compositional design of a generic design agent
Post on 02-Jul-2016
Embed Size (px)
1 Charlton, C A web broker formechanical component selec-tion, in P Rodgers and A Huxor(eds) Distributed web-based AIdesign tools: AID98 Workshop 2Notes, Worcester PolytechnicInstitute, Worcester, MA (1998)
www.elsevier.com/locate/destud0142-694X/01 $ - see front matter Design Studies 22 (2001) 439471PII: S0142-694X(00)00044-2 439 2001 Elsevier Science Ltd All rights reserved Printed in Great Britain
Compositional design of a genericdesign agentFrances M.T. Brazier, Catholijn M. Jonker, Jan Treur and Niek J.E.Wijngaards, Vrije Universiteit Amsterdam, Faculty of Sciences,Department of Artificial Intelligence, De Boelelaan 1081a, 1081 HVAmsterdam, The Netherlands
This paper presents a generic architecture for a design agent, to beused in an Internet environment. The design agent is based on anexisting generic agent model, and includes a refinement of a genericmodel for design, in which strategic reasoning and dynamicmanagement of requirements are explicitly modelled. The genericarchitecture has been designed using the compositional developmentmethod DESIRE, and has been used to develop a prototype design agentfor automated agent design. c 2001 Elsevier Science Ltd. All rightsreserved.
Keywords: design model(s), design automation, software design,artificial evolution, computational model
Design is a task often performed by one or more specialised(human) agents. Architects are, for example, specialised agents:their area of expertise is the design of buildings in given sur-roundings. Design agents negotiate with other agents on requirements andgenerate one or more designs (design object descriptions) on the basis ofthese requirements and additional information received from other agents.Design agents make these design object descriptions (together with allother (intermediate) results of the design process) available to other agentsin the course of the design process, and react to input provided by otheragents as a result.
The World Wide Web provides a rich potential for distributed design inwhich designers and other parties interact. The number of possibilities isimmense: designers can navigate through component libraries available onthe Web1, they can choose between a large number of informal textual andformal types of knowledge representation (and be supported in the
2 Huxor, A, Rogers, P, Clark-son, J and Caldwell, N Knowl-edge-based systems as navi-gation aids to online content andusers, in P Rodgers and AHuxor (eds) Distributed web-based AI design tools: AID98Workshop 2 Notes, WorcesterPolytechnic Institute, Worcester,MA (1998)3 Brazier, F M T, Jonker, C Mand Treur, J Compositionaldesign and reuse of a genericagent model Applied ArtificialIntelligence Journal Vol 14(2000) 4915384 Brazier, F M T, Langen, P HG van, Ruttkay, Z and Treur, JOn formal specification ofdesign tasks, in J S Gero andF Sudweeks (eds) Artificial Intel-ligence in Design 94, KluwerAcademic Publishers, Dordrecht(1994) pp 5355525 Brazier F M T, Dunin-Keplicz B M, Jennings N R andTreur J Formal specification ofmulti-agent systems: a real worldcase, in V Lesser (ed) Proceed-ings of First International Confer-ence on Multi-Agent SystemsICMAS95 MIT Press, Cam-bridge, MA pp 2532 (1995).Extended version in M Huhnsand M Singh (eds) InternationalJournal of Co-operative Infor-mation Systems IJCIS Vol 6 No1 (1997) 6794 (special issue onFormal methods in co-operativeinformation systems: multi-agent systems)6 Brazier F M T, Dunin-Keplicz B, Treur J and Ver-brugge L C Modelling internaldynamic behaviour of BDIagents, in A Cesto and P YSchobbes (eds) Proceedings ofthe Third International Workshopon Formal Models of AgentsMODELAGE97. Lecture Notesin AI Springer Verlag Vol 1760(1999) 36567 Brazier F M T, Jonker C M,Jungen F J and Treur J Distrib-uted scheduling to support a callcentre: a co-operative multi-agentapproach, in H S Nwana and DT Ndumu (eds) Applied ArtificialIntelligence Journal Vol 13 (1999)6590 (special issue on Multi-agent systems). Earlier shorterversion in H S Nwana and D TNdumu (eds) Proceedings of theThird International Conference onthe Practical Application of Intelli-gent Agents and Multi-AgentTechnology PAAM98 The Practi-cal Application Company Ltd,Blackpool (1998) pp 555576
440 Design Studies Vol 22 No. 5 September 2001
process), they can interact with other parties interested in the same Websites as they themselves2, they can perform calculations, to name a few.
