When Razib interviewed Nick Cassimatis about AI, it was an interesting, and I think sensible definition that AI is getting computers to do things that we call intelligent. They went on to discuss that biologists study life, and while there are definitions of what "life" is (are viruses life?), not one studying them ever thinks much about what the definition is, even those studying viruses. Similarly, the SCOTUS may have had a hard time precisely defining pornography, falling back on a rather weak "I know it when I see it" decades ago, as a practical matter, no one working on a porn site stops to ask themselves "Hey, is this pornography?"
So too with intelligence, and therefore, AI. If we went to observe a fifth grade class, or the staff of a catering service, or a station of firefighters for a week each, we would have considerable overlap of which ones we considered intelligent, even absent a definition or a test. We migh add one who did deserve it or leave one off who did, but we would have enormous agreement, and some discussions about the disagreements might resolve quickly. "She is so nervous and timid that it takes a while to see it, but look what she did with that _____ situation." "He is clearly knowledgeable and has worked very hard, but his information is already out-of-date and he doesn't seem that able to adapt."
Life, intelligence, AI, pornography - these are all hard to get words around to define, but as a practical matter, we can plow on, all understanding each other except in marginal circumstances.
3 comments:
It is true that one rarely encounters an argument about such basic questions; we've known since Socrates that it's often the words we are sure we understand that prove most difficult to define.
"Life," at least, was successfully explained by Hans Jonas. (https://grimbeorn.blogspot.com/2022/05/philosophy-on-abortion.html). The definition is not well publicized in large part because it proves inconvenient.
Once a technology is in ordinary, daily use, it is no longer considered artificial intelligence. 30 years ago, a reliable speech recognition system would have been considered AI. Now, it's just something that comes with your phone.
To David Foster's point, what we're calling "AI" right now is linguistic pattern matching that anticipates the next significant word in a phrase. Glorified spell-check. Useful in fairly well-controlled conditions but a disaster in free-form queries for reasons that should be obvious.
Post a Comment