A child's story to illustrate automated reasoning systems using opportunity and history

Abstract

The primary author has performed previous work to create generalized, high-level reasoning software to identify deception on the basis of actions. The work here is to apply that software architecture to an entirely different domain. The original domain was an espionage scenario, and detected deception when communications were in conflict. This present domain is reasoning about a child's story to determine who is lying about the theft of some objects. Applying the previous work to a different domain is an attempt to demonstrate the generality of the architecture. It is also an attempt to further generalize the software, and to formalize additional "common sense" strategies in the detection of deception. This software detects deception on the basis of actions. This is in sharp contrast with present approaches that detect deception based on physiological factors, as well as on verbal and non-verbal cues. Our approach models agents and their actions in a logic programming framework using a theory of agents, a theory of actions, and a theory of reasoning with respect to time. As a test case, a children's mystery is analyzed and implemented. The goal of the story is to identify who stole some items. The software correctly reasons about who the potential suspects are, and ultimately, correctly identifies the chief culprit. Further, it can correctly introspect with regard to previously held beliefs. The program we have developed is able to mimic the thought processes and conclusions of a police investigation. To demonstrate our approach, we have created a scenario loosely based upon a child's book. The basic idea of the story is that a bat is missing, and the goal is to identify who stole the bat. The flow of the story is as follows. There is a practice at the beginning of the story. A particular baseball bat is missing, and presumed stolen. Everyone present at that practice is a suspect in the theft of the bat. There is a subsequent practice at which a glove becomes missing. Everyone present at that subsequent practice is a suspect in the theft of the glove. At this point, there are two separate, but partially overlapping lists of suspects. At a later time, someone is discovered in possession of the glove. That person is assumed to have stolen the glove, and hence, the most likely person to have stolen the bat, since he was a suspect in the theft of both items. The events of our story occur over time. Our program can correctly represent and reason about these events. The events in the story are sequential with respect to time and so the experiments were performed in incremental fashion. They are presented here as four distinct executions. However, each execution completely subsumes the previous exectuion. As such, each execution contains all the results of the earlier executions. Execution 1 is trivial, but is necessary to demonstrate that our program reasons correctly, and does not enter the arena with unfair predispositions. The results of this execution tell us that there are no items missing, and hence no suspects. Subsequently, a practice occurs (execution 2.) There are four persons present at this practice. A bat becomes missing, and is assumed stolen. Those present at the practice are assumed to be suspects. Obviously, the owner of the bat is not a suspect. Therefore, there are three suspects in the theft of the bat. The program correctly identifies those persons. Our third execution takes place at a subsequent practice. There are five persons present at this practice: the same four persons that were present at the previous practice, plus one more person. The significant event that happens at this practice is that a glove is missing, and assumed stolen. There are four suspects in the theft of the glove. Again, the owner of the glove is not a suspect in the theft of the glove. The program correctly identifies who these suspects are, as well as correctly maintaining the suspects in the earlier theft. For our final execution, a specific individual is caught in possession of the glove. That person is therefore presumed to have stolen the glove. Further, that person is also presumed to be the chief suspect in the stealing of the bat since that person is also a suspect in that theft. The other two suspects in the theft of the bat still remain suspects in that theft. However, the person who is identified as having stolen the glove rises to the top of the list of the suspects in the theft of the bat. That person is the chief suspect, and is the only chief suspect in the theft of the bat. If something happened such that that person was no longer considered the chief suspect, then the two remaining suspects would resurface as primary suspects. The program correctly performs these inferences, matching our intuition. We have seen the ability of the program to reason with available, incomplete information. It correctly models our intuition. However, there are three very significant avenues by which this software could be enhanced. First, we could more closely employ an already well established theory of actions. Following this theory more closely, our actions could be more complicated. In addition, our actions could have prerequisites, and consequences. Certain actions could happen in parallel, and other actions could be mutually exclusive. We could predict the consequences of actions, and we could predict future actions.Another significant enhancement would be to follow the tri-axis of police investigations. That is, that suspects should have the means, motive, and opportunity. In our scenario here, we ignored the first two (means and motive), and we trivialized the latter (opportunity.) In our case, we considered that those who were present at practice had opportunity. What if someone was at practice, but was in the concession stand the entire time (meaning that they were nowhere near the bat)? This introduces the idea of degree of opportunity. Or, what about the opportunity a car rider may have had? (That is, someone who rode in the car with the owner and her bat, but who did not attend practice.) This expands our definition of opportunity. Another enhancement would be to apply this software to solve other mysteries. This pursuit would highlight other considerations. Further, the overlap between scenarios may identify opportunities for more general approaches. © Springer-Verlag Berlin Heidelberg 2006.