Logic, as classically formulated, is a crystal-clear, rock-hard structure seemingly ill-adapted to the vagaries of everyday human reasoning processes. In classical logic, truth, consequence and consistency are all-or-nothing affairs; in real life, by contrast, we vacillate, jump to conclusions, change our minds, and are seldom utterly certain about anything. There has thus arisen a perception that if logic is to serve Artificial Intelligence (AI) in any significant way, some method must be found of reconciling the conflicting demands of classical rigour and everyday woolliness.
One possibility which suggests itself is to remould or regiment everyday reasoning into the forms required by classical logic. This is to insist that we replace our ordinary loose ways of speaking by something more precise and logically correct.
Thus, to adapt an example cited in Michael Clarke's paper in this issue, consider the two statements:
If there is sugar in the coffee then the coffee will taste good. If there is sugar in the coffee and there is diesel-oil in the coffee then the coffee will not taste good.
According to classical logic, these statements together entail that there is not both sugar and diesel-oil in the coffee, but intuitively they do not entail this. Using the strategy of regimentation, we note that the utterer of the first statement does not literally mean exactly what he says, but rather something like If there is sugar in the coffee and no other additions then the coffee will taste good.
Thus reformulated, this statement is no longer logically inconsistent with the second statement.
Again, instead of inferring 'Tweety can fly' from 'Tweety is a bird' and 'Normally, birds can fly'--an inference which would have to be rescinded in the event of discovering that Tweety was a penguin --we infer instead the more modest conclusion 'If Tweety is a normal bird, then she can fly', which remains true even in the 'abnormal' case that Tweety is a penguin. (Note, incidentally, that the name 'Tweety' strongly suggests a song-bird, and hence a 'normal', i.e. flying, bird. Few penguin-owners, I venture to suggest, would choose to name their pet 'Tweety'. This additional dimension to the process of nonmonotonic inference seems to have escaped the attention of most writers who have used this popular example.)
This way of doing things, to remodel everyday reasoning so that it fits into the mould of classical logic, has not proved popular with AI researchers. There is a widespread suspicion that such regimentation cannot, in the end, be made to work,