This paper introduces a generic architecture for a design agent that can beintegrated in a Web-based distributed design environment. This architec-ture is based on the Generic Agent Model GAM introduced in 3 and theGeneric Design Model GDM described in 4. Both of these models havebeen devised on the basis of experience, and tested in a number of differentdomains: applications of GAM can be found, for example, in 57; appli-cations of GDM can be found, for example, in 8,9. To tune a generic modelto a specific domain of application, more refined knowledge is needed:both more specific knowledge of the task at hand (by specialisation), andmore specific factual knowledge of the domain (by instantiation). Forexample, domain-specific knowledge is needed to derive whether a givendesign object description satisfies given properties (e.g., requirements), andknowledge that can be used to derive how to refine requirements into morespecific requirements.
The architecture for the design agent has been modelled using the compo-sitional development method for multi-agent systems DESIRE5. This com-positional development method is briefly introduced in Section 1. The gen-eric agent model employed is described in Section 2. The generic modelof design4 employed, in which strategic reasoning and dynamic manage-ment of requirements are explicitly modelled, is described in Section 3.The integration of the two models is discussed in Section 4. Although boththe Generic Agent Model GAM and the Generic Design Model GDM havebeen tested in different application domains, the architecture for a genericdesign agent described in this paper as yet has only been tested in theapplication domain of automated design of Internet agents. Therefore thisapplication domain is used to illustrate the applicability of the proposedapproach. The specific design agent for this application is described inSection 5. Section 6 compares the generic agent model GAM and the gen-eric design model GDM to other agent and design models. Section 7 dis-cusses the results and sketches a perspective of the use of the design agentarchitecture in Web-based environments for Electronic Commerce appli-cations: a current focus of research.
1 Compositional design of agentsThe design agent described in this paper has been developed using thecompositional development method DESIRE for multi-agent systems(Design and Specification of Interacting Reasoning components; cf. 5).Within this method, knowledge of the following three types is dis-tinguished:
8 Brazier F M T, Langen P HG van, Treur J, Wijngaards NJ E and Willems M Modellingan elevator design task inDesire: the VT example, in A ThSchreiber and W PBirmingham (eds) InternationalJournal of HumanComputerStudies, Vol 44 (1996) 4695209 Brazier, F M T, Treur, J andWijngaards, N J E Interactionwith experts: the role of a sharedtask model, in W Wahlster (ed)Proceedings of European Con-ference on AI ECAI96, Wileyand Sons, Chichester (1996) pp241245
441Compositional design of a generic design agent
process composition; knowledge composition; and the relation between process composition and knowledge composition.
The development of a multi-agent system is supported by graphical designtools. Translation to an operational system is straightforward; in additionto the graphical design tools, the software environment includes implemen-tation generators with which specifications can be translated into execut-able code of a prototype system. The three types of knowledge are dis-cussed in more detail below.
1.1 Process compositionProcess composition identifies the relevant processes at different levels of(process) abstraction, and describes how a process can be defined in termsof (is composed of) lower level processes.
1.1.1 Identification of processes at different levels ofabstractionProcesses can be described at different levels of abstraction; for example,the process of the multi-agent system as a whole, processes defined byindividual agents and the external world, and processes defined by task-related components of individual agents. The identified processes are mod-elled as components. For each process the input and output informationtypes are defined. The identified levels of process abstraction are modelledas abstraction/specialisation relations between components: componentsmay be composed of other components or they may be primitive. Primitivecomponents may be either reasoning components (i.e., based on a knowl-edge base), or components capable of performing tasks such as calculation,information retrieval, optimisation. These levels of process abstraction pro-vide process hiding at each level.
1.1.2 Composition of processesThe way in which processes at one level of abstraction are composed ofprocesses at the adjacent lower abstraction level is called composition. Thiscomposition of processes is described by a specification of the possibilitiesfor information exchange between processes (static view on thecomposition), and a specification of task control knowledge used to controlprocesses and information exchange (dynamic view on the composition).
1.2 Knowledge compositionKnowledge composition identifies the knowledge structures at differentlevels of (knowledge) abstraction, and describes how a knowledge structurecan be defined in terms of lower level knowledge structures. The knowl-
10 Wooldridge, M J and Jen-nings, N R Intelligent agents:the