<p>Odd wagers–that is the other thing that really stood out about Watson. Like, on the 2nd day, he was ahead by 30K+ and he wagered something like 987 in Final Jeopardy. </p>
<p>Well, he did answer wrong, and his answer didn’t even fit the category of “U.S. Cities,” which was strange considering he had more time to find the answer. (The question about the two largest airports in which city being named for a WWII battle and WWII hero is a good example of the type of more complex question that is harder for Watson to answer.) </p>
<p>BUT (back to the odd wagers. . .) when considering only the category when making the wager–since Watson knew virtually everything else that day, and there was no way the other players could approach his total even if they doubled their money–it would’ve made sense to wager more. It seems that this type of “wagering logic” would be easier to program than everything else they’ve done.</p>
<p>It would be a fun to analyze the kinds of questions Watson couldn’t answer and write more of them. (I’m sure the IBM guys are already working on the other side of that, too.)</p>
<p>MommaJ, the spouse and I busted a gut laughing about “I, for one, welcome our new computer overlords.” And we are still laughing about the “what is a ho?” line.</p>
<p>Brad Rutter is the most dominant (human) player in the history of Jeopardy. Prior to last night he had never lost a game and had defeated every great champion who came against him, including Ken Jennings and many other big hitters. In fact among Jeopardy players his name had become a verb. Getting “ruttered” means being totally helpless to change the inevitable outcome - getting just totally crushed despite being in top form and giving it all your very best efforts. Last night Brad Rutter himself finally found out what it feels like to get ruttered.</p>
<p>In my experience, human reaction time is close to 0.1-0.3 seconds (from timing athletic events - our watches were always about 0.2 seconds slower than the electronic display). Was something like that programmed into this Watson thing?</p>
<p>^^No, there was no intentional delay programmed into Watson. It was free to respond the very instant the buttons were activated. The only time it delayed was for a few clues on which was still “thinking” about what its answer should be.</p>
<p>Just to show what a geek I am, I actually read an article discussing the buzzer factor, as written in the IBM white paper by David Ferrucci (one of Watson’s proud papa). I am paraphrasing here: the average reaction time of most people to light is around 190ms. But top players in Jeopardy do not actually react to the changing lights, but they anticipate the end of Alex Trebek reading tone so well, that they have been timed under 20, 10 or even 4 ms. Watson buzzer time is between 5ms to 10ms. So top players like Ken Jennings, who has been know to out-buzzed opponents as much as 80% during his streaks, can buzz faster than Watson at times. The big advantage is Watson NEVER buzzed early and is consistent so it rarely makes mistakes like human beings.</p>
<p>The IBM people also discussed that they deliberately built in a slight delay in buzzing time based on how confident Watson is with its top guess. That’s why when the percentage is low, is does not buzz at all. But at 97% or so… well, bow to the computer overlord, y’all!</p>
<p>I admit that I only watch Jeopardy occasionally so I was more riveted by Watson performance than the game itself. After watching the PBS Nova special “The Smartest Machine on Earth” (available online now and highly recommended), I attached the faces of all the IBMers to Watson and rooted for Watson to win just because I want their hard work to be validated. I know, I know… it’s time for me to log off and go into the sun rejoining the human race…</p>
<p>This was the big topic of conversation at our Chamber meeting last night - a little off-topic here, but after the Jeopardy discussion we discussed other potential challenges for Watson…</p>
<p>Watson versus a panel of physicians making diagnoses
Watson on the trading floor on Wall Street buying and selling
Watson making split-second decisions alongside airline pilots based on control readings</p>
<p>That contest might be a little circular. Who would determine what the correct diagnoses really were in order to call answers right and wrong? Probably another panel of physicians.</p>
<p>I think Watson’s oddball betting amounts were IBM humor in action. Watson had a sound betting general strategy - he bet big when he should have and bet small when he should have. But the odd bets like “$947,” or whatever it was, sounded like humor to me.</p>
<p>Actually, I would love to have seen what Watson would have done in a really sticky betting conundrum, but he never faced one. He was always comfortably ahead. </p>
<p>There have been scientific papers published in scholarly journals about Jeopardy betting strategy. Most of them revolve around complex choices and probabilities facing players in second place. Watson never spent much time in second place.</p>
<p>Watson is a cumulative effort of last 30 years of research in AI, language processing and machine learning. I am happy to see that still there are companies like IBM who would sink millions and several man years on a project that could have failed miserably.</p>