WEBVTT

00:00:09.840 --> 00:00:14.900
The graph shows on the
horizontal axis the percentage

00:00:14.900 --> 00:00:19.290
of questions answered,
and on the vertical axis,

00:00:19.290 --> 00:00:25.030
the percentage of answers
that were correct.

00:00:25.030 --> 00:00:33.620
When Watson started, the graph
that shows its performance

00:00:33.620 --> 00:00:35.990
is shown here.

00:00:35.990 --> 00:00:43.910
In 2007, we observed that the
performance of Watson improved,

00:00:43.910 --> 00:00:46.180
and it is shown there.

00:00:46.180 --> 00:00:55.670
And then successively
from 2008 May to 2010,

00:00:55.670 --> 00:00:59.100
we see the successive
improvement of Watson

00:00:59.100 --> 00:01:02.870
in this graph that shows a
trade-off between questions

00:01:02.870 --> 00:01:06.050
answered and precision.

00:01:06.050 --> 00:01:10.090
This is the area of
the best human players.

00:01:10.090 --> 00:01:14.450
Observe that they answer
a high number of questions

00:01:14.450 --> 00:01:17.500
and they answer a very high
number of questions correctly.

00:01:17.500 --> 00:01:22.670
So it was at that time--
2010-- that the people that

00:01:22.670 --> 00:01:25.350
controlled the decision
decided that Watson

00:01:25.350 --> 00:01:27.110
is ready to enter
the competition.

00:01:30.170 --> 00:01:33.950
So the games were scheduled
for February 2011.

00:01:33.950 --> 00:01:36.340
Two games were
played, and the winner

00:01:36.340 --> 00:01:40.200
would be the contestant with the
highest winnings over the two

00:01:40.200 --> 00:01:40.700
games.

00:01:44.570 --> 00:01:47.759
Once we got into the game, it
was nerve-wracking, frankly.

00:01:47.759 --> 00:01:50.400
Brad, if you're ready,
make your first choice.

00:01:50.400 --> 00:01:52.960
Let's take "Alternate
Meanings" for $200, Alex.

00:01:52.960 --> 00:01:56.890
"4-letter word for a
vantage point or a belief."

00:01:56.890 --> 00:01:57.390
Brad.

00:01:57.390 --> 00:01:58.060
What is a view?

00:01:58.060 --> 00:01:58.560
Yes.

00:01:58.560 --> 00:02:01.520
After the first clue of
the game, which Brad won,

00:02:01.520 --> 00:02:05.240
I had just this horrible
feeling at that moment

00:02:05.240 --> 00:02:07.660
that he was as good as
everyone said he was,

00:02:07.660 --> 00:02:11.200
and he was just going to
run the whole board on us.

00:02:11.200 --> 00:02:13.160
"Alternate Meanings," $400.

00:02:13.160 --> 00:02:16.170
"4-letter word for the iron
fitting on the hoof of a horse

00:02:16.170 --> 00:02:19.370
or a card-dealing
box in a casino."

00:02:19.370 --> 00:02:19.870
Watson.

00:02:19.870 --> 00:02:20.760
What is shoe?

00:02:20.760 --> 00:02:21.350
You are right.

00:02:21.350 --> 00:02:22.430
We actually took the lead.

00:02:22.430 --> 00:02:23.430
We were ahead of them.

00:02:23.430 --> 00:02:25.470
But then we started getting
some questions wrong.

00:02:25.470 --> 00:02:26.010
Watson.

00:02:26.010 --> 00:02:26.900
What is leg?

00:02:26.900 --> 00:02:28.450
No, I'm sorry, I
can't accept that.

00:02:28.450 --> 00:02:29.980
What is 1920s?

00:02:29.980 --> 00:02:30.550
No.

00:02:30.550 --> 00:02:31.800
What is chic?

00:02:31.800 --> 00:02:32.680
No, sorry.

00:02:32.680 --> 00:02:33.280
Brad?

00:02:33.280 --> 00:02:33.900
What is class?

00:02:33.900 --> 00:02:35.260
Class, you got it.

00:02:35.260 --> 00:02:36.030
Watson.

00:02:36.030 --> 00:02:37.090
What is Sauron?

00:02:37.090 --> 00:02:39.579
Sauron is right,
and that puts you

00:02:39.579 --> 00:02:43.960
into a tie for the
lead with Brad.

00:02:43.960 --> 00:02:45.340
A lot of people
were feeling good

00:02:45.340 --> 00:02:49.420
because Watson held his own
and was doing pretty well.

00:02:49.420 --> 00:02:51.780
But I wanted to be winning.

00:02:51.780 --> 00:02:54.910
Too close for comfort.

00:02:54.910 --> 00:02:56.040
Ken, you're in third place.

00:02:56.040 --> 00:02:58.500
That means you go first
in "Double Jeopardy."

00:02:58.500 --> 00:03:01.990
The "Double Jeopardy" round
of the first game I thought

00:03:01.990 --> 00:03:03.830
was phenomenal.

