From: AAAI Technical Report FS-01-05. Compilation copyright © 2001, AAAI (www.aaai.org). All rights reserved. Enhancing Crew Coordination with Intent Inference Benjamin Bell and Daniel McFarlane Lockheed Martin Advanced Technology Laboratories 1 Federal Street, Camden, NJ 08102 {bbell,dmcfarla}@atl.lmco.com Abstract Users of complex automation sometimes make mistakes. Vendors and supervisors often place accountability for these mistakes on the users themselves and their limited understanding of how these systems are “supposed” to be used (a perspective that influences how operators are trained and fosters efforts to design systems that explain their actions). Seeking to give users a better understanding of their systems, however, sidesteps the critical issue of how we ought to get computers to shoulder the responsibility of understanding humans. The machine is the piece of this equation that is "supposed" to conform. Our approach has the potential to make what people do naturally be the right thing. This work derives somewhat from related research in Intent Inference, which has focused on representing the tasks and actions of a user so that a system can maintain a representation of the user’s intent and can tailor its decision support accordingly. There is a growing need to extend the notion of intent inference to multi-operator venues, as systems are targeted increasingly at collaborative environments and operators are called upon to perform in multiple roles. In this paper we describe an approach to Crew Intent Inference that relies on models representing the goals and actions of individual operators and the overall team. We focus on the potential for intent inference to enhance coordination and present example scenarios highlighting the utility of intent inference in a multi-operator domain. Introduction Crews are often simultaneously responsible for several critical interdependent task activities. They also usually have a suite of complex automated support systems that provide some assistance but must themselves be managed or supervised by the crew. Failure to coordinate these activity dependencies or errors in supervising the automation that coordinates these dependencies can have serious consequences. We follow Malone & Crowston’s definition of coordination as “managing dependencies between activities” (1994). This definition guides an approach to solving coordination problems based on identifying the dependencies in a crew work context and devising ways to manage the coordination processes required to accommodate them. In this paper we describe an approach that promotes more effective crew coordination by applying our ongoing work in team intent inference. Importance of Coordination People intuitively employ strategies during coordinated activity, whether in routine every-day tasks or highly complex technical procedures. Coordination among operators in high-risk engineering environments is an area of intensive study, because of the enormous potential consequences of lapses in coordination (e.g., for airline flightcrews, power plant crews). In this paper we focus on a mission aircrew, such as would be found on aircraft that perform command and control (C2), airborne early warning (AEW), or intelligence, surveillance and reconnaissance (ISR) missions. Coordination of military teams has received considerable attention from government research sponsors, notably in the area of air operations (see Salas, Bowers & Cannon-Bowers, 1995, for a summary of ten years of DoD-sponsored research in aircrew coordination). Safety is of great concern in aviation because crew-related error is believed to be a contributing factor in at least 69% of civil aviation accidents (International Civil Aviation Organization, 1989) and perhaps as many as 80% of civil aviation accidents (Sarter & Alexander, 2000). Military operations show similar patterns: during FY 1994–1995, human error was a factor in 71% of Air Force, 76% of Army, and 74% of Navy/USMC aircraft mishaps (General Accounting Office, 1996). Besides safety, there is a critical and obvious link between crew coordination and effective military operations (Barker, et al., 1997; LaJoie & Sterling, 1999; US Army, 1995). Structure and Breakdown of Coordination Research into coordination has revealed patterns of coordination and some of the conditions under which Copyright © 2001, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. coordination breakdowns are more likely. Activity can be coordinated by a centralized supervisor, by a team of cooperating peers, or by a distributed supervisory team (Palmer, et al., 1995; Wilkins, et al., 1998). Studies have identified patterns that subjects exhibit while coordinating activity (Decker, 1998; Malone & Crowston, 1990, 1994; Patterson, et al., 1998; Schmidt, 1994) and have observed subjects attenuating their coordination activity in response to the demands of the current situation (Schutte & Trujillo, 1996). Findings