Control of Agent Based Systems Slide 1 Good Afternoon and welcome to the Control of Agent Based Systems briefing at DARPATech 99. My name is Jim Hendler and I am Program Manager within the Information Systems Office. Slide 2 Although the number and availability of information sources are increasing, current military command and control systems including those supplemented with commercial off-the-shelf technology require manual manipulation and user-specification of all details. Once accessed, information must be filtered in the context of the situation. Current systems lack the ability to get contextual information or use it to automate filtering. Furthermore, as dependence on information technology increases, so too does the potential benefit to adversaries capable of damaging, disrupting, or polluting our information systems. The Agent technology promises to lead to a revolutionary new model of computing beyond mainframe and client-server architectures. This new model will consist of entities that are autonomous, adaptable, and cooperative. It will result in computing systems that are dramatically easier to use, as well as, systems that are self-configuring, self healing, and evolvable. Slide 3 There are many definitions of what an agent is depending on its context. In our program, we will confine the definition to these four capabilities. 1. Embedded in, and "aware" of, an environment (Communicative) 2. Dynamic in its behaviors (Autonomous) 3. User enabled/steered, but "empowered" to act for user (Capable) 4. Able to improve its behavior over time (Adaptive) Slide 4 In terms of where we are and where we would like to be, there are technical gaps which need to be filled. For example, we are now able to search the internet for information on just about any topic. However, the tedious tasks of filtering, deciphering, and translating the sea of information provides little use to those of us who needs specific and precise information. The ideal future scenario would be for the user to tell the computer some desirable end-goal and the computer would act as an "agent" to provide that information or means to it. We need to provide customized software solutions that can be rapidly developed at low cost and execute on readily available hardware without overloading conventional processors. Slide 5 Agents or agent systems produced by different developers cannot cooperate in any meaningful way. Cooperation among agents is critical to building powerful applications to support military capability because without cooperation, each new task must be handled by a monolithic agent designed for it. However, multiple agents in an uncertain environment can lead to a system that is in disarray and destructive. A number of alternative control strategies have been suggested, but further research is required to fully elucidate characteristics of these strategies. Slide 6 The challenge of the CoABS program is to be able to prove operational feasibility of such multi-agent systems. Slide 7 To increase the military user's productivity and, by extension, our military capability, the next generation of software must be able to automatically accept abstract tasking, get needed information, decide how to solve simple problems, help the user solve difficult problems, and take action on the user's behalf. Arising from research in distributed artificial intelligence, intelligent agents are software modules designed to provide these capabilities. For the military, intelligent agents are critical force multipliers that free personnel from having to do simple tasks which can be automated and assist personnel with difficult tasks. As our military forces are drawn down, intelligent agents are becoming increasingly important for retaining our ability to meet crises effectively. Slide 8 Military TIEs stress integration OOTW Neo-Challenge Problems defined Ballistic and Theatre Missile Defense TMD demonstrated at FBE-E Coalition Force Interoperability International Workshop on Coalition Planning 5/99 Scientific TIEs stress scaling Negotiation Experiments 1st results favor auctions Mathematical Analyses New results for agent mobility Control Scheme Comparison Analysis of time/ Experiments designed Slide 9 Current experiments cluster near the axes on the orthogonal planes of 3 agent factors. They are number of agents, number of agent types, and agent complexity (#of actions an agent can take). We currently have to sacrifice one factor in order to improve in another. Slide 10 However, CoABS will aim to prove that it is possible to increase all three factors with new type of software technology. Slide 11 The CoABS Grid will provide a basis and a test bed for technologies developed under this program. The Grid could provide various services including communication protocol, brokerage services, semantics, and support for new systems development. Slide 12 The CoABS program is providing funding to the research community to cooperate in developing the underlying science for agent technologies. CoABS will develop software components and techniques to provide further development and realization of a true agent society. Slide 13 To bring the agent technology to realization, we will have to develop a fundamentally new kind of software technology and methodology. The workshops we have planned will bring new ideas and provide opportunity for the leading researchers to share innovative solutions. Slide 14 The DARPA CoABS program is focusing on critical military need to search, filter, manipulate, and present information in an autonomous and timely manner. The research will culminate in interoperability of legacy systems as well as scaling of multi-agent systems. In addition, through iterative development and evaluation, the CoABS research will provide necessary software tools and techniques to meet this critical need.