"No battle plan survives contact with the enemy." Military forces are constantly faced with unexpected situations and must choose the best course of action despite great time pressure, incomplete and uncertain information, and weighty consequences of their actions. They must collectively understand the problems they face, identify alternative courses of action, select one or more courses of action, and implement it. This requires gathering information needed to understand the problem and identify potential alternative solutions. Obtaining the needed information may require access to global and local information seamlessly across heterogeneous networks of uncertain reliability (Information Access Hard Problem). In a time-critical situation they must not be swamped with information of little relevance; they need important and relevant information delivered in an efficient and effective form (Information Load Management Hard Problem). They must obtain information from many sources, including coalition members, assess the quality of the information they receive, and decide whether it is actionable (Quality of Information Hard Problem). When deciding which actions to take, they must evaluate the assets available and the relative importance of other demands on those assets (Finite Assets Hard Problem). They must also ensure that the problem is understood by everyone involved in defining its solution and that the courses of action are understood by everyone involved in executing or supporting them (Shared Understanding Hard Problem).

The information systems that support military problem solving must function in a heterogeneous and unreliable environment. Software agents provide the flexibility required in these situations. Like people, software agents can learn and are goal directed, distributed, collaborative, reactive, and adaptive. They can distribute themselves across heterogeneous networks to reduce the consequences of uncertain reliability. They can process and integrate many sources of information, assist in the assessment of information quality, manage information delivery with priorities for relevance, and ensure that information is disseminated to those who need it. The challenge is to develop software agent technologies that recognize and understand the problems faced in the military context in sufficient detail and with sufficient accuracy to provide trustworthy support to military problem solving.

The capabilities of software agents to learn and reason will be of little value if the agents lack knowledge about the military context, including its political, social, technical, and cultural aspects. The future networked environment will allow agents to retrieve considerable information about context, including the current state of the network itself. Research conducted in ITA Project 12 (Network-enabled Shared Understanding) will define the technical infrastructure required for software agents to interpret semantically tagged information about military missions and plans, the communications and sensor networks, and military resources. These elements define the environment in which missions are planned and executed.

A greater challenge that we will tackle is how to acquire knowledge about events as they occur during mission planning and execution. We cannot expect military personnel to describe and explain everything that occurs in some sort of diary for the benefit of software agents. Instead, agents must interpret the actions and communications of both military and non-military personnel. Agents will obtain information about their actions and communication through their interactions with networked information systems and through networks of sensors. Agents will construct models or interpretations of these actions and communications. We do not propose to solve the problem of human speech understanding, but we will investigate ways that statistically-based heuristics can interpret human discourse. We will also develop formal models (based on argumentation) of the communication required when executing a mission plan. These models define when and what information should be provided to team members and when events are diverging sufficiently from the plan to suggest the need for re-planning. Of course, agents (like humans) may misinterpret human actions or construct alternative uncertain interpretations, requiring human guidance. People are accustomed to providing guidance about event interpretation through conversations, and simple ways must be found for software agents to expose their interpretations and receive human guidance with minimal disruption to human activity.

Cultural and subtle linguistic differences can be a major source of misinterpretations. People from different cultures (including military and non-military cultures) develop different habits or patterns of communication even when they speak essentially the same language, and communication between cultures offers many opportunities for miscommunication. This project will investigate ways to reduce the frequency of miscommunications. We will empirically investigate examples and causes of miscommunication across culture in military contexts. These investigations will include surveys and interviews of military personnel. Guided by this information and analyses, we will seek ways that agents can help people avoid or correct miscommunications.

Automated information gathering from sources across the network (and, in part, about the network) is an essential capability of a software agent infrastructure to support military problem solving. This project will investigate ways to gather information automatically and to disseminate it to both humans and agents that need it. Providing information to people requires investigating methods of presentation that communicate to people efficiently and effectively and query methods that enable people to describe easily and clearly the information they require. The project will explore both reactive dissemination, in which people request and receive information, and proactive dissemination, in which agents anticipate human information needs based on their understanding of the problem solving context. We will investigate ways to adaptively adjust the amount and presentation format of information depending on the estimated time and cognitive capacity of the human for making effective decisions. This research will leverage Project 12’s progress in investigating Network-enabled Shared Understanding.