from these studies can help identify specific coordination activities that automation systems should detect and support. For instance, Patterson et al. (1998) identified patterns that give rise to what they term “coordination surprises” that include agents working autonomously, gaps in agents’ understanding of how their partners work, and weak feedback about current activities and planned behavior. Factors that contribute to coordination failures frequently involve an increased workload or need for accelerated activity and more rapid decision-making. Often, a chain of events escalates coordination difficulties. A typical instance is when a process being monitored exhibits some anomalous behavior and the automation system begins generating alerts. The malfunction will often induce off-nominal states throughout other, related subsystems, and the operator monitoring these processes will be confronted by a growing queue of alerts. Demands on the operator’s cognitive activity increase as the problem cascades, and paradoxically, reporting and resolving these alerts introduces a heightened demand for coordination with his team members. These system-generated alerts are intended to deliver critical "as-it-happens" information about process status that the user needs to make good decisions about how to react to the problem. However, from the user's perspective each alert is also an interruption, and people have cognitive limitations that restrict their capacity to handle interruptions (McFarlane, 1997, 1999; Latorella, 1996, 1998). The need for more coordination, therefore, presents itself precisely when the operator is least able to sacrifice any additional attention. This process has been formalized as the Escalation Principle (Woods & Patterson, 2001); the authors further observe that both the cascading anomalies and the escalation of cognitive demands are dynamic processes, making it difficult for operators and supervisors to stay ahead of the situation. Impact of automation on crew coordination Patterns of coordination lapses such as the Escalation Principle are likely to become more ubiquitous in aircrew coordination through the effects of five related trends. First, improved sensor packages are bringing more information onboard, and better communications links are pulling in data collected from other sources. As a result, operators are facing increased volumes of information. Second, aircraft roles are becoming more multi-mission and operator duties are as a result becoming more diverse and interdependent. Third, military and supporting agencies are emphasizing joint and coalition operations, which incur more demanding requirements for coordination among personnel from different services and nations. Fourth, military planners are focused on “network-centric warfare”, which will generate both a more distributed team structure and an accelerated operational tempo. Fifth, computer automation in systems continues to grow in both ubiquity and complexity. This automation requires intermittent human supervision and its presence results in a much higher rate of computer-initiated interruptions of operators. New systems, developed to enhance mission effectiveness (e.g., more powerful sensors, improved communications links), often carry implications for crew coordination but are seldom designed to support coordinated activity (Qureshi, 2000). The increased use of automation has imposed new dynamics on how crewmembers work together. Automation has changed the nature of crew communication in subtle ways (Bowers, et al., 1995). For instance, automated systems that capture the attention of an operator may compromise established coordination processes (Sheehan, 1995). The use of automation may adversely affect crew coordination, possibly leading to unsafe conditions (Diehl, 1991; Eldredge, Mangold & Dodd, 1992; Funk, Lyall & Niemczyk, 1997; Wiener, 1989; Wiener, 1993; Wise, et al, 1994). Designing systems that support coordination A tactical picture can be radically altered in minutes or even seconds. The level of coordination that an operator must sustain under an accelerated tempo is likely to increase with the amount of information he must manage (leaving less time and attention to devote to coordination). Automated systems are not responsive to such changes in workload, and in fact frequently worsen the operator’s situation by generating a rapid stream of alerts which may fail to communicate their relevance, priority, or reliability (FAA, 1996; Palmer, et al., 1995; Rehmann, et al., 1995). Operators and crew rely instead on maintaining coordination directly, by talking to each other (NTSB, 1990; Patterson, Watts-Perotti, & Woods, 1999). When conversation interrupts the operator’s task, he later must resume the interrupted task by mentally recalling his previous state or by referring to his written log. Recasting these issues as design principles, we would like systems that: • monitor the mission and anticipate