Individual Ability-Based System Design of Dependable Human-Technology Interaction

This paper highlights the importance of considering especially individual differences in intelligence when designing systems and interfaces due to their impact on operator performance in new and unfamiliar situations. For this purpose, an approach is introduced which allows assessing performancerelevant abilities of the operators on the basis of their performance on everyday life tasks. In order to increase the overall human-machine system dependability, guidelines are derived about appropriate reconfigurations of the technical system and/or its interface on the basis of the assessed performancerelevant abilities. The impact of this new approach to dependable system and interface design is discussed. 1. MOTIVATION: HUMAN-CENTERED TECHNOLOGY Research in the field of human-centered technology is often motivated by developing technical systems which optimize the overall system performance in normal situations, in unanticipated circumstances and during system breakdowns. Two approaches to achieve this goal can be distinguished and are described in the following. 1.1 Definition of an optimal “ level of automation” Some human-centered technology researchers aim at defining an optimal “level of automation”, i.e., the level of autonomy with which the technical system pursues its functions and at which the human-machine system performs best (e.g., Endsley & Kaber, 1997; Parasuraman, Sheridan, & Wickens, 2000). Other researchers focus on specifying when the operator should be supported with automated functions to balance his/her current workload and, therewith, achieve a high overall performance level (e.g., Byrne & Parasuraman, 1996; Hancock, Chignell, & Lowenthal, 1985), which is defined on the basis of the task’s degree of fulfillment. Both fields, i.e., the static definition of an ideal level of automation and the adaptive allocation of functions to the human operator or the machine, have also been combined (see e.g., Kaber & Endsley, 2004). 1.2 Interface design Other approaches to enhance the performance of humanmachine systems aim at ameliorating the communication between the user and a machine. For this purpose, various guidelines for optimal interface design have been published. Examples are the Direct Manipulation Interface (DMI, Shneiderman, 1983), the Ecological Interface Design (EID, Vicente & Rasmussen, 1992), the Intuitive Interfaces (Baerentsen, 2000) or the Delegation-Type Interfaces (e.g., Parasuraman, Galster, Squire, Furukawa & Miller, 2005). The DMI has been defined by Shneiderman (1983) as an interface allowing the user to directly manipulate objects presented on the display. This manipulation should correspond at least loosely with the according manipulations in the physical/real world. The overall goal of the DMI is to make the interaction easier to learn, to give the operator incremental and rapid feedback, to complete tasks in less time and to make the overall system more dependable by reducing the number of human errors performed. The EID (Vicente & Rasmussen, 1992) extends the DMI and aims at providing optimal support for each level of cognitive control. The concept of cognitive control is based on the Skills, Rules, and Knowledge (SRK) model introduced by Rasmussen (1983), who distinguished three ways of interaction between human beings and their environment depending on the degree of novelty of a situation: 1) Skill-based behavior (SBB) is highly automated, unconscious behavior, representing fluid sensory-motor performance. The perceptual-motor system controls the human behavior. 2) Rule-based behavior (RBB) is controlled by rules or procedures, which are rules of thumb or effective knowhow. These rules are empirically derived informal cues that discriminate between the perceived action possibilities and allow choosing the supposedly best one without investigating great cognitive effort. 3) Knowledge-based behavior (KBB) takes place in unfamiliar, unanticipated situations, in which no rules are available (Vicente & Rasmussen, 1992). In such new situations, the human being formulates goals based on analyzing the environment and the overall aims. Based on these goal formulations, plans are developed and selected to achieve the goals. The effects of different plans are tested based on internal representations or by experiments. Proceedings of the 17th World Congress The International Federation of Automatic Control Seoul, Korea, July 6-11, 2008 978-1-1234-7890-2/08/$20.00 © 2008 IFAC 14779 10.3182/20080706-5-KR-1001.0974 To support these three levels of cognitive control, the following guidelines are provided by the EID (Vicente & Rasmussen, 1992): 1) SBB can be supported best if the interface provides the means to act directly on the display. The information on the display should be isomorphic to the structure of corresponding movements. 