What do all death threats have in common? Only their meaning, it seems. And meaning is not like radioactivity or acidity, a property readily discriminated by a well-tuned detector. The closest we have come yet to creating a generalpurpose meaning-detector is IBM’s Watson, which is much better at sorting by meanings than any earlier artificial intelligence system, but notice that it is not at all simple, and would still (probably) misidentify some candidates for death threats that a child would readily get. Even small children recognize that when one laughing kid yells to another, “So help me, I’ll kill you if you do that again!” this is not really a death threat. The sheer size and sophistication of Watson are at least indirect measures of how elusive the familiar property of meaning is.
Philosopher Daniel Dennett's Book Intuition Pumps