situations where coordination may be degraded; • • • • adapt to coordination-threatening conditions by more conspicuously intervening to keep crew members in sync; respond to a lapse in coordination by rapidly and smoothly bridging crewmembers back into coordinated activity; help operators resume interrupted tasks by assisting with context restoration and updating logs to reflect recent events; empower users with efficient and effective direct supervisory power over automation and the mixedinitiative interaction that it requires. Enhancing Coordination with UserCentered Approaches to Decision Support User-centered approaches explore how humans interact with computers and how to create computers that better support users. When a user experiences a problem, there is a general tendency to either blame the user, or posit an anomaly in the system. A user-centered perspective would characterize the problem as a failure of the system to capture and execute what the user wants to make happen. Addressing this limitation requires that systems maintain a representation of these user goals. Efforts to develop systems that enhance coordination have generally focused on helping operators manage information overload. The basic strategy has been to solve the problem of too much information by displaying less of it, and to reduce workload through automation. Automation systems though tend to be least helpful when workload is highest (Bainbridge, 1983). Another problem with these approaches is that systems do not adequately recognize what information is most relevant in the current situation. What automation systems typically lack is an understanding of context (Woods, et al., 1999), a key determiner of the relevance of information. Context includes the current environment or tactical picture as well as the team’s and operators’ current tasks. A capability to recognize an operator’s task state could be used to monitor and report any deviations from planned procedures to keep supervisors, peers, and subordinates informed (Patterson, et al., 1998). An additional aspect to understanding context is appreciating the dependencies that obtain among operators (Malone & Crowston, 1994). Promoting effective coordination therefore hinges on systems that (1) understand what the operators are doing and what the intent of the overall mission team is; and (2) recognize the dynamics and dependencies of the situation and anticipate effects on operators. Modeling Users and their Tasks Attempts at understanding user activity have emphasized user modeling and task modeling. User modeling seeks to develop a declarative representation of the user’s beliefs, skills, limitations, current goals, and previous activity with the system (Linton, Joy, & Schaefer, 1999; Miller, 1988; Self, 1974). Task modeling focuses on representing the goals a user is likely to bring to the interaction, the starting points into that interaction, and the paths a user might follow in getting from a start state to a goal state (Deutsch, 1998; Mitchell, 1987; Rich, 1998; Rouse, 1984). Task modeling relies on an explicit representation of the steps (perhaps orderindependent) that a user might take, of the transitions among these steps, and of the language that a user employs in specifying those steps and transitions. Steps in the task need not be physical; approaches such as Cognitive Work Analysis represent mental tasks as well (Sanderson, Naikar, Lintern & Goss, 1999; Vicente, 1999). Modeling user activity is not the goal of our work but is a necessary element in developing systems with intent inference capability as a step toward promoting better crew coordination. Intent Inference Intent-inference is an approach to developing intelligent systems in which the intentions of the user are inferred and tracked. Intent inference involves the analysis of an operator’s actions to identify the possible goals being pursued by that operator (Geddes, 1986). Systems that are intent inference-capable can anticipate an operator's future actions, automatically launch the corresponding tools, and collect and filter the associated data, so that much of the information required by the operator can be at hand when needed. This approach relies on a task model, or state space (Callantine, Mitchell, & Palmer, 1999) to follow human progress through work tasks, monitoring human-system and human-human interaction within the task environment (e.g., a flight deck). To leverage task models in creating intent-aware systems, mechanisms are needed to capture user actions through observation mechanisms that supply information about operator activity in real time (from control inputs, keystrokes, the content of information queries, or spoken dialogue). These observations are coordinated with task models to help the system generate inferences about the operator’s current tasks and goals, as well as to anticipate the operator’s next actions. Operator actions can also be inferred in cases