Electric Elves: Immersing an agent organisation in a human organisation (AAAI2000)

This paper firstly introduces the term of agentization in the organization. The idea is using software agents (agentization) to support for the large-scale human organizations. For example, dynamic teaming of such heterogeneous agents will enable organizations to act coherently, to robustly attain their mission goals, to react swiftly to crises, and to dynamically adapt to events.

Advances could potentially assist all organizations, including the military, civilian disaster response organizations, corporations, and universities and research institutions.

Agents here act as proxies for each person within an organisation, presented on behalf of the people or resources they represent.

Applying agent technology in human organization provides these following challenging:

·         A key research question of adjustable autonomy. It means agents acting as proxies for people must automatically adjust their own autonomy such as avoiding critical errors or possibly by letting people make important decisions.

·         The agent system must be up and running 24/7 as human organization operates continually over time.

·         People have a wide and rich variety of capabilities, interests and preferences, etc, as well as their associated tasks. Therefore agents acting as proxies must represent and reason with such capabilities and interests. It is difficult to arrange a powerful matchmaking capability to match people with similar interests.

·         Human organizations are often huge. This mean we need to scale-up in the number of agents compared against typical multiagent systems in current operation.

Electric Elves project investigates the above research issues and the impact of agentization on human organization in general, using their own Intelligent System Division of USC/ISI as a testbed. The working prototypes is about 10 agent proxies running continuously and automatically manages some tasks such as selecting teams for giving a demonstration, scheduling and rescheduling meetings, monitoring the location of users.

The project used Teamcore, a domain-independent, decentralised, teamwork-based integration architecture to construct the organizations of software agents. The Teamcore proxy considers only the team-level activities. It additionally used the Friday agent to serve the user’s interests in dealing with Teamcore proxy. The Friday agent is to ensure proper consideration of each person’s individual preferences.  The Friday uses C4.5 to learn a decision tree from user’s feedbacks in training mode. Then Friday applies its learned knowledge in circumstances to act autonomously in actual use mode.

The Teamcore used the capabilities matcher and the interest matcher combined with the statistical information retrieval (IR) techniques with knowledge-based matchmaking based on logical inference to form the team.

The Electric Elves project has ran the agent proxies to coordinate real meeting schedules for a group of five agentized people. They found it smoothly and helpful, except the limitation of using C4.5 for agent’s learning paths.

This paper has been written in 2000, presented a vision for a complex, heterogeneous agent organization that automates numerous tasks central to human organizations. The Teamcore’s infrastructural teamwork could potentially facilitate such agents to work together in teams. The paper has not yet attempted to apply the agentization of the large-scale human organisations. They said they would do it but I doubt about that.

The design of everyday things - chapter 1

Most accidents are attributed to human error, but in almost all cases the human error was the direct result of poor design. This chapter introduces about the principles of design such as affordances, feedback and visibility. The fundamental principles of designing for people are to provide a good conceptual model and to make things visible. A good conceptual model allows us to predict the effects of our actions.

The right things have to be visible in design to indicate what parts operate and how, to indicate how the user is to interact with the device. Visibility indicates the mapping between intended actions and actual operations. Visibility indicates crucial distinctions so that you can tell salt and pepper shakers apart, for example.

How do people cope with studying so much visual perception. Part of the answer lies in the way of mind works, in the psychology of human thought and cognition.

Quick review for Forming Coalition chapter from Michael Wooldridge book

This chapter focuses on forming coalitions analysis, based on the realm of cooperative game theory. The cooperative game states the situation that there are a number of agents with their own utility. Each agent can work their own to earn a certain amount of money. But also they can work with others, by cooperating, to generate some surplus income, over and above the amount that could be earned by individuals. Agents can form as a team to obtain a certain utility, which can be shared among coalition members. The game itself is completely silent about how this utility should be distributed among coalition members. In addition, the agreement of how to divide the ‘pay-check’ is done by coalition members.

Three key issues have been raised: coalition structure generation, solving the optimisation problem of each coalition to maximise social welfare, and how to divide the value of the solution for each coalition as fair as possible.

Some representative solutions have been reviewed for each key issue. However, this is NP-Hard problem as the complexity is exponential in the number of agents. It is likely impossible to give an optimal solution if the number of agents is high volume.

Thinking: just a simple real scenario can raise lots of key issuesL.

A blog for PhD life

A big day stepped into this type of blog revolution. This could be a summary for whatever I would have done for my PhD life in IAM group.

Please help and support.