2) In order to support RBB, the interface should provide cues or signs which optimally map the constraints of the work domain in question. 3) To support KBB, the interface should display the relational properties of the work domain in the form of an abstraction hierarchy, which serves as an externalized mental model (see e.g., Vicente & Rasmussen, 1990). This mental model provides appropriate support for planning activities and thought experiments. The applicability of these guidelines have e.g. been empirically tested by Vicente, Christoffersen, and Pereklita (1995), and yielded experimental support. 1.3 Relevance of individual differences The known interface design guidelines and approaches to statically or dynamically adapt the automated functions to optimize the performance of the human-machine system only consider the user in a very general way but ignore differences between users (but see research on dynamically allocating automated functions to the machine or the operator depending on his/her current level of workload as conducted e.g., by Parasuraman, 1990). The importance of considering individual differences is especially at hand when discussing the EID: The three levels of cognitive control proposed by the SRK model closely resemble the three phases of skill acquisition (see e.g., Fleishman, 1972): The first phase of skill acquisition takes place when the user is confronted with a situation the first time: Attention is focused on thoroughly understanding the task in question, building a cognitive representation, and working out a (potentially successful) solution. Performance is slow and error-prone. The description of this phase resembles the KBB. When an adequate cognitive representation of the task has been built, the learner proceeds to the second phase of skill acquisition and easier ways of achieving the same result are defined. Rules are worked out and fine-tuned. RBB takes place. In the last phase of the skill acquisition process, performance is fast and accurate. The task is fully automated and can be completed without much attention. SBB is controlling human behavior. Ackerman (1988) explained the performance of human beings in these three levels of cognitive control (according to Rasmussen, 1983) or phases of skill acquisition (according to Fleishman, 1972) on the basis of individual differences in relevant abilities (see Fig. 1). The author proposed that performance in the first phase of skill acquisition (or KBB) is determined by general intelligence, performance in the second phase, i.e., the RBB by perceptual speed and the performance in the third phase, i.e., the SBB by motor abilities. General intelligence was defined by Ackerman (1988) in accordance to Humphreys (1979), as the ability to acquire, store, retrieve, combine, compare, and use information in new, other contexts. The core cognitive activity of perceptual speed is to generate rules to effectively solve easy tasks; hence, perceptual speed is the speed with which such simple rules can be implemented and compiled. Last, psychomotor abilities are independent from information processing and represent individual differences in the speed/accuracy of motor responses to tasks without information processing demands. Hence, highly intelligent operators will be advantaged, and make less serious errors when having to deal with an unknown and unanticipated situation compared to a less intelligent operator (see also Fig. 1): The ability/performance correlation is greater for the more intelligent users. This is especially important, as, according to Vicente and Rasmussen (1992) the problem solving behavior is particularly safetycritical when confronted with new situations. Ackerman’s skill acquisition theory (1988) implies that the new situation is less safety-critical for highly intelligent users but reducing the complexity of the situation and, thus, the need for general intelligence to be involved in working out the best solution, will also make the situation less safety-critical (Ackerman, 1988). This advantage of the highly intelligent users will disappear with the degree of familiarity of the situation, as other abilities will then determine performance. The relationship between the impact of errors and the higher cognitive processes involved has e.g., been shown empirically by Hammond, Hamm, Grassia, and Pearson (1987). Fig. 1. Phases of skill acquisition adapted from Ackerman (1988) and the impact of general intelligence on the performance/ability correlation. As intelligence is not considered a single ability construct (see also Section 4.1), but a complex structure of cognitive abilities, not only general intelligence but also the operator’s structure of intelligence is to be considered. The content abilities (i.e., verbal, numerical and figural intelligence) will impact the quality of interaction between the operator and the system’s interface, while the operation abilities (i.e., unknown 1st phase: highly intelligent user