A Fuzzy Expert System for Task Distribution in Teams under Unbalanced
Workload Conditions
José A. Benito Calleja and Jimmy Troost
Thales Nederland, Hengelo, The Netherlands jose.benito@nl.thalesgroup.com, jimmy.troost@nl.thalesgroup.com

Abstract
Inappropriate workload levels on the team members of a naval force have been detected as a problem that can threaten the performance and safety of future naval operations. A suitable distribution of tasks among the members of a team is a crucial issue in order to prevent high and low workload levels. In this paper, we propose a rule-based expert system, the Task
Distribution Expert System (TDES), which assists team leaders to manage mental workload in a team by suggesting appropriate task assignments. The TDES emulates the behavior of a team leader deciding which member of the team should perform a task and how.
The system handles mental workload as an uncertain fuzzy concept comprising three fuzzy variables that represent the way mental workload affects performance. Automation issues and different recommendations for effective workload management in teams are analyzed and incorporated. A prototype demonstrates the system.

1. Introduction
Naval Command and Control (C2) systems support organizations formed by a number of people cooperating on a multitude of tasks simultaneously to achieve overall goals. Future naval C2 systems are characterized by less people, more information available, shorter decision cycles and more complex and varied tasks [1]. The nature and number of those tasks can lead an operator, a team, and finally an organization to low and high mental workload situations. These inappropriate mental workload conditions threaten the performance of future naval operations. Traditionally, workload has been addressed at the task level [2]. This is based on the fact that if a task is modified, its associated demands will change. Some aspects that have been considered at this level are:

information filtering, information representation and intelligent decision support systems. However, workload must also be addressed at other levels [3]: operator level, team level and organization level. At the operator level, not only one task but a set of tasks is assigned to an operator; there is a need for creating a plan for effectively finishing them considering their importance and relationships. At the team level, there is not only one operator performing tasks but a group of them; workload on an operator can be managed by suitable distribution of tasks among the members of the team. Finally, the organization level considers the balance of tasks between teams.
The Distributed Adaptive Task Management
System (DATMS) [4] has been proposed as a solution to tackle mental workload at the team and operator levels. The DATMS consists of 4 modules: Workload
Measure (WM), Workload Prediction (WP), Task
Distribution (TD) and Task Scheduling (TS). The WM and WP modules estimate the workload of an operator: the former monitors the operator’s dynamic state directly whereas the latter uses the tasks assigned to an operator and his static characteristics. The TD module merges workload values with instructions and suggestions from the operators to assign tasks to an operator, a machine or a cooperation of both. Finally, the TS module, one per operator, receives the tasks assigned to its associated operator and helps him to decide which tasks should be performed and when.
In this paper, we propose a rule-based expert system as a solution for the TD module. The rule-based expert system comprises a representation of mental workload, and a number of facts and rules that perform the functionality of the TD module by emulating the behavior of a leader doing task assignment among the members of a team.
This paper is organized as follows. In section 2, we analyze the main issues involved: mental workload, workload management in teams and automation. In section 3, we describe the context and requirements of the TD module. In sections 4 and 5, we justify and

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

describe the workload representation and the main components included in the rule-based expert system.
In section 6, we demonstrate the expert system. Section
7 concludes this paper.

2. Domain analysis
2.1. Mental workload
Many studies have been performed in order to define the concept of mental workload but researchers have not achieved an agreement on a single definition that suits all the disciplines. Traditionally, workload has been defined as imposed task demands, level of performance, mental effort exerted by an operator, or the operator’s perception [5].
Since the workload experienced by an operator affects the way he performs, it is relevant to determine the relationship between workload and performance.
Workload has typically been related to performance according to an inverted U relation [6] (see Fig. 1).
Three regions can be differentiated: first, a region of poor performance due to low workload that reduces attention, an increase in workload leads to an increase in performance; second, a region of acceptable workload in which the operator performs properly; and third, a region of poor performance due to the excessive load on the operator, an increase in workload decreases performance.
There are different methods of measuring mental workload. These methods can be classified into three groups [7]: performance, subjective and physiological.
Performance measures assume the inverted U relation between mental workload and performance. Subjective measures are obtained from reports, in real-time or after the tasks have been finished, by the operator performing the tasks or by a supervisor. Physiological measures are based on the role of the nervous system in controlling all behavior. Since every method has

their own drawbacks, the aggregation of different measures allows the achievement of a more reliable indicator that mitigates the specific limitations of individual measures [8].
The availability of workload measures makes feasible the estimation of an operator’s mental workload state, and their further use to prevent situations that diminish performance.

