1 00:00:09,840 --> 00:00:15,210 This is a lecture of how IBM built a computer that plays 2 00:00:15,210 --> 00:00:18,100 the very popular game of Jeopardy!. 3 00:00:18,100 --> 00:00:23,160 In 2004, IBM Vice President Charles Lickel and coworkers 4 00:00:23,160 --> 00:00:24,890 were having dinner at a restaurant. 5 00:00:24,890 --> 00:00:28,160 All of a sudden, the restaurant fell silent. 6 00:00:28,160 --> 00:00:30,770 Everyone was watching the game Jeopardy! 7 00:00:30,770 --> 00:00:32,210 on the television. 8 00:00:32,210 --> 00:00:35,040 A contestant, Ken Jennings, was setting the record 9 00:00:35,040 --> 00:00:39,690 for the longest winning streak of all time, 75 days. 10 00:00:39,690 --> 00:00:41,920 Why was everyone so interested? 11 00:00:41,920 --> 00:00:42,420 Jeopardy! 12 00:00:42,420 --> 00:00:46,080 is a quiz show that asks complex and clever questions like puns, 13 00:00:46,080 --> 00:00:48,940 obscure facts, and uncommon words. 14 00:00:48,940 --> 00:00:51,380 It originally aired in 1964. 15 00:00:51,380 --> 00:00:53,330 It covers a huge variety of topics, 16 00:00:53,330 --> 00:00:55,630 and it's generally viewed as an impressive feat 17 00:00:55,630 --> 00:00:57,600 to do well in this game. 18 00:00:57,600 --> 00:01:01,040 No computer system has ever been developed that could even come 19 00:01:01,040 --> 00:01:03,630 close to competing with humans on Jeopardy!. 20 00:01:07,030 --> 00:01:11,370 IBM Research strives to push the limits of science. 21 00:01:11,370 --> 00:01:14,560 It has a tradition of addressing inspiring and difficult 22 00:01:14,560 --> 00:01:16,640 challenges over the years. 23 00:01:16,640 --> 00:01:20,300 In the mid '90s, it built Deep Blue, a computer, 24 00:01:20,300 --> 00:01:23,060 to compete against the best human chess players, 25 00:01:23,060 --> 00:01:25,530 and then in the mid '90s, Deep Blue 26 00:01:25,530 --> 00:01:28,530 beat Garry Kasparov, who was at that time 27 00:01:28,530 --> 00:01:31,260 the world champion in chess. 28 00:01:31,260 --> 00:01:34,650 Later it built Blue Gene, a computer 29 00:01:34,650 --> 00:01:36,740 to map the human genome. 30 00:01:40,600 --> 00:01:44,310 In 2005, a team at IBM Research started 31 00:01:44,310 --> 00:01:48,680 creating a computer that would compete at Jeopardy! 32 00:01:48,680 --> 00:01:51,870 Of course, no one knew at the time how to beat humans 33 00:01:51,870 --> 00:01:54,610 or if it was even possible. 34 00:01:54,610 --> 00:01:57,610 Six years later, a two-game exhibition match 35 00:01:57,610 --> 00:01:59,039 aired on television. 36 00:01:59,039 --> 00:02:01,390 The winner would receive a million dollars. 37 00:02:03,900 --> 00:02:09,030 The contestants were Ken Jennings, the longest winning 38 00:02:09,030 --> 00:02:13,550 champion, whose longest winning streak was 75 days; 39 00:02:13,550 --> 00:02:17,940 Brad Rutter, who was the biggest money winner of over 3.5 40 00:02:17,940 --> 00:02:21,850 million; and Watson, a supercomputer with 3,000 41 00:02:21,850 --> 00:02:26,000 processors and a database of 200 million pages of information. 42 00:02:27,770 --> 00:02:31,720 This is Jeopardy!-- The IBM Challenge. 43 00:02:33,690 --> 00:02:39,110 And now, here is the host of Jeopardy!, Alex Trebek. 44 00:02:41,480 --> 00:02:42,560 Thank you, Johnny Gilbert. 45 00:02:42,560 --> 00:02:44,540 Thank you, ladies and gentlemen. 46 00:02:44,540 --> 00:02:48,280 And welcome, everyone, to a very special Jeopardy! event. 