00:03:03.830 --> 00:03:05.010
Watson went on a terror.

00:03:05.010 --> 00:03:05.900
Watson.

00:03:05.900 --> 00:03:07.220
Who is Franz Liszt?

00:03:07.220 --> 00:03:07.950
You are right.

00:03:07.950 --> 00:03:09.070
What is violin?

00:03:09.070 --> 00:03:09.600
Good.

00:03:09.600 --> 00:03:11.490
Who is The Church Lady?

00:03:11.490 --> 00:03:12.330
Yes.

00:03:12.330 --> 00:03:13.290
Watson.

00:03:13.290 --> 00:03:14.610
What is narcolepsy?

00:03:14.610 --> 00:03:15.200
You are right.

00:03:15.200 --> 00:03:21.600
And with that, you
move to $36,681.

00:03:21.600 --> 00:03:25.329
After the first game, we have
a pretty good commanding lead,

00:03:25.329 --> 00:03:27.140
but we've seen enough
Jeopardy to know

00:03:27.140 --> 00:03:28.270
that this is not a lockout.

00:03:32.660 --> 00:03:37.510
Welcome to the deciding game
in the IBM Challenge here

00:03:37.510 --> 00:03:38.370
on Jeopardy!.

00:03:38.370 --> 00:03:40.920
The second game was
a whole other story.

00:03:40.920 --> 00:03:43.290
We struggled some
at the beginning.

00:03:43.290 --> 00:03:45.040
"Actors Who Direct" for $1,000.

00:03:45.040 --> 00:03:46.870
"A Bronx Tale."

00:03:46.870 --> 00:03:47.390
Brad.

00:03:47.390 --> 00:03:49.880
Who is Robert De Niro?

00:03:49.880 --> 00:03:50.460
Correct.

00:03:50.460 --> 00:03:53.010
That was an awful
category because Watson

00:03:53.010 --> 00:03:56.380
got every single one right,
but was just a little too slow.

00:03:56.380 --> 00:03:57.000
Ken.

00:03:57.000 --> 00:03:57.740
Who is Denzel Washington?

00:03:57.740 --> 00:03:58.230
You got him.

00:03:58.230 --> 00:04:00.790
Ken and Brad were getting
the answer in a second.

00:04:00.790 --> 00:04:02.580
"Rocky II", "III", & "IV."

00:04:02.580 --> 00:04:03.080
Brad.

00:04:03.080 --> 00:04:04.230
Who is Sylvester Stallone?

00:04:04.230 --> 00:04:04.970
Correct.

00:04:04.970 --> 00:04:09.310
Even with 2,800 cores, it took
us 2 1/2 seconds or 3 seconds,

00:04:09.310 --> 00:04:12.060
and that just gives you a sense
of how incredible the human

00:04:12.060 --> 00:04:14.610
brain is-- how quickly it
could understand the category,

00:04:14.610 --> 00:04:15.570
what's being asked.

00:04:15.570 --> 00:04:15.670
Boom!

00:04:15.670 --> 00:04:16.459
And get the answer.

00:04:16.459 --> 00:04:17.240
Who is Sean Penn?

00:04:17.240 --> 00:04:17.860
Right.

00:04:17.860 --> 00:04:19.870
Ken was doing very well.

00:04:19.870 --> 00:04:24.260
And the risk was if Ken gets
a "Daily Double," that's big.

00:04:24.260 --> 00:04:26.080
Gets it right, he's
going to be well ahead.

00:04:26.080 --> 00:04:28.670
And then with that kind of lead
going into "Final Jeaoprady"--

00:04:28.670 --> 00:04:31.300
if he bets enough, he could
end up winning the match.

00:04:31.300 --> 00:04:34.450
"A camel is a horse
designed by this."

00:04:34.450 --> 00:04:34.950
Ken.

00:04:34.950 --> 00:04:35.760
What's a committee?

00:04:35.760 --> 00:04:36.260
Good.

00:04:36.260 --> 00:04:37.650
"Familiar Sayings" for $2,000.

00:04:37.650 --> 00:04:41.390
We've gotta find that
last "Daily Double."

00:04:41.390 --> 00:04:42.740
I was nail-biting.

00:04:42.740 --> 00:04:44.340
It was just intense.

00:04:44.340 --> 00:04:46.440
Because these guys were good.

00:04:46.440 --> 00:04:46.940
Ken.

00:04:46.940 --> 00:04:47.710
Who is the brain?

00:04:47.710 --> 00:04:48.570
Brain, yes.

00:04:48.570 --> 00:04:49.550
There was a crucial
moment in the game.

00:04:49.550 --> 00:04:50.930
There was still a "Daily
Double" on the board

00:04:50.930 --> 00:04:53.450
and it was starting to
become pretty clear that it

00:04:53.450 --> 00:04:55.380
was in the "Legal E's" category.

00:04:55.380 --> 00:04:58.100
Let's go to "Legal
E's" for $1,200.

00:04:58.100 --> 00:04:59.980
"This person is
appointed by a testator

00:04:59.980 --> 00:05:03.840
to carry out the directions
and requests in his will."