2.2. Workload management in C2 teams
The success of a C2 team requires that team members collaborate on a large number of tasks simultaneously to accomplish a mission. Current practice in the naval C2 domain is based on systematic assignment of tasks to operators in an inflexible manner, generating periods of high workload for some operators while others may remain almost inactive with low workload. Much higher precision in task assignments with proper distribution of workload is required to enhance team performance [2]. To support that issue, naval C2 organizations are moving from a rigid structure to a more flexible one based on role crossing: besides being trained to perform tasks according to their role, operators are qualified to carry out tasks that do not match their role [1].
In addition to the role crossing concept, different recommendations have been proposed to improve team performance [9]. Two of them should specially be considered in high workload environments. First, team leaders in high workload situations should delegate tasks in order to reduce multi-tasking and improve decision-making and vigilance. Second, implicit coordination, defined as coordination through the anticipation of the needs of others, should be promoted. In order to achieve effective workload management in teams, the above issues are essential and must be incorporated in a system responsible for task distribution. 2.3. Adaptive and adaptable automation

Figure 1. Relation between workload and performance. Automation tools release a team member from part or even the whole load of a task. The control of the automation level can be placed in the user (adaptable system), or in the system (adaptive system). Adaptive systems allow the achievement of greater speed of performance, reduced human workload, more consistency, a greater range of flexibility in behaviors and can require less training time. However, by taking the operator out of the control “loop”, some human performance problems can arise: diminished situational and system awareness, over-reliance, skill decrease,

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

performance degradation and reduced user acceptance
[10].
Adaptable systems maintain operators in active charge of the automation level, keeping them more actively “in the loop” and, hence, more aware of how the system is performing, reducing the undesirable effects described above. However, adaptable systems may increase operators’ workload, because humans may not have the time or expertise to select the best forms of automation [10].
From previous experience, it is clear that users want to remain in charge of actions; in particular, they want to remain in charge of the task allocation functionality
[11]. However, they also want the benefits that automation can provide. An integrated approach, capable of combining the benefits of both methods while diminishing their drawbacks, is required.

to be allocated (adaptive behavior). To diminish the problems of adaptive systems, the TD module must cautiously consider the application of excessive levels of automation and show some further adaptable behavior by allowing a team member to change the level and kind of automation of his tasks [10].
The above requirements and expected behavior lead us to consider a rule-based expert system, the Task
Distribution Expert System (TDES), as a solution for the TD module. The TDES emulates the behavior of a team leader doing task assignment among the members of a team. There are some advantages over a traditional program: ability to explain reasoning, learning, and easy management of knowledge. Efficiency, quality, consistency and completeness are advantages over a human expert; however, they cannot be replaced in aspects like creativity and common sense [12].

3. Task Distribution module

4. Workload representation

The Task Distribution (TD) module of the DATMS
[4] aims to assist a team leader to handle workload at the team level by suggesting proper assignment of tasks. Proper assignment of tasks involves either the allocation of incoming tasks to a team member, machine or a cooperation of both, or the retrieval of tasks from a team member, their modification and reallocation. Appropriate assignment of tasks strives after keeping an acceptable team members’ workload while tasks are carried out properly.
In order to carry out its task assignment responsibility, the TD module must combine task characteristics, team members’ features and their statements. Therefore, the TD module needs to interact with three kinds of external entities: a task monitor that inspects the current situation and generates tasks to handle it; a team monitor that examines the different members of the team and generates a description of them (fundamental extracted features are their roles and their mental workload levels); and team members
(specially the team leader) that enter statements that control and influence the way in which the TD module operates and the decisions it makes.
The TD module must show adaptable and adaptive behavior. On one hand, since team leaders want to remain in charge of the task allocation functionality, the system must be configurable as a decision support system with different automation levels on team leaders’ demand
(adaptable
behavior).
To
accommodate those levels of automation, the TD module must be capable of reasoning, learning and explaining its own behavior as well as accepting direction from leaders. On the other hand, the TD module has to suggest the automation level for a task

In order to determine inappropriate levels of workload, the TDES needs an internal representation of how workload affects performance, i.e., an internal representation of the relationship between workload and performance. We use a representation based on the inverted U relation (see Fig. 1).