Agents can do more than simply provide information. They can also offer advice based on an analysis of the situation and alternative solutions. This advice will not be useful, however, unless people accept it. People must be able to inquire about agents’ advice and be reassured that the agents used all the available information, considered all reasonable alternatives, and weighed the alternatives appropriately. This project will investigate these issues in the domain of policy conflicts. We will develop formal techniques for policy reconciliation and conflict resolution, and we will develop capabilities for explaining conflict sources and recommendations regarding their resolution.

The technical objectives of this project require a focus on human capabilities and performance during mission planning and execution. Mission participants are expected to include both military and civilian personnel from multiple networked nations and organizations. Mission success depends on their ability to plan and execute together in stressful situations despite differences in goals, objectives, policies, procedures, resources, infrastructure, language, customs, training, and culture. The project’s objectives are to investigate ways to understand the problems mission participants must solve with minimal explanation from humans, anticipate potential miscommunications due to differences in participants’ backgrounds, and intervene in ways that are useful and acceptable to people. All three of these objectives require grounding in empirical investigations to inform the research and to validate the results.

The project is divided into three interrelated tasks. The first task is Agent Assistance for Information Management led by Katia Sycara from CMU. This task will investigate ways to recognize and anticipate the information needs of military personnel while planning and executing military missions, gather the needed information from all available networked resources, and adaptively disseminate it. When disseminating the information, agents will consider factors such as the sources of information, coalition policies regarding information sharing, the intended participants, the intended purpose of the information, and the intended recipient. Agents will attempt to deliver just the right information to the right persons at the right time. When agents offer information or recommendations, people will be able to inquire about the sources of the information or the reasons for the recommendations. We anticipate that this capability will allow the development of trust in agents’ contributions.

The second task is Interpreting Team Collaboration led by Peter Wagget from IBM UK. This task will investigate ways to interpret the communication and actions of teams engaged in simulated missions. When people work together they act and speak in ways intended to ensure that all team members understand what they are doing and what is happening around them. Networked sensors could capture many of their actions and communications. This task is investigating ways to interpret the captured actions and communications, relate them to a team’s plans, and construct a dynamic model of the team’s context. This model can then be used by the agents developed by Task 1 to anticipate the team’s need for assistance or to disambiguate human requests for assistance.

The third task is Context and Misunderstanding led by Iya Whiteley from SEA. This task will investigate potential sources of miscommunication that may arise in coalition operations (especially when communicating remotely over networks) and methods or tools for avoiding miscommunication. The task will conduct surveys and interviews with military personnel engaged in multinational field experiments to identify their experiences with miscommunication and the sources of those miscommunications.

• Task 1, Project 10: Agent Assistance for Information Management
• Task 2, Project 10: Interpreting Team Collaboration
• Task 3, Project 10: Context and Misunderstanding

The U.S. and UK anticipate that future military actions will involve coalitions of many nations and require integrated and coordinated deployment of political, economic, cultural, and military action. Military forces will need to collaborate with people from diverse backgrounds, both in terms of their national culture and their military experience. The Project 10 research plan addresses these needs in several ways. We will study the miscommunications that occur among people from different backgrounds and identify ways to reduce or repair miscommunications. We will also investigate technologies that help people gather and integrate information from diverse sources, that recognize the human problem-solving context, that deliver information to them when they need it and in an understandable format, and that recommend and explain solutions to problems.

