Position Paper for the AgentLink SIG on Intelligent Information Agents
Brussels
24 -25 September 1998

New directions in multi-agent research at the University of Salford

Agent research at the University of Salford has predominately been focused on a heterogeneous multi-agent 
architectures as demonstrated with our simulated co-operative agents and mobile robots in such projects as 
MACTA (Multiple Automata for Complex Task Achievement). A new project MACTA-LAB that has just 
started will build on the success of MACTA, applying the architecture to laboratory automation. 

At the beginning of this year the Agent research group at the University of Salford moved to the new Centre 
for Virtual Environments. With our added new skills base and additional new resources we have expanded the 
breadth of our work to cover further types of agent research. Projects now in progress include work on 
intelligent virtual environments, a virtual agent toolkit, and distributed synthetic agents for education. Most of 
these projects are at the beginning of their life cycle, but we have already come across some issues that face 
the wider agent community and might be resolved by suitable collaboration with others. 

Synthetic Information Agents

With the advent of intelligent agent based retrieval systems, extracting information from large complex 
distributed systems, many of which are heterogeneous, it is easy for the scientist to overlook how this 
information is offered to the user. Information can be represented in many forms depending on its nature and 
the application it is intended for. 

In the past, conventional human computer interaction has been sufficient when developing software 
applications, but with the complex nature and quantity of information, and the tools used to gather it, we need 
intelligent agents to answer queries, provide guidance, and interpret information. Information agents are also a 
paradigm for how the human user interacts with a multi-agent or conventional information retrieval system. 

Humans are social animals and they respond most effectively in a social context, and therefore agents with 
personalities that interact naturally with humans will become increasingly important. Take for example 
STEVE (SOAR Training Expert for Virtual Environments) [Rickel and Johnson] developed at USC/ISI, this 
anthropomorphic agent helps train people to use complex pieces of equipment in fully immersive virtual 
environments. Successful pedagogical agents like STEVE use a wide range of new technologies such as a 
text-to-speech engine, voice recognition, and interactive virtual reality (VR) to name just a few. Although VR 
may not be needed for all pedagogical agents the extensive use of diverse technologies in such projects means 
that few organisations have the resources or complete 'know how' to start working on such projects alone. 
This problem could well be addressed by encouraging collaborative research ventures. 

New multi-agent systems have agents performing different tasks, some with local knowledge that other agents 
do not have. A lot of work has gone into agent communication but not so much consideration has gone into 
letting the user easily interrogate an agent network; allowing them to home in on the information they require. 
The concept of an interface agent, who is a single point of contact between the user and the other agents in a 
multi-agent system, is appealing, following the idealistic model of the desktop digital butler. As we have 
found at Salford the concept of a desktop agent is not always suitable; such as in virtual environments that 
may contain multiple real and virtual agents. This problem also occurs on agents running on a machine shared 
by multiple users, the agent must keep track of each user's profile and also recognise which user it is attending 
to.

New techniques in information retrieval 

As well as working with synthetic agents we have been doing some preliminary research into agent 
information retrieval (IR) for the World Wide Web (WWW). When this project started we were disappointed 
at the lack of foundation work that had been done on information retrieval in open systems. Conventional 
techniques for finding information on the WWW have their roots in the world of large databases. Robots 
using a variety of similar techniques would extract keywords out of a web page and store them in a large 
database waiting to be queried. The search engines of today still apply the same principle as the first did, just 
on a larger scale with some novel 'bells and whistles' added on. Users' expectations for finding information 
that meets their requirements are higher than ever with the popularisation of the WWW, and the volume of 
information is increasing and changing constantly. Despite the innovation of novel agent front-ends for search 
engines giving us IR shopping agents, and agents tracking stocks and shares, to mention just two examples, 
we see the development of new techniques in information retrieval as the cornerstone of overcoming the 
problem of information overload. We would like to explore and find new ontologies for IR for open systems, 
such as the application of AI planning to information gathering.  

Conclusions

The term 'intelligent information agents' throws a broad net over many areas of work within the agent field, so 
we hope it is of no surprise that our interests are wide and varied. We are aware that some of these questions 
have been echoed previously by others, and people in the field of agent research and development identify 
with these issues. 

We have tried to concentrate on the area of synthetic characters, which relative to the traditional research our 
agent group does is a new area. But as a secondary interest we are also hoping that AgentLink will help 
breathe some fresh life into the basic area of IR in open systems, which is sadly being overlooked. Agent IR 
research is currently integrating new agent technology with old IR technology and in most cases this is 
insufficient, and a new holistic approach needs to be taken.   

Typically synthetic agents are associated with the entertainment industry; this paper has tried to highlight 
some of the applications that they have in 'serious' domains. As more autonomous agents are gradually 
adopted by the real world, performing an increasing number of tasks in the background transparent to the 
user, control will be transferred from direct to indirect manipulation. Users will demand a new paradigm of 
social interaction with new and upcoming agent applications. It is also felt that there is a social stigma 
attached to synthetic agents within the agent community that needs to be addressed. The Salford University 
AgentLink node would like to set up a working group to help nurture the research and development of 
synthetic agents within the agent community. Depending on the decided nature of this initiative, it could take 
part within the 'Intelligent Information Agent SIG' or as a separate body working within AgentLink.


Aylett, R., et al. 1998 Virtual Teletubbies: reapplying a robot architecture to virtual agents, To appear in the 
Virtual Reality SIG, Exeter. Centre for Virtual Environments, University of Salford.

Doyle, P., and Hayes-Roth, B. 1997. Agents in Annotated Worlds, Technical report KSL 97-09, Knowledge 
Systems Laboratory, Department of Computer Science, Stanford University.

Elliot, C., and Brzezinski, J. 1998. Autonomous Agents as Synthetic Characters, AI magazine, AAAI Press, 
19(2): Summer 1998,13-30

Laurel, B. 1997, Interface agents: Metaphors with character. In Software Agents, ed J.M. Bradshaw. Menlo 
Park, Calif.: AAAI Press

Rickel, J., and Johnson, L. 1997. Integrating Pedagogical Capabilities in a Virtual Environment Agent. In 
Proceedings of the First International Conference on Autonomous Agents, 30-88. New York: Association of 
Computing Machinery