4.1. Workload segments
The workload representation could be either a continuous description or a segment description. A continuous description represents workload with an inverted U function allowing the system to handle an exact mapping between workload and performance.
However, as it requires workload measures to be provided quantitatively, and some methods only supply qualitative values, this approach restricts the number of methods that can be used. In addition, it does not allow a team leader to assess workload using natural language. A segment description represents workload with regions that characterize qualitatively the relation between workload and performance. It tackles the two disadvantages of the continuous description: methods providing either qualitative or quantitative values can be used after an adaptation process that adjusts their values to the defined segments; and, natural language can be used by relating terms to the segments. The main drawback of this approach is the loss of information when converting workload numerical values into regions. However, the specific shape of the relation between workload and performance, characterized by two clearly differentiated levels with two transitions, diminishes this disadvantage.

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

In order to adjust the inverted U, the TDES needs to distinguish, at least, between three segments (see Fig.
2): the underload segment, a region of low workload in which the operator performs poorly; the capable segment, a region of acceptable workload in which the operator carries out tasks properly; and the overload segment, a region in which the workload becomes excessive leading to deficient performance.

4.2. Fuzziness
The three-segment representation is not sufficient for the TDES to handle workload properly due to two reasons. First, the ambiguity of the capable segment: when an operator is performing within the capable region, the system is unable to determine whether he is close to the underload segment, close to the overload segment or in the middle part. Second, the rigid boundaries of the segments do not allow the system to capture the transitions that exist between them.
In order to cope with the above issues we apply fuzziness to the three-segment representation.
Fuzziness occurs when the boundary of a piece of information is not clear-cut [13]. Terms like underload, capable and overload can be regarded as fuzzy. For example, assuming quantitative workload values between 0 and 100 mapped to the inverted U relation:
5 is definitely underload, 50 is capable, and 95 is overload; however, 20 has some possibility of being underload as well as capable, and 70 has some possibility of being overload as well as capable.
We handle fuzziness using fuzzy set theory. Unlike classical set theory where membership of an element to a set can be clearly described, in fuzzy set theory, membership of an element to a set can be partial [13].
Therefore, we regard workload as a fuzzy concept associated with the fuzzy term set {underload, capable, overload}, where each term represents a specific fuzzy set. Linear representations of the workload fuzzy sets that match the inverted U relation are shown in Fig. 3.
A further advantage of fuzziness is the option of

associating modifiers like not, very, extremely and somewhat, as well as the operators and and or with the fuzzy terms in order to define new fuzzy terms on demand using natural language [14]. For instance, the fuzzy terms very capable and degrade may be defined to point out especially appropriate workload conditions and overall poor workload conditions respectively. The distribution of the term very capable is obtained by applying the mathematical function associated with the modifier very (typically the square root function) to the fuzzy set of the term capable. The fuzzy set of the term degrade can be defined as the union of the fuzzy sets of the terms underload and overload (underload or overload). Fig. 4 shows the distribution of the terms very capable and degrade using the previous representation of the term capable.

4.3. Uncertainty
Uncertainty is utilized in the TDES in order to model the certitude about workload assessments made by the workload measures or the team members. We handle uncertainty by assigning a crisp numerical value on a scale from 0 to 1, where 1 indicates that the system can be certain about the workload assessment, and 0 indicates that it can be completely uncertain about the workload assessment [14].

Figure 3. Grade distribution of the workload fuzzy sets.

Performance

High underload capable

overload

Low
Low

High
Workload

Figure 2. Workload segments.

Figure 4. Grade distribution of derived workload fuzzy sets.

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

5. TDES components
The TDES, like every rule-based expert system, comprises five components [12]: the fact base that holds the data of interest; the knowledge base that contains the problem solving knowledge as rules; the inference engine, a set of algorithms that work through rules and facts and control the process of deriving conclusions; the interface that allows the communication with the external entities; and the working memory that keeps the rules that can be applied. Fact base and knowledge base are described in the next subsections.

