Autonomous Adaptive Information Systems


The infosphere – as envisioned by the DoD in such programs/documents as Joint Battlespace Infosphere, (JBI), Network Centric Warfare, and Joint Vision 2020 – is an amazingly complex network of information producers, processors, and users. In JBI, a fuselet is a process that is supposed to take information provided to it via publish/subscribe infrastructure and produce new, processed information for a specific user. Unfortunately, the probability of specific information and/or information sources being available, predictable, and timely is unknown. While the infosphere is tasked with providing persistent information objects, the infosphere, due to its lack of control over information sources, cannot guarantee that the current information is most recent or best available. Thus, these systems are, by nature of the infosphere, susceptible to loss of individual information sources, which can significantly impair the ability of the system to accomplish its goal. Most systems are currently designed to work with a limited set of information source/type configurations so that even when the information it needs to execute correctly is available, it is limited in reaching its goal by its own rigid information configuration requirements. To overcome this, we need to develop systems that can adapt to a dynamic information environment. Specifically, we propose to develop the theories, techniques, and tools to allow systems to adapt to the changing information environment via system reorganization.


Our proposed solution is based on the concept of a cooperative multiagent system, or a multiagent team. The team consists of agents playing the roles of information producers, information sources and information processors. Information processor agents understand how to fuse particular types of information and raw data to create new information that is usable by specific users such as field commanders; they are roughly equivalent to a JBI fuselet except they do not know how to get their raw data or information. Information producer agents represent the actual sources of raw data and information in the infosphere. Information source agents understand the information they are required to generate and how to interface with information producers to obtain the necessary raw data. However, an information source agent does not necessarily know all the ways that a particular type of information can be produced. Therefore, a particular system many employ many information source agents to generate the same type of information. Some may generate the information more accurately while others may generate it more quickly or even may be able to derive it from different sources. The key to the process is being able to pick and choose the appropriate information source at the appropriate time for the right task. Thus the assignment can be equated to choosing the right multiagent organization for a particular task (i.e., reorganization). Additionally, if an information source is lost during the process, the team must be able to reorganize in the middle of its operation.

This research proposes a layered approach to investigating the necessary multiagent reorganizational capability.

  1. A formal model of team goals and organization will be developed. This model will be based on team goals and valid organizational structures (the required roles and relationships) and will form the foundation upon which all reasoning and decision-making will take place.
  2. Appropriate reasoning techniques to determine (1) when reorganization is necessary and (2) how to go about performing the reorganization will be developed.
  3. The theory will be made practical by integrating the organizational model and reasoning techniques into existing software engineering methods, such as Multiagent Systems Engineering (MaSE), for developing multiagent and multiagent systems.

Anticipated Results

The results of this research will be evaluated both theoretically and experimentally. The theory of reorganization based on goals and roles will be shown to be both sound and complete. Specifically, it will be shown that, for a given system, the theory will produce an organization that is capable of reaching its goal, if such an organization exists. In the experimental evaluation, the effectiveness of this approach will be demonstrated by conducting experiments with two exemplar systems; the first will be a traditional “non-reorganizing” multiagent team while the second will incorporate the proposed organizational models and reasoning techniques. An example application will be chosen from appropriate infospheric applications.

If successful, the impacts of this research will be significant. Not only will it provide a theory for reorganization of information based multiagent teams, but it will also provide a practical methodology for employing that theory in real infospheric applications. The results will also provide a foundation for more extensive research into other forms of system reorganization in the other areas such as teams of uninhabited air or ground vehicles. While most existing research into adaptive multiagent teams has been limited or fairly shallow, this research will provide the foundation for a comprehensive theory of reorganization based on team goals and will provide a practical guide to implementation via its grounding in existing software development methods.

Related Publications


This research is sponsored by AFOSR.