I thought Watson’s failures were more interesting than its successes-how the heck did it get “Toronto” for an answer to the first day’s Final Jeopardy? You would think the category being “U.S. Cities” would eliminate that immediately as a potential answer.
Yes.
In my early years I watched the "moving target" which artificial intelligence had become. Everybody had their own ideas of what would constitute "intelligence". As soon as people understood how to accomplish any given level of performance, the definition shifted to something more challenging. People were never impressed by a machine accomplishing something that they could understand.
Playing championship level chess is one example. Once it was done, people treat it like the brute force effort that it was. The computers, being able to analyze millions of positions per second, only needed a little algorithmic help to prune the search trees to make the most of that ability. People were not so impressed when they understood how it was done.
Recognizing that effect, I decided to change my personal definition of "arificial intelligence". It is any accomplishment by artificial means which CANNOT be fully understood.
I would assume that Watson maintained a "log" of its activities during the competition. Searching through that log, it should be possible for the creators of Watson to figure out what went wrong and to make corrections so that in future Watson will know the answer to that question. (Or, more correctly, the question to that answer.)
At some point in the future it will become impractical to make or examine such an activity log. At that point, we will have created artificial intelligence. Just as with real intelligence, it won't be correct all the time. And just as real intelligence, it won't be possible to understand exactly how it works.
I'm willing to believe that such an intelligence may require an artificial brain perhaps orders of magnitude more complex than the human brain.