unlimited environment in which Intelligent
  Processes can be deployed
• General Planning and Learning programs will be
  much easier to apply and evaluate

19
Example - automated planning

GOAL gt
PLANNING AGENT
Sensors
Reasoning Mechanism World Model (facts, objects,
actions, rules)
gt PLAN gt
Effectors
20
Example - automated planning

• Knowledge Acquisition
• facts - from effective SW information retrieval
• actions - pre- and post conditions of semantic
  web services
• Online Ontologies - allow the Agent to interpret
  the acquired facts and actions
• Sensors - e.g. results of searches or
  communications
• Effectors - e.g. execution of semantic web
  services

21
My theory...

• The biggest problems in AI planning currently
  are lack of Accessibility and Usability of the
  technology
• Timely maturing of 4 research areas
• -- Semantic Web
• -- knowledge engineering, sharing and re-use
• -- Planning language conventions
• -- Planning KE
• will solve these problems..

22
Conclusions

• The Semantic Web looks like becoming a reality in
  the future (when?). It will be a bit like an
  enormous distributed OODB with composable
  services
• The Semantic Webs dynamic, boundless aspects
  coupled with structured descriptions of info and
  processes will present great opportunities for
  research and developers of AI technology (as well
  as much else!)

23
References

• You can get these slides from my web site
• Any other resources are easy to get - eg search
  on Semantic Web Conference...