Response to Hanks and McDermott: Temporal Evolution of Beliefs and Beliefs about Temporal Evolution

Steve Hanks and Drew McDermott (1986) describe a temporal projection problem that they believe (1) exhibits an important kind of reasoning for practical Artificial Intelligence systems and (2) cannot be handled by the existing non-monotonic inference systems. I don’t share either of these beliefs. In this note, I will point out why I am neither bothered by their temporal projection problem nor convinced by their analysis. The problem involves reasoning about a gun, known to be loaded at a time, and fired at a person at a later time. We want to know if the person ceases to live. We are willing to assume that if the gun remained loaded, then the firing was effectively fatal. But the kind of reasoning that permits us to conclude that the gun remained loaded until it was fired would also appear to allow reasoning to the conclusion that the person remained alive, even after the firing. One choice is to reason that the property of being alive persists, and hence, the property of being loaded does not. The other choice is to reason that the property of being loaded persists and the property of being alive does not. Symbolically, in an impoverished temporal representation (but one that is adequate for our purposes), we have (see Figure 1):