Magnus Boman

Norms in artificial decision making

A method for forcing norms onto individual agents in a multi-agent system is presented. The agents under study are supersoft agents: autonomous artificial agents programmed to represent and evaluate vague and imprecise information. Agents are further assumed to act in accordance with advice obtained from a normative decision module, with which they can communicate. Norms act as global constraints on the evaluations performed in the decision module and hence no action that violates a norm will be suggested to any agent. Further constraints on action may then be added locally. The method strives to characterise real-time decision making in agents, in the presence of risk and uncertainty.

Distributed monitoring and control of office buildings by embedded agents

We describe a decentralized system consisting of a collection of software agents that monitor and control an office building. It uses the existing power lines for communication between the agents and the electrical devices of the building, such as sensors and actuators for lights and heating. The objectives are both energy saving and increasing customer satisfaction through value added services. Results of qualitative simulations and quantitative analysis based on thermodynamical modeling of an office building and its staff using four different approaches for controlling the building indicate that significant energy savings can result from using the agent-based approach. The evaluation also shows that customer satisfaction can be increased in most situations. The approach here presented makes it possible to control the trade-off between energy saving and customer satisfaction (and actually increase both, in comparison with current approaches).

Implementing an agent trade server

An experimental server for stock trading autonomous agents is presented and made available, together with an agent shell for swift development. The server, written in Java, was implemented as proof-of-concept for an agent trade server for a real financial exchange.

General risk constraints

The risk evaluation process is integrated with procedures for handling vague and numerically imprecise probabilities and utilities. A body of empirical evidence has shown that many managers would welcome new ways of highlighting catastrophic consequences, as well as means to evaluate decision situations involving high risks. When events occur frequently and their consequences are not severe, it is relatively simple to calculate the risk exposure of an organization, as well as a reasonable premium when an insurance transaction is made, relying on variations of the principle of maximizing the expected utility. When, on the other hand, the frequency of damages is low, the situation is considerably more difficult, especially if catastrophic events may occur. When the quality of estimates is poor, e.g. when evaluating low-probability/highconsequence risks, the customary use of quantitative rules together with unrealistically precise data could be harmful as well as misleading. We point out some problematic features of evaluations performed using utility theory and criticize the demand for precise data in situations where none is available. As an alternative to traditional models, we suggest a method that allows for interval statements and comparisons, which does not require the use of numerically precise statements of probability, cost, or utility in a general sense. In order to attain a reasonable level of security, and because it has been shown that managers tend to focus on large negative losses, it is argued that a risk constraint should be imposed on the analysis. The strategies are evaluated relative to a set of such constraints considering how risky the strategies are. The shortcomings of utility theory can in part be compensated for by the introduction of risk constraints.

Saving Energy and Providing Value Added Services in Intelligent Buildings: A MAS Approach

In a de-regulated market the distribution utilities will compete with added value for the customer in addition to the delivery of energy. We describe a system consisting of a collection of software agents that monitor and control an office building. It uses the existing power lines for communication between the agents and the electrical devices of the building, such as sensors and actuators for lights, heating, and ventilation. The objectives are both energy saving and increasing customer satisfaction through value added services. Results of qualitative simulations and quantitative analysis based on thermodynamical modeling of an office building and its staff using four different approaches for controlling the building indicate that significant energy savings, up to 40 per cent, can be achieved by using the agent-based approach. The evaluation also shows that customer satisfaction can be increased in most situations. In fact, this approach makes it possible to control the trade-off between energy saving and customer satisfaction (and actually increase both in comparison with current approaches).

A multi-agent system for controlling intelligent buildings

Gives a high-level description of a project aimed at investigating the usefulness of the agent metaphor and the notion of multi-agent systems for the design of control systems for intelligent buildings. The use of the agent approach was initially motivated by the close mapping that it offered between the entities of the application domain and the entities of the software. The concurrent non-deterministic nature of the activities inside the building was another factor that led to the development of concurrent autonomous entities. We argue that the suggested multi-agent system architecture can be easily adapted to almost any building. Also, the implementation of the system allows for dynamic re-configuration of the agents, without disrupting the operation of the system. Finally, simulation results indicate that the approach is viable and that considerable energy savings are possible, while at the same time providing added value for the customer. Also, the approach enables a much more fine-grained control of the tradeoff between energy saving and customer satisfaction than current approaches do.

Imposing security constraints on agent-based decision support

The principle of maximising the expected utility has had a large influence on agent-based decision support. Even though this principle is often useful when evaluating a decision situation, it is not always the most rational decision rule and other candidates are worth considering. A decision making agent may want, for example, to exclude particular strategies which, in some sense, are too risky with respect to specific thresholds. A theory is presented for situations where a decision making agent, human or machine, has to choose between a finite set of strategies having access to a finite set of autonomous agents reporting their opinions on the strategies. The approach considers a decision problem with respect to the contents and the credibilities of the reports, and the main emphasis is on how to perform analyses in decision situations where the available information is vague or numerically imprecise.

Multi-agent systems, time geography, and microsimulations

In this chapter we consider the role virtual conferencing has to play in realising a successful Inhabited Information Space (IIS). For any IIS to be successful it needs to weave together many different constituent elements and present these in a coherent and seamless manner. For example, Maher et al (2000) describe how many different components are used together to create a virtual design studio for architectural collaboration. For the IIS to function all the elements must work both individually and collectively. Communication is one of the basic building blocks for an IIS, and can be in many modes across many media. Approaches to virtual conferencing offer support for communication across a number of media and can be utilised in an IIS. These approaches are also starting to additionally offer support for collaboration. By providing an introduction to and overview of various possibilities for virtual conferencing we aim to show how these solutions can provide the required and appropriate support for communication and collaboration between inhabitants in a shared information space. Of course virtual conferencing solutions exist at many levels of sophistication and fidelity. Communication media can range from text through 3d graphics to video representations. The aim of this chapter is to present these many and varied possibilities, drawing on the experience of the author as well as insights into the past, present and future. In this way it is possible to see how diverse a range of IISes can make use of virtual conferencing functionality.

FROM LOCAL ASSESSMENTS TO GLOBAL RATIONALITY

We present a theory and a tool for the treatment of problems arising when a decision making agent faces a situation involving a choice between a finite set of strategies, having access to a finite set of autonomous agents reporting their opinions. Each of these agents may itself be a decision making agent, and the theory is independent of whether there is a specific coordinating agent or not. Any decision making agent is allowed to assign different credibilities to the statements made by the other autonomous agents. The theory admits the representation of vague and numerically imprecise information, and the evaluation results in a set of admissible strategies by using criteria conforming to classical statistical decision theory. The admissible strategies can be further investigated with respect to strength and also with respect to the range of values that makes them admissible.

Parrondo strategies for artificial traders

On markets with receding prices, artificial noise traders may consider alternatives to buy-and-hold. By simulating variations of the Parrondo strategy, using real data from the Swedish stock market, we produce first indications of a buy-low-sell-random Parrondo variation outperforming buy-and-hold. Subject to our assumptions, buy-low-sell-random also outperforms the traditional value and trend investor strategies. We measure the success of the Parrondo variations not only through their performance compared to other kinds of strategies, but also relative to varying levels of perfect information, received through messages within a multi-agent system of artificial traders.