The planned research defines and addresses the hard problem of problem solving in a networked environment, which combines five of the ITA hard problems. Military forces are constantly faced with unexpected situations and must choose the best course of action despite great time pressure, incomplete and uncertain information, and weighty consequences of their actions. They must collectively understand the problems they face, identify alternative courses of action, select one or more courses of action, and implement it. This requires gathering information needed to understand the problem and identify potential alternative solutions. Obtaining the needed information may require access to global and local information seamlessly across heterogeneous networks of uncertain reliability (Information Access Hard Problem). In a time-critical situation they must not be swamped with information of little relevance; they need important and relevant information delivered in an efficient and effective form (Information Load Management Hard Problem). They must obtain information from many sources, including coalition members, assess the quality of the information they receive, and decide whether it is actionable (Quality of Information Hard Problem). When deciding which actions to take, they must evaluate the assets available and the relative importance of other demands on those assets (Finite Assets Hard Problem). They must also ensure that the problem is understood by everyone involved in defining its solution and that the courses of action are understood by everyone involved in executing or supporting them (Shared Understanding Hard Problem). Below we describe how this project addresses some of these hard problems.

Information Load Management Hard Problem

The planned research directly addresses the key elements of this hard problem. First information agents will support (stressed) users by alerting them to critical information so the users do not suffer cognitive load when trying to find and understand critical information in the vast amount of available information. Second, the agents will take into consideration the cognitive state of the user, including activity phase, intent and cognitive load, when making decisions about information retrieval and presentation. Thirdly, by providing unobtrusive proactive information assistance, the agents free the user from having to deliberate about what information he currently needs and then inquire from the assistant. Fourth, by providing automated information dissemination based on the state of the team problem solving and progress in execution, the agents free the users from having to keep track of which team members need specific information, a cognitive activity that incurs high workload.

Information load management can also be facilitated by avoiding miscommunications. Information is not communicated, misunderstood, or delayed as a result of miscommunication, increasing the information load management requirements of participants. By understanding the contexts under which miscommunication is most likely to occur, we can suggest ways in which training or changes in processes can minimize miscommunication and, consequently, the information load, of the war fighter, the commander, or the planner.

Quality of Information Hard Problem

The Task 1 research investigating information quality, information reputation, and their use in constructing query execution plans directly addresses the quality of information hard problem. Information quality includes its fidelity to ground truth, its utility in task performance, and its fulfillment of user needs. We will develop agents that access, fuse, filter and disseminate information from different information sources. This software will collect user assessments of how well the agent information matches users’ perceived needs. The retrieval algorithms will consider the reputation (credibility and trustworthiness) of information sources. Retrieval algorithms that correctly estimate resource credibility will yield higher quality information.

Miscommunications constitute, by definition, reductions in the quality of information. We propose to develop ways of assessing the impact of different types of miscommunication in different contexts, which will lead naturally to metrics for the utility of information transmitted via spoken and written language in these situations.

Shared Understanding Hard Problem

This research contributes to shared understanding among coalition members in several ways First, automated communication of information to team members can contribute to better shared understanding because recipients have the information needed even in situations with high workloads in which people may forget to convey needed information to a team mate. Second, agents provide users with greater cognitive capacity to assimilate information by adapting the information presentation to the users’ cognitive state. Third, the research proposes models and algorithms for conflict identification, reconciliation and resolution of a broad range of policies that may vary between coalition partners and cause conflicts. Agents can intervene in both planning and plan enactment to advise team members of conflicts and propose possible reconciliations. The agent advice makes the humans aware of these discrepancies thus enhancing their situational understanding.

During a coalition mission, shared understanding involves knowing what coalition partners are doing that is relevant to the mission and plan. Today this knowledge is obtained by people communicating with other people, and the content of these messages is unknown to computers unless someone enters the information in an application. Lacking this information, computers cannot provide integrated and comprehensible views of mission progress that elucidate the contributions of all partners. This research automatically interprets the actions and communication of team members, generating higher level descriptions of their activities. By interpreting this communication, computers can participate in this common understanding and make it available to higher levels of the organizational structure.

Miscommunications directly lead to divergence in situation awareness and the shared understanding that serves as a basis for collaborations across coalitions. By understanding the different types of miscommunication that occur in cross-cultural situations, and their contexts and impact, we can help minimize these, thus improving shared situation awareness and mutual understanding.