47 00:02:48,280 --> 00:02:50,010 For the next three days, we're going 48 00:02:50,010 --> 00:02:53,430 to be coming to you from this IBM Research facility just 49 00:02:53,430 --> 00:02:55,490 outside of New York City. 50 00:02:55,490 --> 00:02:57,210 And let me tell you why. 51 00:02:57,210 --> 00:02:59,590 A little over three years ago, the folks at IBM 52 00:02:59,590 --> 00:03:02,320 came to us with a proposal that they considered 53 00:03:02,320 --> 00:03:05,550 to be the next grand challenge in computing. 54 00:03:05,550 --> 00:03:08,280 And that was designing a computer system 55 00:03:08,280 --> 00:03:10,120 that could understand the complexities 56 00:03:10,120 --> 00:03:12,700 of natural language well enough to compete 57 00:03:12,700 --> 00:03:15,240 against Jeopardy!'s best players. 58 00:03:15,240 --> 00:03:17,760 Well, they think they've succeeded, 59 00:03:17,760 --> 00:03:19,270 and that's why we're here today. 60 00:03:19,270 --> 00:03:23,340 So you are about to witness what may prove to be an historic 61 00:03:23,340 --> 00:03:27,200 competition -- an exhibition match pitting an IBM computer 62 00:03:27,200 --> 00:03:30,710 system against the two most celebrated and successful 63 00:03:30,710 --> 00:03:33,050 players in Jeopardy! history. 64 00:03:33,050 --> 00:03:34,620 Sounds like a lot of fun, doesn't it? 65 00:03:34,620 --> 00:03:37,570 Developed and programmed especially for this moment, 66 00:03:37,570 --> 00:03:40,840 making its first appearance on our national television 67 00:03:40,840 --> 00:03:43,740 program, ladies and gentlemen, this is Watson. 68 00:03:49,210 --> 00:03:52,510 Just as I expected, that was a very warm reception. 69 00:03:52,510 --> 00:03:55,470 And I'm sure Watson would've appreciated 70 00:03:55,470 --> 00:03:57,910 the applause, except for one thing. 71 00:03:57,910 --> 00:04:01,340 Watson can neither hear nor see. 72 00:04:01,340 --> 00:04:04,360 It will be receiving all of its information electronically. 73 00:04:04,360 --> 00:04:08,120 And as a matter of fact, what you're looking at right now 74 00:04:08,120 --> 00:04:10,670 is not the real Watson. 75 00:04:10,670 --> 00:04:12,790 This is an avatar. 76 00:04:12,790 --> 00:04:15,400 This is a representation of Watson. 77 00:04:15,400 --> 00:04:18,769 Watson, of course, is a sophisticated computer system 78 00:04:18,769 --> 00:04:21,740 too big and too heavy to fit behind that lectern 79 00:04:21,740 --> 00:04:23,520 on our stage. 80 00:04:23,520 --> 00:04:25,070 But it is close by. 81 00:04:25,070 --> 00:04:26,510 It's right next door. 82 00:04:26,510 --> 00:04:30,260 And a little while ago, I paid it a visit. 83 00:04:30,260 --> 00:04:31,920 The very first thing you will notice 84 00:04:31,920 --> 00:04:36,470 as I come into Watson's space is noise. 85 00:04:36,470 --> 00:04:38,600 There is a lot of noise. 86 00:04:38,600 --> 00:04:42,370 A little bit comes from Watson himself, but most of the noise 87 00:04:42,370 --> 00:04:45,030 comes from two very large refrigerator 88 00:04:45,030 --> 00:04:47,980 units that help to keep Watson cool. 89 00:04:47,980 --> 00:04:53,310 Now as you can see, Watson has been set up in two units. 90 00:04:53,310 --> 00:04:57,220 Each half contains five separate racks. 91 00:04:57,220 --> 00:05:05,190 Each of these racks contains 10 IBM Power 750 servers. 92 00:05:05,190 --> 00:05:06,840 Now when you link all of these servers 93 00:05:06,840 --> 00:05:09,770 together, as they have done for Watson, 94 00:05:09,770 --> 00:05:13,090 you create a deep analytic system 95 00:05:13,090 --> 00:05:18,100 that is the equivalent of 2,800 powerful computers tied 96 00:05:18,100 --> 00:05:21,860 together in a super high speed network.