CNET, the research center of France Télécom, has been engaged, for a couple of years, in the development, experimentation, and evaluation of different agent technologies in various contexts: telematic services (mediation, information retrieval), cooperative human-computer dialogue, telecom network supervision.

The present contribution comes from two CNET's groups concerned with agent-based software encapsulation and with rational agent designing for intelligent spoken dialogue, respectively.

The mobile agent technology has been mainly investigated in the General Magic environment, and the concept of agent-oriented programming (Agent0 à la Shoham) has been applied to the domain of telecommunication network supervision.

As regard the intelligent agent technology, a logical theory of rational interaction specifying a communicating intelligent agent has been developed and implemented (using a first-order modal logic theorem prover, also designed at CNET) as the reasoning unit of a cooperative spoken dialogue system. This system demonstrates the feasibility of spoken query of the France Télécom shared-revenue information voice services. It displays different abilities of intelligent user-friendly human-computer interaction : spoken language understanding and generation, reasoning, negotiation, cooperation, etc.

Along these lines, our interests in the scope of FIPA are twofolds:

1/ Software agents and interagent communication:

Basic communication primitives and "speech acts" for interagent communication, such as KQML (8), cooperation and negociation protocols such as the contract net protocol (9) and aspects of agent society such as delegation and brokering (16).

2/ Rational agents and human-agent communication:

Semantic information representation and cognitive architectures (knowledge representation, cognitive primitives, rational behavior principles, reasoning, automated inference) (12), speech and natural language interfaces for human-agent communication (1), agent programming human interface, and tool kits for implementing agent-based applications (14).

We believe that intelligent agent technology will play a central role in the future offer of advanced Telecom services. In this context, FIPA could be an adequate framework for identifying and/or defining standard descriptions for agent knowledge, capabilities, competence domains, and communication interfaces, either with humans or with other agents.

Possible topics to be examined at the meeting:

- Requirements for rational agents, corresponding cognitive primitives and knowledge representation models, and implementation implications.

- Impact of natural language (especially spoken) communication on the designing of rational agents.

- Contract Net Protocol (which is one of the reference protocol used by the multi-agent systems in Distributed AI) as an item of section 9 (Composition operators for basic messages types).

- Relationships between the cognitive architecture of agents and their social and communication capabilities.

T. BOURON, D. SADEK

France Télécom - CNET - LAA

22307 Lannion Cedex France

bouron, sadek@lannion.cnet.fr

Participants:

Thierry BOURON is head of the "Expert Systems and Agents" group at the "AI Applications" department (LAA/EIA/AIA), which is in charge of the studies on software agents, machine learning and 2nd generation expert systems. His thesis work was about the designing of a communication model based on the concepts of speech acts and commitment, which has been used to elaborate a social agent model including different levels of organization.

David SADEK is head of the "Human-computer Spoken Dialogue" group at the "Research in Speech Communication" department (LAA/TSS/RCP). His work concerns the designing of formal models of cognitive attitudes and rational action, and their application to the context of human-computer dialogue. He has proposed a logical theory of rational interaction, which is implemented as the kernel of an effective rational-agent-based cooperative spoken dialogue system.