00:05:03.840 --> 00:05:04.400
Watson.

00:05:04.400 --> 00:05:05.670
What is executor?

00:05:05.670 --> 00:05:06.600
Right.

00:05:06.600 --> 00:05:08.710
Same category, $1,600.

00:05:08.710 --> 00:05:14.210
Answer-- "Daily Double."

00:05:14.210 --> 00:05:17.740
That was the moment
when I knew it's over.

00:05:17.740 --> 00:05:21.400
Let us look at the
results of the game.

00:05:21.400 --> 00:05:23.630
Looking at these numbers,
it is fair to say

00:05:23.630 --> 00:05:28.100
that Watson's victory over
the two best human players

00:05:28.100 --> 00:05:29.840
has been a decisive one.

00:05:29.840 --> 00:05:34.340
In the first day,
Watson's winnings

00:05:34.340 --> 00:05:40.790
were more than double the sum
of the winnings of both players.

00:05:40.790 --> 00:05:47.050
And in the second day, it
was more than their sum.

00:05:47.050 --> 00:05:53.150
Overall, the winnings for
Watson were almost double

00:05:53.150 --> 00:05:56.090
the sum of the winnings
of both players--

00:05:56.090 --> 00:05:57.800
a rather decisive victory.

00:06:00.520 --> 00:06:04.050
So what's next for Watson?

00:06:04.050 --> 00:06:08.010
So Watson is ideally suited
to answer questions which

00:06:08.010 --> 00:06:10.960
cover a wide range
of material and often

00:06:10.960 --> 00:06:15.250
have to deal with inconsistent
or incomplete information.

00:06:15.250 --> 00:06:19.390
This makes it particularly
appropriate for applications

00:06:19.390 --> 00:06:20.920
like medicine.

00:06:20.920 --> 00:06:24.050
Note that the amount of
medical information available

00:06:24.050 --> 00:06:26.800
is doubling every five
years, and a lot of the data

00:06:26.800 --> 00:06:29.310
is unstructured.

00:06:29.310 --> 00:06:35.380
As a result, Watson has been
utilized for cancer diagnosis

00:06:35.380 --> 00:06:38.150
and selecting the best
course of treatment.

00:06:38.150 --> 00:06:41.130
And Watson has
been in discussions

00:06:41.130 --> 00:06:46.710
with MD Anderson and Memorial
Sloan-Kettering Cancer Centers

00:06:46.710 --> 00:06:51.310
to apply the Watson technology
for cancer diagnosis

00:06:51.310 --> 00:06:53.780
and selecting the best
course of treatment.

00:06:57.920 --> 00:07:01.370
So what is the analytics
edge of Watson?

00:07:01.370 --> 00:07:05.530
First, we have observed that
the major technology of Watson

00:07:05.530 --> 00:07:07.690
is to combine many
algorithms that

00:07:07.690 --> 00:07:10.470
increase the accuracy
and the confidence

00:07:10.470 --> 00:07:12.090
of the overall outcome.

00:07:12.090 --> 00:07:14.410
Any one of these algorithms
would not have worked,

00:07:14.410 --> 00:07:20.940
but their combination gives
Watson a considerable edge.

00:07:20.940 --> 00:07:23.920
The second important observation
that gives Watson an edge

00:07:23.920 --> 00:07:26.300
is that it approaches the
problem in a very different way

00:07:26.300 --> 00:07:29.390
than how a human
approaches a problem.

00:07:29.390 --> 00:07:33.140
A key aspect of the technology
is hypothesis generation,

00:07:33.140 --> 00:07:35.980
and then using different
and distinct algorithms

00:07:35.980 --> 00:07:39.170
to explore this hypothesis
and combining them,

00:07:39.170 --> 00:07:43.320
it leads to answers
that provide an edge.

00:07:43.320 --> 00:07:45.580
Finally, and quite
importantly, Watson

00:07:45.580 --> 00:07:48.440
has a considerable
ability to deal

00:07:48.440 --> 00:07:51.670
with massive amounts of data,
often in unstructured form.

00:07:51.670 --> 00:07:55.440
And the vast majority of the
data-- 90% of it-- is, in fact,

00:07:55.440 --> 00:07:56.409
unstructured.

00:07:56.409 --> 00:07:59.300
So the combination
of these ideas

00:07:59.300 --> 00:08:02.830
give Watson a significant edge.

00:08:02.830 --> 00:08:08.790
It is no accident that IBM has
invested so far more than $24

00:08:08.790 --> 00:08:12.680
billion to invest in
Watson technologies,

00:08:12.680 --> 00:08:16.190
and it has created
a company within IBM

00:08:16.190 --> 00:08:20.080
called Watson Solutions,
whose key objective is

00:08:20.080 --> 00:08:22.580
to apply the Watson
technology and the edge

00:08:22.580 --> 00:08:26.570
that it creates to a variety
of verticals, one of which

00:08:26.570 --> 00:08:28.280
is medicine.