where direct observation is not feasible (Geib & Goldman, 2001). These inferences and predictions are central to the proactive property of intentaware systems, allowing them to execute proactive information queries, focus operator attention on upcoming tasks, and diagnose and report operator errors. An early example of a complex intent-inference application was The Pilots’ Associate, a DARPA program that provided foundational progress in intentinference (Banks & Lizza, 1991) and the use of task models to infer goals from actions (Brown, Santos, & Banks, 1998). Crew-Intent Inference While operator intent inference is a vital factor in helping users smoothly negotiate increasingly complex interactions with computers, perhaps more important is extending intent inference to cases involving teams, where one operator’s activity influences the work of others (e.g., Alterman, Landsman, Feinman & Introne, 1998). The “members” of the team may not understand themselves to be such, since the team approach can be applied to the study of dynamic systems of discrete participants, such as in intent inference for Free Flight (Geddes, Brown, Beamon, & Wise, 1996). But for this discussion “team” will imply a shared goal among cooperative players who may represent a team of operators (co-located or distributed), such as an aircrew or command staff. Observations of individual operators’ actions can be used to infer the goals of the entire team. Conversely, knowledge of the entire team’s goals can be used to predict the actions and information needs of individuals in the team. A system that tracks the team’s goals can facilitate stronger coordination among crew members and can support improved workload balancing. In addition, systems that are team intent-aware can remediate detected differences between individual and team goals. Intent Inference for Crew Coordination In previous work (Bell, Franke & Mendenhall, 2000), we have advanced team-level Intent Inference as an method for providing autonomous information gathering and filtering in support of a shipboard theater missile defense cell. Our approach, called Automated Understanding of Task / Operator State (AUTOS), calls for individual and team task models, observables to track operator activity, and intelligent agents to gather taskspecific information in advance of the operators’ requests for such information. The task models characterize the discrete activities that are expected to occur while performing a given task. By progressing through the model as activity is detected, AUTOS can optimize information flow to best serve the operator’s intent. Direct observations are gathered dynamically by automatically examining the actions the operator is taking: button clicks, mouse movements, and keystrokes. Indirect observations arise from an examination of the semantic content resulting from the operator’s actions, including analyses of information that the operator displays and analyses of the verbal communication among operators within the command center. The AUTOS design solution emphasizes user control for coordinating interruptions. Operators control when they will pause what they are doing to review the system-initiated results of the proactive automation. When the automation has something new, the user interface creates an iconified representation of it and presents it at the bottom of the user's work window. The user can quickly see the availability and general meaning the new information. They then control the decision of when and how to interrupt their current activity to get the new information. This solution focuses on user control and is a type of negotiation support for coordinating human interruption. A negotiation-based solution was shown by McFarlane (1999) to support better overall task performance than alternative solutions except for cases where small differences in the timeliness of handling interruptions is critical. Limitations of Intent Inference for Coordination Lessons learned from this work suggest the challenges posed by approaches that rely on task models, among them, the intrinsic complexities of the processes being modeled, the uncertainties in recognizing user intent, and the unresolved issues surrounding scalability. Limitations on task modeling constrain intent inference, since the latter relies on matching observed actions against the former to generate hypotheses about what the user is trying to do. Coordination is an especially nonlinear, non-deterministic activity that task models of the type explored in our previous work would be hardpressed to accommodate. The daunting and unsolved problem of how to costeffectively develop plausible task models can be partially overcome by reducing the need for some of the task modeling. In particular, tasks associated with coordination can be tracked using generic models of coordination, and imminent coordination lapses can be detected and corrected using these same general-purpose models. Proposed