1
00:00:06,000 --> 00:00:09,100
IBM's Watson is at the forefront of a new 
era of computing:

2
00:00:09,110 --> 00:00:12,100
Cognitive computing. It's a radically new 
kind of

3
00:00:12,110 --> 00:00:14,700
computing, very different from the programmable 
systems

4
00:00:14,710 --> 00:00:17,020
that preceeded it. As different as those 
systems were from

5
00:00:17,020 --> 00:00:20,000
the Tabulating Machines of a century ago.

6
00:00:20,100 --> 00:00:24,100
Conventional computing solutions, based on 
mathematical principals

7
00:00:24,200 --> 00:00:26,600
that emanate from the 1940s, are programmed

8
00:00:26,610 --> 00:00:30,700
based on rules and logic, intended do derive 
mathematically precise answers

9
00:00:31,600 --> 00:00:33,950
often following a rigid decision tree approach.

10
00:00:33,950 --> 00:00:36,500
But, with today's wealth of big data, and 
the need

11
00:00:36,510 --> 00:00:39,400
for more complex evidence-based decisions, 
such a

12
00:00:39,410 --> 00:00:43,900
rigid approach often breaks, or fails to 
keep up with available information.

13
00:00:43,910 --> 00:00:47,200
Cognitive computing enables people to create 
a profoundly new

14
00:00:47,300 --> 00:00:49,500
kind of value, finding answers in insights

15
00:00:49,600 --> 00:00:51,500
locked away in volumes of data.

16
00:00:51,600 --> 00:00:54,400
Whether we consider a doctor diagnosing a 
patient,

17
00:00:54,410 --> 00:00:58,000
a wealth manager advising a client on their 
retirement portfolio,

18
00:00:58,100 --> 00:01:01,200
or even a chef creating a new recipe, they 
need new approaches

19
00:01:01,210 --> 00:01:05,200
to put into contacts the volume of information 
they deal with

20
00:01:05,210 --> 00:01:07,010
on a daily basis in order to derive value 
from it.

21
00:01:07,010 --> 00:01:10,990
This process serves to enhance human expertise.

22
00:01:10,990 --> 00:01:13,010
Watson and its cognitive capabilities mirror 
some of the key

23
00:01:13,010 --> 00:01:16,990
cognitive elements of human expertise.

24
00:01:16,990 --> 00:01:19,990
Systems that reason about problems like a 
human does

25
00:01:19,990 --> 00:01:21,500
when we, as humans, seek to understand something,

26
00:01:21,510 --> 00:01:25,000
and to make a decision, we go through four 
key steps:

27
00:01:25,010 --> 00:01:28,400
first, we observe visible phenomena and bodies 
of evidence;

28
00:01:28,410 --> 00:01:32,200
second, we draw on what we know to interpret 
what we're seeing,

29
00:01:32,210 --> 00:01:34,800
to generate hypothesis about what it means;

30
00:01:34,810 --> 00:01:38,900
third, we evaluete which hypothesis are right 
or wrong;

31
00:01:38,910 --> 00:01:42,900
finally, we decide, choosing the option that 
seems best,

32
00:01:42,910 --> 00:01:44,400
and acting accordingly.

33
00:01:44,410 --> 00:01:48,200
Just as humans become experts by going through 
the process of observation,

34
00:01:48,210 --> 00:01:51,800
evaluation and decision making, cognitive 
systems like Watson

35
00:01:51,810 --> 00:01:55,500
use similar processes to reason about the 
information they read.

36
00:01:55,510 --> 00:01:58,200
Watson can also do this at massive speed 
and skill.

37
00:01:58,210 --> 00:02:00,400
So, how does Watson do it?

38
00:02:00,410 --> 00:02:03,600
Unlike conventional approaches to computing, 
which can only handle

39
00:02:03,700 --> 00:02:07,100
neatly organized structures data, such as 
what is stored in a database,

40
00:02:07,110 --> 00:02:11,400
Watson can understand unstructured data, 
which is 80% of data today:

41
00:02:11,410 --> 00:02:14,980
all of the information that is produced primarily 
by humans for

42
00:02:14,700 --> 00:02:17,400
other humans to consume. This includes everything 
from

43
00:02:17,410 --> 00:02:22,400
literature, articles, reasearch reports, 
to blogs, posts and tweets.

