Opportunities for AI in the use of Computer-Based

We have had experience and considerable success in the design of computer-based systems which enable people with cognitive disabilities to partially overcome them. This success leads to problems for which AI may make substantial contributions. In addition, our experience in related fields may well be helpful to AI researchers who wish to work with individuals who have cognitive impairments. Introduction For the past decade, our labs have focused on the application of computer science to assist individuals who have cognitive disabilities due to brain injury and mental retardation. We have developed an approach to brain injury rehabilitation which we call Computer Based Cognitive Prosthetics. While we have not applied AI theory, our findings and experience may be of interest to people in the AI community. The idea of using computer technology to assist people with cognitive deficits was recognized independently in the mid-80s by three groups working in different disciplines: Cole et al (1990) 5 in computer science; Friedman at CMU working in electrical engineering and robotics, and Levine and Kirsch (1987) 2’3 from the UMich working in biomedical engineering and neuropsychology respectively. Cole’s early work addressed the predicament of people who had received a brain injury and had little prospect of being cured of various cognitive disabilities. These include disorders of memory, concentration, organization, planning, task sequencing, language, and eye-hand coordination, among others. Typically, these very cognitive disabilities prevented these individuals from learning and using off-the-shelf software. His lab, the Institute for Cognitive Prosthetics (ICP), has developed Computer-Based Cognitive Prosthetics (CBCP) as a means of providing personal productivity software to individuals who needed assistance in performing every day activities. Surprisingly, this approach is novel to the field of brain injury rehabilitation, where computers are used primarily for cognitive skill drills. Use of Computer Technology in Cognitive Rehabilitation Computers are now widely used in cognitive rehabilitation treatment. Most of the software available consists of cognitive skill drills, one skill at a time. Unfortunately, skill definition has not kept up with trends in experimental neuropsychology and neuroscience, and some cognitive skill dimensions rely on decades old work. Rehabilitation therapists will also use offthe-shelf computer games as a means of retraining cognitive skills. Off-the-shelf educational software is also used in rehab. 19 From: AAAI Technical Report FS-96-05. Compilation copyright © 1996, AAAI (www.aaai.org). All rights reserved. When there is a recognition that a cognitive skill loss may be long-term, patients are taught to use new methods of performing an activity, an approach which is called a "compensatory strategy". For memory loss, a therapy may involve using a To Do List, an appointment book and other manual means. There is a widespread feeling that patients should not be made to depend on technology for compensatory strategies. A few leaders in the field suggest the use of diverse electronic aids which are often found in electronic technology stores. However, both of these approaches have substantial problems. Success at skill training too often does not transfer to real life activities. Electronic aids are too difficult to learn and too unforgiving in error conditions. There would seem to be many opportunities for applying computer technology to rehabilitation of cognitive disorders. ICP’s approach to acquired cognitive disabilities from brain injury Our early work addressed the question of whether computer software could bridge cognitive deficits. The goal of our work with patients is to enable them to perform activities that they have not been able to perform since their brain injury. This problem is important because of the difficulty current rehabilitation techniques have in restoring the individual to their pre-injury level of activity. The research strategy focused conditions which were particularly difficult for current clinical state of the art. Traumatic brain injury (TBI) provided this setting. TBI typically causes diffuse damage, --to several lobes of the brain, and to many cognitive dimensions -but only partial damage. This leaves the individual with simultaneous deficits in several dimensions, but with none of the dimensions being totally eliminated. Also, TBI patients were expected to have no further recovery beyond 2 years of the accident. This would provide a baseline for our work. Cognitive engineering models of Card Moran & NewelP, and of Olsen and Olsen5 seemed to provide a heuristic. This work conceptualized an individual’s target activity as a lengthy sequence of cognitive tasks on a time frame of hundredths of a second to a second for expert users. One can then conclude that a disruption of the sequence for even a single cognitive skill would disrupt activity performance. We hypothesized that computer software could be constructed to perform the specific cognitive tasks in the sequence. In this way, the computing system would become prosthetic, taking over human functioning which the individual previously could do with their natural ability, but because of disease process was no longer able to