Approach Our approach integrates three capabilities: (1) predict states under which heightened coordination will be needed; (2) detect and report the onset of coordination lapses; (3) recognize and facilitate operator attempts to re-establish coordination. Predicting the states under which heightened coordination will be needed requires that operator workload and the tactical environment be monitored. Coordination lapses generate symptoms that help people recognize and correct them. Unusual delays, duplication of effort, and interference are routine indicators. Monitoring might also reveal an operator performing the wrong tasks or executing tasks out of sequence. Subtle cues can serve as early indicators, and include an operator performing a task that is “correct” but was not assigned, an operator executing a task other than what he has reported, or (subtler yet) an operator performing a task other than what he believes to be performing. Workload levels can be estimated directly by examining system outputs (number of alerts being displayed, number of targets being tracked) and system inputs (submode changes, keystrokes, trackball movements). The tactical picture can be assessed (with respect to workload) by considering the complexity of the mission plan, current mission activity, volume of incoming message traffic, presence of hostile forces, etc. Domain-tailored models of crew-level tasks would allow even more explicit specification of the conditions under which coordination becomes critical. Obvious cases where rapid and closely coordinated activity is called for are threats to the aircraft, such as when a missile launch is detected, though more routine situations will also indicate a heightened need for coordination. For instance, a team-level model might recognize that a piece of equipment is inoperative and recommend to the crew a re-distribution of tasks to minimize the effects of the equipment failure on the mission. Detecting the onset of coordination lapses could be enabled by monitoring frequency of exchanges over the ICS, recording whenever an operator sets the display to mirror another station, and detecting when control of a sensor is tasked to another operator. Sufficiently welldefined operator and team models could detect when two operators are out of sync, performing duplicate tasks, or interfering with each other. Once detected, the coordination problem must be corrected, which introduces a potential paradox of having to generate an alert for operators who are already faced with too many interruptions (McFarlane, 1999). Recognizing and facilitating operator attempts to reestablish coordination is perhaps the most subtle of the three proposed capabilities. Coordination can be regained through requests for clarification, selfannounced reports of current activity, reference to a mission log or other archival material, direct observation of on-going activity, or by reasoning from indirect observations. A system could recognize such activities by monitoring for specific commands (such as displaying a mission log, searching an archive, or requesting a playback). Explicit representations of coordination activity (Jones & Jasek, 1997) could also be applied to recognizing attempts to restore coordination. Example A P-3 Orion, the U.S. Navy’s maritime anti-submarine warfare (ASW) platform, has mission stations for the tactical coordinator (TACCO) – who also may be the mission commander (MC), navigator/communicator (NAV/COMM), acoustic sensor operators (SS/1-2), and non-acoustic sensor operator (SS/3). The most intuitive, effective, and widely-used means for maintaining crew coordination is verbal interaction over the internal communications system (ICS). The configuration of the aircraft prevents operators from seeing each other (Figure 1) and the displays restrict what each operator can view. Figure 1. Operator stations SS/1-2 (left) looking forward toward flight deck. SS/3 is immediately forward on right; TACCO is just aft of flight deck, on left; NAVCOMM is aft of flight deck, on right For instance, the SS/1-2 stations cannot show the tactical picture that is displayed to the TACCO (Figure 2). A seasoned TACCO will maintain close coordination via the ICS (which every operator listens to), reminding stations of tasks-in-progress and calling out updates to the tactical situation. There are no features designed into the automation specifically for supporting coordination. On P3s upgraded under the antisurface warfare improvement program (AIP), operators have the ability to view what other stations are seeing and a workload sharing feature enables some sensors to be controlled from any station. But assuming multimission roles (an objective of AIP) introduces additional responsibilities (such as rapid precision targeting and detection of transporter-erector-launcher sites) and with those, a greater need for coordination that simple display mirroring cannot compensate for. Below are three