5.1. Fact base
The fact base of the TDES contains the relevant data to assign tasks. Facts are inserted, updated and removed by the external entities to the expert system and by the knowledge base as a result of the inference process. Facts introduced by the external entities can be classified into three groups: team member facts, task facts and statement facts. A team member fact holds the relevant features of a team member: identifier, role, skill and workload. These facts are inserted and updated by the team monitor entity and, altogether, describe the team. A task fact holds relevant information of a task: identifier, role, automationLevelcomplexity (list of pairs that indicates the different levels of automation and their associated demands), importance and urgency. These facts are inserted and removed by the task monitor entity and, altogether, define the load imposed on the team. A statement fact holds information extracted from an instruction or suggestion that a team member has entered in the system. Two kinds of statement facts are considered: task-assignment facts and workload-assessment facts.
When a team member suggests a particular assignment of a task, a task-assignment fact is derived that contains: task identifier, team member identifier, automationLevel, and certainty. By using the maximum level of certainty, the team leader keeps complete control over the task allocation. When a team member enters his perception of the workload that he or others are experiencing, a workload-assessment fact is derived that contains: team member identifier, workload, and certainty.

5.2. Knowledge base
The knowledge base contains rules that simulate in a condition-action (if-then) basis the behavior of a human team leader doing task assignment. From a functional point of view, four kinds of rules are

considered: observation rules, orientation rules, filtering rules and decision-making rules.
The observation rules are in charge of detecting when the task assignment process must be thrown.
They comprise six rules that monitor six situations: “if a new task is entered then allocate it”, “if a team member’s workload is underload then increase his load”, “if a team member’s workload is overload then reduce his load”, “if the team’s workload is overload then decrease the load of the team leader” (see section
2.2), “if a team member requests a task then evaluate that assignment”, and “if the team leader requests a task assignment then perform it” (team leader in charge of the task allocation functionality).
The orientation rules are responsible for finding feasible task assignments. They comprise three groups of rules that react to the first four situations monitored by the observation rules (the last two situations do not require orientation):
• Increase the load of a team member: the knowledge included in these rules aims to propose ways of increasing the workload of a team member. Two different rules addressing two different mechanisms to raise workload are considered: “if increase the load of a team member then decrease the automation level of his tasks”, and “if increase the load of a team member then retrieve a task from another team member and assign it to him”. The retrieval of a particular task is evaluated by further rules that consider workload and role aspects.
• Reduce the load of a team member: the goal of the knowledge included in these rules is to propose ways of decreasing the workload assigned to a team member. Again, two mechanisms are implemented by two different rules: “if reduce the load of a team member then increase the automation level of his tasks”, and “if reduce the load of a team member then retrieve one of his tasks and assign it to another team member”. This is handled by the rules that address the next situation.
• Allocate an incoming task: the knowledge included in these rules aims to propose feasible allocations for a new task. The rules consider different aspects like ‘find a team member that matches the role and is in the underload state’, ‘find a team member that matches the role and is in the capable state’,
‘perform a shift in the automation level of the task’, and ‘assignment to a team member that does not match the role’.
For every assignment proposed by the orientation rules, two factors are derived: certainty and quality.
Certainty indicates the confidence the expert system has in the assignment. It is the result of the fuzzy inference process and the uncertainty associated to the

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

assessments. Quality evaluates the assignment independently of the workload context of the team: quality = Q(role, state, automationBias, roleCrossing); where role points out whether the role is matched, state indicates the workload state, automationBias shows the task automation shifting, and roleCrossing points out the level of role crossing within the team, and affects how much role influences quality. The highest quality is obtained when the task is assigned without shifting its automation level (automationBias=0), to a team member that matches its role (role=true), and has capable workload state (state=capable).
The filtering rules are in charge of sending to the team leader the most relevant task assignment proposals made by the orientation rules. The way the filtering rules provide proposals depends on two parameters of the expert system configured by the team leader: quantity and mode. Quantity determines the number of proposals. Mode establishes the way the expert system operates: optimal or sub-optimal. In the optimal mode, the filtering rules wait for the orientation rules to finish searching for feasible assignments and provide those with the highest certainty and quality values. In the sub-optimal mode, as the orientation rules propose feasible assignments, the filtering rules send those whose quality and certainty values are higher than predefined thresholds.
Finally, the decision-making rules comprise two rules that are responsible for selecting the task assignment to be executed. Which one is fired depends on whether the team leader decides on an assignment:
“if the team leader selects an assignment then perform that assignment”, or “if the team leader does not select an assignment then perform the assignment with the highest quality and certainty”.

