Modeling cognitive errors in the realization of an activity of the everyday life

The Domus Laboratory is a multidisciplinary lab aimed at research in Domotics and Mobile Computer Science and located at Universite de Sherbrooke. Among the ongoing projects, let's mention the cognitive assistance to people suffering from Alzheimer type dementia, traumatic brain injury and other cognitive disorders. This assistance is dispensed through the realization of intelligent homes and mobile applications. When the resident carries out an activity of the daily living (ADL) in an inadequate way, the apartment explains to him which action it must realize to complete its ADL. In order to help according to the deficit, we must first be able to describe the typical inadequate actions carried out by people with cognitive disorders. One solution to understand the underlying mechanisms of the erroneous behavior is by cognitive modeling. More precisely, in this project we focus on simulating behavior of patients with brain lesions. The methodology employed first of all consists in classifying the various types of possible human errors observed in the literature. The second step consists in modeling ADL carried out by normals in a suitable way. Then the next step consists in disturbing in an automatic way the realization of the ADL in order to simulate the lesions. And finally, we compare our results with the clinical studies to validate the model. From our explored and reviewed errors taxonomies in the literature (for example see Reason 1990, Norman 1981, Schwartz 1998) we chose for our model Schwartz’s Multi-Level Action Test (MLAT) error taxonomy. We chose it because it’s already aimed at people with cognitive disabilities. Also, various studies using this (or a similar) classification have been published which offer a great deal of hints of the underlying causes of these errors and information on the distributions of error. The most important categories of the MLAT for our modeling purpose are omission of a key step, error of sequence, object substitution an action addition. More precisely, errors of sequence include anticipation-omission, and perseveration and reversal. The second step consists in realizing activities carried out adequately using ACT-R cognitive architecture. ACT-R is computational framework based on psychological experiments to simulate the behavior of the human cognitive system. The first selected activity was that “make a coffee with milk and sugar”, an activity which one frequently finds in the literature. This activity was modeled by integrating procedural knowledge, i.e. the steps to make the coffee and declarative knowledge, i.e. the objects and their associated properties in ACT-R. The actual status consists in adding the different type of errors in the coffee-making model. This will be carried out on one hand by adding procedural rules generating the errors, and on the other hand by applying levels of activation threshold to knowledge to allow slips and lapses. Theses mechanisms used are provided by the architecture of ACT-R. For example, omissions errors occur when the model wanders off-task when the attention is lost or when there’s a failure for objects to reach threshold. Perseveration are encoded as a failure to remember that the step as already been executed. Noise can also cause anticipations by skipping a preferred action before the main step, for example trying to pour water in a thermos before opening it. Similarities between objects could also cause the wrong object to be activated which result in an object substitution (e.g. stirring coffee with a fork instead of a spoon).