brief vignettes illustrating potential threats to coordination and how an intent inference-enabled automation system could recognize and help overcome (or avoid) coordination breakdowns. Equipment failure: During a mission there always exists some likelihood that an operator could become overtasked, a potential that becomes more acute when equipment failures occur. For instance, failures of the ontop position indicator (OTPI) are not uncommon, and in the absence of the OTPI the TACCO calculates buoy positions manually, which induces an increased workload. Coordination becomes especially important at such times, and an intent-aware system would be able to draw relationships between specific failure modes and the overall effects on the mission, further identifying how to retask all mission crewmembers to overcome the failure in a coordinated, synchronized fashion. Figure 2. SS/1-2 stations (left) show a different tactical picture using different data, displays, and format than TACCO (right). Workload imbalance : Additional on-board sensors and workload sharing introduce more multi-tasking responsibilities. For instance, a TACCO may task SS/1 with controlling the Advanced Imaging Multi Spectral (AIMS) camera and SS/3 with operating the inverse synthetic aperture radar (ISAR). These operators’ other obligations may at the same time begin to grow (for instance, SS/1 must also listen to acoustic data transmitted from deployed buoys while analyzing the acoustic waveforms displayed on his console, while supervising and delegating to SS/2). Intent-inference capabilities would help the automation systems recognize what tasks crewmembers were engaged in, what level of tasking they were confronting, and what suitable workload sharing solutions to recommend to TACCO. Even simple graphics depicting the system’s assessment of each station’s loading would provide the MC with a quick, visual summary and offer early warnings of imminent coordination lapses. Context Switching: With P-3 crews routinely engaging in multi-tasking, a typical situation might have the crew suspend the ASW mission, fly to a release point to launch an air-to-surface weapon, and return to the prior location to resume the ASW task thirty or forty minutes later. Transitioning back into the ASW task requires recalling the Last Known Position for each track, re-entering the frequencies being tracked, looking at the historicallygenerated tracks, and checking any relevant data links for updated information. When an operator's task is interrupted by a more urgent task, his previous context is restored in part by reference to his written mission log, though he must manually restore equipment configurations if any were changed in the interim (see Malin et al. (1991) for another example of the use of logs for recovery after interruption). Intent-inference would enable systems that detect changes in the mission, verify with each operator that his equipment is properly reconfigured, and automatically save the current state (settings, mission log, tactical picture). The interrupted task in this case need not lie dormant – intelligent sentinel agents would instead monitor incoming data source such as Link-11/Link-16, record imagery, and monitor comms channels. Resuming the interrupted task is analogous in some respects to a shift change (Patterson & Woods, 1997), so restoration of context could be handled in part by offering each operator a “playback” tailored to that operator’s responsibilities. The system should also be able to restore all configurations to their prior states, and integrate historical information with updates filtered from the information streams that were autonomously monitored while the task was in suspense. Conclusion Mission operations invariably demand close coordination among aircrew members serving different roles and requiring information support from a disparate array of assets. New responsibilities for operators must be accompanied by systems that possess an understanding of crew roles and coordination. The simple vignettes in this paper offer a brief glimpse of how intent inference-enabled systems could promote more effective aircrew coordination. The brief examples portrayed situations that are relatively routine and non-critical. Tactical environments with real-time mission profiles and hostile threats have an even more pronounced impact on the need for crew coordination (as alluded to by the Escalation Principle). The need for intent inference approaches to promoting aircrew coordination is rendered all the more urgent when these real-world conditions are considered. References Alterman, R., Landsman, S., Feinman, A., and Introne, J. (1998). Groupware for Planning. Technical Report CS-98-200, Computer Science Department, Brandeis University, 1998. Bainbridge, L. (1983). Ironies of Automation, Automatica, 19, pp. 775-779. Banks, S.B., and Lizza, C.S. (1991). Pilot's associate: A cooperative, knowledge based system