Counter Deception

This report describes research on the problem of providing counter- deception support for expert systems. An expert system is deceived only if its product e.g., a situation assessment contains errors that derive from intentional actions of the enemy to mislead the system and which give the enemy an advantage. The possibility of deceiving these systems constitutes a serious threat to national security, particularly as more and more expert systems are being fielded to support military planning, intelligence analysis, and decision- making. Since the government speculates that it may be relatively easy to deceive some expert systems, there is a need for basic research on the possibility of strengthening expert systems so that they are less sensitive to deception. If the enemy can acquire a copy of the expert system then his problem is that much simpler however, he does not need a copy to be an effective deceiver Ultimately, the technical problem is that the expert system lacks an explicit representation for deception and thus has no computational basis to examine its conclusions in this light. We believe that deception efforts actually target not only the expert system but also the sensor suite that provides data for the expert system and the human analyst, planner, or decision maker that uses the tool. It is this larger system, comprised of the sensors, expert system, and human users that must be supported. Counterintelligence Countermeasures. edc