6. Prototype
In order to demonstrate the TDES, a prototype has been developed using FuzzyClips [14], a tool for implementing expert systems that supports fuzziness and uncertainty. The prototype includes a fuzzy representation of workload according to Fig. 3, and a fact base and a knowledge base containing the data and rules described in section 5.
Workload
balance, inappropriate workload avoidance, role crossing and task automation shifting aspects are tested with a scenario characterized by a team composition and the tasks created by a task generator. The team consists of one operator of type A and one operator of type B. Operators’ features and the

characteristics of their tasks are used to simulate their workload values [4]. Operators are modeled as workload consumers: they consume a particular amount of workload in a period of time. The task generator creates, over time, a static number of B tasks that impose acceptable workload on the B operator, and a configurable number of A tasks that impose diverse workload levels on the A operator. A and B tasks can be carried out using three automation levels (being the intermediate level the default) that establish three levels of load. Four different situations are generated:
• S1: low demanding. The default workload imposed by the A tasks is very low; a rigid assignment
(neither role crossing nor automation shifting) would lead to low workload levels on the A operator. The TDES must decrease the automation level of the A tasks or send some B tasks to the A operator in order to increase his workload.
• S2: medium demanding. The workload caused by the A tasks is acceptable; a rigid assignment would lead to acceptable workload levels. Therefore, the
TDES does not need to apply role crossing or automation shifting.
• S3: high demanding. The workload imposed by the
A tasks is high; a rigid assignment would lead to high workload on the A operator. The TDES must increase the automation level of the A tasks or send some A tasks to the B operator to reduce the A operator’s workload.
• S4: very high demanding. The workload caused by the A tasks is very high; a rigid assignment would again lead to high workload. The TDES must both increase the automation level of the A tasks and send some A tasks to the B operator to reduce the A operator’s workload.
Fig. 5(a) shows the workload experienced by the operators as a result of a rigid assignment. The B operator experiences acceptable workload in all the situations whereas the A operator suffers different workload conditions: low in the first situation, acceptable in the second situation, and high in the third and forth situations. Fig. 5(b) shows the workload experienced by the operators as a consequence of the assignment performed by the TDES prototype. Both operators experience acceptable workload in all the situations. Table 1 describes the allocation of tasks performed by two configurations of the TDES prototype (low and high role crossing) over a period of time in the different situations:

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

Figure 5. Workload values: (a) for the rigid allocation and (b) for the TDES prototype.
Table 1. Task allocation results.

• S1: the TDES has increased the load of the A operator by applying low automation levels (1.1 on average) in the low role crossing condition, and by sending 5 B tasks to the A operator in the high role crossing condition. To compensate the decrease in load of the B operator in the latter condition, the B tasks assigned to the B operator have lower automation levels (1.6 vs. 2.0 on average).

• S2: the TDES has not modified the default load of the A operator; neither role crossing nor task automation shifting has been carried out.
• S3: the TDES has decreased the A operator’s load mainly by applying high automation levels (2.8 on average) in the low role crossing condition, and by sending 7 A tasks to the B operator in the high role crossing condition. To compensate the increase in load of the B operator in the latter condition, the B

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE

tasks assigned to the B operator have higher automation levels (2.3 vs. 2.0 on average).
• S4: the TDES has decreased the A operator’s load in both role crossing conditions by applying high automation (2.8 and 2.7 respectively) and moving some A tasks to the B operator (10 and 12 respectively). The B tasks assigned to the B operator have higher automation to compensate the increase in load (2.5 and 2.6 on average).