application. IEEE Expert, 18–29, June 1991. Barker, J., McKinny, E., Woody, J., and Garvin, J.D. (1997). Recent Experience and Tanker Crew Performance: The Impact of Familiarity on Crew Coordination and Performance. In Proceedings of the 9th International Symposium on Aviation Psychology, Ohio state university, Columbus, OH. Bell, B.L., Franke, J., and H. Mendenhall (2000). Leveraging Task Models for Team Intent Inference. In Proceedings of the 2000 International Conference on Artificial Intelligence, Las Vegas, NV. Brown, S.M., Santos, E., and Banks, S.B. (1998). Utility TheoryBased User Models for Intelligent Interface Agents. In Proc. of the 12th Canadian Conf on Artificial Intelligence (AI '98), Vancouver, British Columbia, Canada, 18-20 June 1998. Callantine, T.J., Mitchell, C.M., and Palmer, E.A. (1999). GTCATS: Tracking Operator Activities in Complex Systems. Technical Memo NASA/TM-1999-208788. NASA Ames Research Center, Moffet Field, CA. Decker, K. (1998). Coordinating Human and Computer Agents. In W. Conen, and G. Neumann, (Eds.), Coordination technology for Collaborative Applications - Organizations, Processes, and Agents. LNCS #1364, pages 77-98, 1998. Springer-Verlag. Diehl, A.E. (1991). The Effectiveness of Training Programs for Preventing Aircrew "Error". In Proceedings of the Sixth International Symposium of Aviation Psychology, Ohio St. Univ., Columbus, OH. Deutsch, S. (1998). Multi-disciplinary Foundations for Multipletask Human Performance Modeling in OMAR. In Proceedings of the 20th Annual Meeting of the Cognitive Science Society, Madison, WI, 1998. Eldredge, D., Mangold, S., and Dodd, R.S. (1992). A Review and Discussion of Flight Management System Incidents Reported to the Aviation Safety Reporting System. Final Report - DOT/FAA/RD-92/2, Washington, DC: Federal Aviation Administration. FAA (1996). The Interfaces Between Flight Crews and Modern Flight Deck Systems. Report of the FAA Human Factors Team. Washington, D.C.: FAA. Funk, K.H., Lyall, E.A., and Niemczyk, M.C. (1997). Flightdeck Automation Problems: Perceptions and Reality. In M. Mouloua and J.M. Koonce (Eds.) Human-automation interaction: Research and practice. Mahwah, New Jersey: Lawrence Erlbaum Associates. Geddes, N.D. (1986). The use of individual differences in inferring human operator intentions. In Proceedings of the Second Annual Aerospace Applications of Artificial Intelligence Conference, 1986. Dayton, OH. Geddes, N.D., Brown, J.L., Beamon, W.S., & Wise, J.A. (1996). A shared model of intentions for free flight. In Proc. of the 14th IEEE/AIAA Digital Avionics Systems Conference. Atlanta, GA. 27-31 October. Geib, C., and Goldman, R.P. (2001). Issues for Collaborative Task Tracking. In Proc. of the AAAI Fall Symposium on Intent Inference for Collaborative Tasks, AAAI Press, forthcoming. General Accounting Office (1996). Military Aircraft Safety: Significant Improvements Since 1975 (Briefing Report, 02/01/96, GAO/NSIAD-96-69BR. International Civil Aviation Organization (1989). Human Factors Digest No. 1 , ICAO Circular 216- AN/ 131, December 1989. Jones, P. M. and Jasek, C. A. (1997). Intelligent support for activity management (ISAM): An architecture to support distributed supervisory control. IEEE Transactions on Systems, Man, and Cybernetics, 27, No. 5, 274-288. LaJoie, A.S. & Sterling, B.S. (1999) Team and Small Group Processes: An Annotated Bibliography, Army Research Institute Research Report, 2000-01. Latorella, K. A. (1996). Investigating interruptions: An example from the flightdeck. In Proceedings of the Human Factors and Ergonomics Society 40th Annual Meeting. Latorella, K. A. (1998). Effects of modality on interrupted flightdeck performance: Implications for Data Link. Proceedings of the Human Factors and Ergonomics Society 42th Annual Meeting. Linton, F., Joy, D., Schaefer, H.P. (1999). Building user and expert models by long-term observation of application usage. In Proceedings of the 7 th International Conference on User Modeling, 20-24 June, Banff, CA. Malin, J. T., Schreckenghost, D. L., Woods, D. D., Potter, S. S., Johannesen, L., Holloway, M., & Forbus, K. D. (1991). Making Intelligent Systems Team Players: Case Studies and Design Issues, Vol. 1, Human-Computer Interaction Design (NASA Technical Memorandum 104738): National Aeronautics & Space Administration (NASA), Washington, DC. Malone, T. and Crowston, K. (1990). What is coordination theory and how can it help design cooperative work systems? In Proceedings of the 1990 ACM Conference on ComputerSupported Cooperative Work, 357-370. Malone, T. and Crowston, K. (1994). The Interdisciplinary Study of Coordination, ACM Computing Surveys, Vol. 26, No. 1, pages 87-119. McFarlane, D. C. (1997). Interruption of People in HumanComputer Interaction: A General Unifying Definition of Human Interruption and Taxonomy (NRL Formal Report NRL/FR/5510—97-9870): Naval Research Laboratory, Washington, DC. McFarlane, D. C. (1999). Coordinating the Interruption of