HUMAN ERROR MODELING PREDICTIONS: INCREASING OCCUPATIONAL SAFETY USING HUMAN PERFORMANCE MODELING TOOLS

The use of computer-aided job analysis tools has been increasing in the recent past as a result of decreases in computational costs, augmentations in the reality of the computer-aided job analysis tools, and usefulness of the output generated from these tools. One tool set known as integrated Human Performance Modeling (HPM) is a human- out-of-the-loop (HOOTL) computational methodology used to generate predictions of complex human-automation integration and system flow patterns. These tools provide computational representations of humans incorporating physical, cognitive, perceptual, and environmental characteristics. Increasingly complex automation leads to a new class of errors and error vulnerabilities. Hollnagel's (1993) Contextual Control Model (CoCoM) will be used as the human error theory behind a HOOTL simulation using Air Man-machine Integration Design and Analysis System (Air MIDAS) to evaluate complex human- automation integration considerations currently underway at NASA Ames Research Center. This paper will highlight the importance of the physical and cognitive link of a specific task and will outline attempts being made to understand the factors underlying human error, a critical consideration of human-complex system performance. 1. JOB ANALYSIS: PHYSICAL VERSUS COGNITIVE MODELS Current job analysis activities focus on the development of procedures that integrate ergonomic stresses across body parts of major interest (e.g., lower back, upper extremities, and neck) and allow in-plant teams to rank the seriousness of exposures across different jobs (Medsker & Campion, 1997). The themes that are examined in these exposures range from job design issues (self management, participation, task variety, significance and identity), to job interdependence of tasks, to job composition (flexibility), to job context (training, support, cooperation among members) and to job process issues (workload, social support, member cooperation). This method of analyzing the job is often subjective in nature and over-relies on the physical performance of a task making the process limited in cross-domain application. In these ergonomic programs, there is little significance given to the physical-cognitive job interactions that may occur. This method of analyzing a job assumes that human behavior is sequential when viewed in hindsight but this orderliness is really just an artifact of the asymmetry of time (Hollnagel, 2000). 2. HUMAN-OUT-OF-THE-LOOP (HOOTL) SIMULATIONS Many different forms of Human-Out-Of-The-Loop (HOOTL) simulations exist - these can range from anthropometric simulations of human performance to procedural static models, through to more complex dynamic representations of human performance within an operating environment. These latter techniques include integrated human performance models which use computer models of human performance where human characteristics, based on empirical research, are embedded within a computer software structure to represent the human operator (Laughery & Corker, 1997; Gore 2000). This virtual operator is then set to interact with computer-generated representations of the operating environment. HOOTL simulations can therefore be used at an earlier process in the development of a product, system or technology than waiting for the concept to be fully designed and used in practice (human in the loop tests). The system model development process allows the designer of the product, system or technology to fully examine many aspects of human-system performance with the new technologies. The model of human performance enables predictions of emergent behavior based on elementary perception, attention, working memory (WM), long-term memory (LTM) and decision-making models of human behaviors. This modeling approach focuses on micro models of human performance that feed-forward and feedback to other constituent models in the human system depending on the contextual environment that surrounds the virtual operator. These complex HOOTL simulation tools permit researchers to formulate procedures, generate hypotheses, and identify variables for Human in the Loop simulations (Gore & Corker, 2000b). One criticism of HOOTL tools has been that the software only predicts input-output behavior in mechanistic terms (Craik, 1947). The integrated and emergent structure of the tools however does more than solely represent input-output behavior, it attempts to prescribe how sequences of actions are planned and not simply prescribe a sequence of actions. The framework integrates many aspects of human performance