7. Conclusions
In this paper we presented a rule-based expert system, the Task Distribution Expert System, which aims to prevent, detect and cure inappropriate mental workload situations on the members of a naval C2 team. Inappropriate workload situations will arise in the naval C2 domain as a consequence of the number and nature of tasks that must be carried out.
The TDES assists team leaders to manage mental workload in a team by suggesting task assignments.
Workload is handled as an uncertain fuzzy concept allowing a simple, natural and extensible description of the way different levels of workload affect performance. Role crossing and task automation shifting aspects are considered in order to achieve high precision in task assignments. The results of a prototype of the TDES showed that the system is able to keep acceptable workload levels for different workload demanding situations, by applying role crossing and/or task automation shifting.
So far all the knowledge included in the TDES has been built from the analysis of the domain. Such knowledge may well be incomplete and inaccurate.
Learning can allow the TDES to improve its knowledge. Therefore, further work will be directed to investigate and introduce a learning mechanism in the
TDES so that the prevention, recognition and reaction to inappropriate workload situations can be enhanced.
Inappropriate workload conditions are not unique to the naval C2 domain. Many government and private organizations operate in similar conditions and, therefore, could benefit from the TDES.

8. Acknowledgment
Thanks to the National Research Council of Canada for the license to the software NRC FuzzyClips. This research has been supported by a Marie Curie

fellowship of the European Community programme
“Mutual Adaptive Human Computer Interaction” under contract number HPMI-CT-2002-00221.

9. References
[1] D.S. Alberts and R.E. Hayes, Power to the Edge:
Command and Control in the Information Age, Washington,
DC: US DoD C2 Research, 2003.
[2] G.A. Osga, “Human-centered shipboard systems and operations,” in H.R. Booher, Handbook of Human Systems
Integration, Hoboken, NJ: Wiley Interscience, 2003.
[3] T. Lenox et al., “Support of teamwork in human-agents teams,” in Proc. IEEE Int. Conf. Syst., Man, Cyb., San
Diego, CA, 1998, pp. 1341-1346.
[4] José A. Benito Calleja and Jimmy Troost, “Dealing with high workload in future naval command and control systems,” in Proc. IEEE Int. Conf. Syst., Man, Cyb., The Big
Island, HI, 2005.
[5] B.N. Huey and C.D Wickens, Workload Transition:
Implications for Individual Team Performance, Washington,
DC: National Academy Press, 1993.
[6] D. De Waard, The Measurement of Drivers' Mental
Workload, PhD thesis, Haren, The Netherlands: University
Groningen (TRC), 1996.
[7] N.L. Miller, J.J. Crowson and J.M. Narkevicius, “Human characteristics and measures in systems design,” in H.R.
Booher, Handbook of Human Systems Integration, Hoboken,
NJ: Wiley Interscience, 2003.
[8] K.F. Van Orden, “Workload assessment and management for the Multimodal WatchStation,” in Proc. 44th Conf.
Human Factors and Ergonomics Soc., San Diego, CA, Aug.
2000.
[9] J.B. Sexton. Golden Rules of Group Interaction in High
Risk Environments: Evidence-based Suggestions for
Improving Performance, Rüschlikon, Switzerland: Swiss Re
Centre for Global Dialogue, 2004.
[10] C.A. Miller, H.B. Funk, R. Goldman, J. Meiser and P.
Wu, “Implications of adaptive vs. adaptable UIs on decision making: why automated adaptiveness is not always the right answer,” in Proc. 1st Int. Conf. Augmented Cognition, Las
Vegas, NV, July 2005.
[11] C.A. Miller and H.B. Funk, “Associates with etiquette: meta-communication to make human-automation interaction more natural, productive and polite,” in Proc. 8th Europ.
Conf. Cognition Sci. Approaches in Process Control,
Munich, Germany, Sept. 2001.
[12] P. Jackson, Introduction to Expert Systems, 3rd ed.,
Reading, UK: Addison-Wesley, 1999.
[13] E. Cox, The Fuzzy Systems Handbook, San Diego, CA:
AP Prof., 1999.
[14] FuzzyClips Version 6.04A User’s Guide, National
Research Council, Canada, 1998.

Proceedings of the 2005 International Conference on Computational Intelligence for Modelling, Control and Automation, and International Conference on
Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’05)
0-7695-2504-0/05 $20.00 © 2005

IEEE