People in Human-Computer Interaction, Human-Computer Interaction - INTERACT'99, Sasse, M. A. & Johnson, C. (Editors), Published by IOS Press, Inc., The Netherlands, IFIP TC.13, 295-303. Miller, J.R. (1988). The role of human-computer interaction in intelligent tutoring systems. In M.C. Polson and J.J. Richardson (Eds.), Foundations of Intelligent Tutoring Systems, 143–190. Hillsdale, NJ: Lawrence Erlbaum. Mitchell, C.M. (1987). GT-MSOCC: a research domain for modeling human-computer interaction and aiding decision making in supervisory control systems. IEEE Transactions on Systems, Man, and Cybernetics, 17(4), 553-570. NTSB (1990). Aircraft Accident Report: United Airlines Flight 232, McDonnell Douglas DC-10-10 N1819U, Sioux Gateway Airport, Sioux City, Iowa, July 19, 1989. Washington: NTSB, Report No. NTSB-AAR-90-06. Qureshi, Z.H. (2000). Modeling Decision-Making In Tactical Airborne Environments Using Cognitive Work AnalysisBased Techniques. In Proc. of the 19th Digital Avionics System Conference, Philadelphia, PA, October, 2000. Palmer, N.T., Rogers, W.H., Press, H.N., Latorella, K.A., and Abbott, T.S. (1995). A crew-centered flight deck design philosophy for high-speed civil transport (HSCT) aircraft. NASA Technical Memorandum 109171, Langley Research Center, January, 1995. Patterson, E.S., and Woods, D.D. (1997). Shift Changes, Updates, and the On-Call Model In Space Shuttle Mission Control. In Proceedings of the 41st Annual Meeting of the Human Factors and Ergonomics Society, September, 1997. Patterson, E.S., Watts-Perotti, J., and Woods, D.D. (1999). Voice Loops as Coordination Aids in Space Shuttle Mission Control. Computer Supported Cooperative Work, 8(4): 353371. Patterson, E.S., Woods, D.D., Sarter, N.B., & Watts-Perotti, J. (1998). Patterns in cooperative cognition. In Proc. of COOP '98, Third International Conference on the Design of Cooperative Systems. Cannes, France, 26-29 May. Rehmann, A. J., Mitman, R. D., Neumeier, M. E., and Reynolds, M. C. (1995). Flightdeck Crew Alerting Issues: An Aviation Safety Reporting System Analysis. (DOT/FAA/CTTN95/18). Washington, DC: U.S. Department of Transportation, Federal Aviation Administration. Rich, E. (1998). User modeling via stereotypes. In M. Maybury and W. Wahlster (Eds.), Readings in Intelligent User Interfaces, 329-342. San Francisco: Morgan Kaufmann. Rouse, W.B. (1984). Design and evaluation of computer-based support systems. In G. Salvendy (Ed.), Advances in Human Factors/Ergonomics, Volume 1: Human-Computer Interaction. New York: Elsevier. Salas, E., Bowers, C. A., & Cannon-Bowers J. A. (1995). Military team research: Ten years of progress. Military Psychology, 7, 55-75. Sanderson, P, N. Naikar, G. Lintern and S. Goss (1999). Use of Cognitive Work Analysis Across the System Life Cycle: Requirements to Decommissioning. In Proc. of 43rd Annual Mtg of the Human Factors & Ergonomics Society, Houston, TX, 27-Sep – 01-Oct. Sarter, N.B. and Alexander, H.M. (2000). Error Types and Related Error Detection Mechanisms in the Aviation Domain: An Analysis of ASRS Incident Reports. International Journal of Aviation Psychology, 10(2), 189206. Schmidt, K. (1994). Modes and mechanisms for cooperative work. Risø Technical Report, Risø National Laboratory, Roskilde, Denmark. Schutte, P.C., and Trujillo, A.C. (1996). Flight crew task management in non-normal situations. In Proceedings of the Human Factors and Ergonomics Society 40th Annual Meeting, 1996. pp. 244-248. Self, J.A. (1974). Student models in computer-aided instruction. International Journal of Man-Machine Studies, 6, 261-276. Sheehan, J. (1995). Professional Aviation Briefing, December 1995. US Army (1995). Aircrew Training Program Commander's Guide to Individual and Crew Standardization, Training Circular 1210, October, 1995. Vicente, K (1999). Cognitive Work Analysis. Lawrence Erlbaum Associates, Mahwah, NJ. Wiener, E.L. (1989). Human factors of advanced technology ("glass cockpit") transport aircraft. (NASA Contractor Report No. 177528). Moffett Field, CA: NASA Ames Research Center. Wiener, E. L. (1993). Intervention strategies for the management of human error. NASA Contractor Report Number NAS1264547. Moffett Field, CA: NASA Ames Research Center. Wilkins, D.C., Mengshoel, O.J., Chernyshenko, O., Jones, P.M., Bargar, R., and Hayes, C.C. (1998). Collaborative decision making and intelligent reasoning in judge advisor systems. In Proceedings of the 32nd Annual Hawaii International Conference on System Sciences. Wise, J.A., D.W. Abbott, D. Tilden, J.L. Dyck, P.C. Guide, & L. Ryan. (1994). Automation in the Corporate Aviation Environment. In N. McDonald, N. Johnston, & R. Fuller (Eds.), Applications of Psychology to the Aviation System, Cambridge: Avebury Aviation. pp 287-292. Woods, D.D., Patterson, E.S., Roth, E.M., and Christoffersen, K. (1999). Can we ever escape from data overload? In Proc. of the Human Factors and Ergonomics Society 43rd annual meeting, Sept-Oct, 1999, Houston, TX.