Reasoning about Intended Actions

In most research on reasoning about actions and reasoning about narratives one either reasons about hypothetical execution of actions, or about actions that actually occurred. In this paper we first develop a high level language that allows the expression of intended or planned action sequences. Unlike observed action occurrences, planned or intended action occurrences may not actually take place. But often when they do not take place, they persist, and happen at an opportune future time. We give the syntax and semantics for expressing such intentions. We then give a logic programming axiomatization and show the correspondence between the semantics of a description in the high level language, and the answer sets of the corresponding logic programming axiomatization. We illustrate the application of our formalism with respect to reasoning about trips. Introduction and Motivation In reasoning about actions (for example, (?; ?) and reasoning about narratives we often reason about action sequences that are executed in a particular situation, or actions that happened at particular time points. Alternatively, there have been some work on reasoning about natural actions (?) and actions that are triggered. In this paper we consider intended execution of actions and formalize how to reason about such intentions. To motivate this further, consider a narrative where an agent intended to execute action a at time point i. A commonsense reasoner looking back at this intention would conclude that the agent must have executed action a at time point i. To ground this example, consider that the wife of our reasoner says that she intends to leave work at 5 PM. At 6 PM the commonsense reasoner would conclude that his wife must have left at 5 PM. Now suppose the reasoner checks his email and finds an email from his wife saying that she has been held up in a meeting and later gets information that the meeting ended at 5:30. The reasoner would then conclude that his wife must have left at 5:30 PM. I.e., her intended action, since it became impossible at the initially intended time point, must have persisted and executed at the next time point when it became executable. Copyright c © 2005, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. Now let us generalize this to a sequence of actions where an agent intends to execute a sequence a1, . . . , an at time point i. Now what if it happens (the world evolved in such a way) that the executability condition of ak is not true at the time point where ak−1 ended. Does this mean the agent abandoned his intention to execute ak, . . . , an? It seems to us that most agents, if they failed to execute their intended action ak after the execution of ak−1, would execute ak in the next possible time point when it became executable. As before, let us consider a more grounded example. John is supposed to have taken flight A to B and then take a connection from B to C. Suppose Peter finds out that John’s flight from A to B was late. Once Peter knows when exactly John reached B, his reasoning would be that John would have taken the next flight from B to C. In other words, failure to go from B to C at a particular time point, does not mean that John would have abandoned his intention to go from B to C; rather most likely he would have just done it at the next possible time point. This actually happened to one of the authors. He correctly guessed that his wife would take the next flight (after missing a connection) and was able to meet her at the airport when the next flight arrived. In most earlier work on reasoning about actions and narratives (for example, (?)), if one or many of the actions in a given sequence a1, . . . , an are not executable or otherwise prevented from execution then the reasoning process rigidly assumes that either the actions were not executed or considers the domain to be inconsistent. The formulation there is appropriate with respect to the assumptions in those languages. Here we consider the new notion of “intended (or planned) execution of actions”, which needs a different formalization. In this we can take pointers from prior studies on intentions (?; ?). In particular, intentions have been studied from the point of view of the design of rational agents (?), and they are one of the three main components of BDI (Belief-Desire-Intention) agents. In (?), various properties of ‘intentions’ of a rational agent is discussed. In particular the author says: Summarizing, we can see that intentions play the following important roles in practical reasoning • Intentions drive means-ends reasoning. If I have formed an intention, then I will attempt to achieve the intention, ... • Intentions persist. I will not usually give up on my intentions without good reason – they will persist, ...