Commonsense Reasoning
Jon Barwise, Jeremy Seligman · Cambridge University Press eBooks · 1997
Among the problems that have beset the field of artificial intelligence, or AI, two have a particularly logical flavor. One is the problem of nonmonotonicity referred to in Part I. The other is the so-called frame problem. In this lecture, we suggest that ideas from the theory presented here, combined with ideas and techniques routinely used in state-space modeling in the sciences, suggest a new approach to these problems. Nonmonotonicity The rule of Weakening implies what is often called monotonicity: If Г ⊢ Δ then Г, α ⊢ Δ. The problem of nonmonotonicity, we recall, has to do with cases where one is disinclined to accept a constraint of the form Г, α ⊢ Δ even though one accepts as a constraint Г ⊢ Δ. The following is an example we will discuss in this lecture. Example 19.1. Judith has a certain commonsense understanding of her home's heating system – the furnace, thermostat, vents, and the way they function to keep her house warm. Her understanding gives rise to inferences like the following. (α 1 ) The thermostat is set between sixty-five and seventy degrees. (α 2 ) The room temperature is fifty-eight degrees. ⊢ (β) Hot air is coming out of the vents. It seems that α 1 , α 2 ⊢ β is a constraint that Judith uses quite regularly and unproblematically in reasoning about her heating system. However, during a recent blizzard she was forced to add the premise (α 3 ) The power is off.