44
00:02:22,410 --> 00:02:25,600
While structured data is governed by well-defined 
fields that contain

45
00:02:25,610 --> 00:02:30,980
well-specified information, Watson relies 
on natural language, which is governed by

46
00:02:30,980 --> 00:02:32,980
rules of grammar, context and culture.

47
00:02:32,980 --> 00:02:37,100
It's implicit, ambiguous, complex, and a 
challenge to process.

48
00:02:37,110 --> 00:02:39,980
While all human languages are difficult to 
parse,

49
00:02:39,980 --> 00:02:42,980
certain idioms can be particularly challenging.

50
00:02:42,980 --> 00:02:46,200
In English, for instance, we can feel blue 
because it's raining cats and dogs.

51
00:02:46,210 --> 00:02:48,900
While, we're filling in a form, someone asks 
us to "fill out".

52
00:02:48,910 --> 00:02:53,200
When it comes to text, Watson doesn't just 
look for key words matches or synonyms,

53
00:02:53,210 --> 00:02:57,500
like a search engine, it actually reads and 
interpret texts like a person.

54
00:02:57,510 --> 00:02:59,900
It does this by breaking down a sentence 
grammatically,

55
00:02:59,910 --> 00:03:03,700
relationally, and structurally, discerning 
meaning from the semantics

56
00:03:03,710 --> 00:03:04,600
of the written material.

57
00:03:04,700 --> 00:03:08,300
Watson understand context. This is very different 
than simple

58
00:03:08,310 --> 00:03:11,800
speech recognition, which is how a computer 
translates human speech

59
00:03:11,810 --> 00:03:16,600
into a set of words. Watson tries to understand 
the real intent of the user's language,

60
00:03:16,610 --> 00:03:20,800
and uses that understanding to possibly extract 
logical responses, and draw

61
00:03:20,810 --> 00:03:24,700
inferences to potential answers through a 
broader ray of linguistic models

62
00:03:24,710 --> 00:03:25,700
and algorithms.

63
00:03:25,710 --> 00:03:28,900


64
00:03:28,910 --> 00:03:32,400
When Watson goes to work in a particular 
field, it learns the language,

65
00:03:32,410 --> 00:03:35,300
the jargon, and the mode-of-thought of that 
domain.

66
00:03:35,310 --> 00:03:39,200
Take the term "cancer" for instance. There 
are many different types of cancer,

67
00:03:39,210 --> 00:03:42,200
and each type has different symptoms and 
treatments.

68
00:03:42,210 --> 00:03:46,200
However, those symptoms can also be associated 
with diseases

69
00:03:46,210 --> 00:03:49,000
other than cancer. Treatments can have side-effects, 
and affect

70
00:03:49,010 --> 00:03:51,400
people differently, depending on many factors.

71
00:03:51,410 --> 00:03:55,100
Watson evaluates standard of care practices, 
and thousands of pages

72
00:03:55,200 --> 00:03:58,400
of literature that capture the best science 
in the field,

73
00:03:58,410 --> 00:04:01,800
and from all of that, Watson identifies the 
therapies that offer the best

74
00:04:01,810 --> 00:04:05,100
choices for the doctor to consider in their 
treatment of the patient.

75
00:04:05,110 --> 00:04:08,400
With the guidance of human experts, Watson 
collects the knowledge

76
00:04:08,410 --> 00:04:12,900
required to have literacy in a particular 
domain, what's called a Corpus of knowledge.

77
00:04:12,910 --> 00:04:15,700
Collecting a Corpus starts with loading the 
relevant body

78
00:04:15,710 --> 00:04:19,300
of literature into Watson. Building the Corpus 
also requires

79
00:04:19,310 --> 00:04:22,960
some human interventation to go through the 
information and

80
00:04:22,960 --> 00:04:26,500
discard anything that is out of date, poorly 
regarded or immaterial to

81
00:04:26,510 --> 00:04:30,040
the problem domain. We refer to this as Curating 
the content.

82
00:04:30,040 --> 00:04:33,800
Next, the data is pre-processed by Watson, 
building indices

83
00:04:33,810 --> 00:04:37,040
and other metadata that make working with 
that content more efficient.

84
00:04:37,040 --> 00:04:41,800
This is known as Ingestion. At this time, 
Watson may also create a

85
00:04:41,810 --> 00:04:45,040
knowledge graph to assist in answering more 
precise questions.

86
00:04:45,040 --> 00:04:47,040


87
00:04:47,040 --> 00:04:50,500
Now that Watson has ingested the corpus, 
it needs to be trained

88
00:04:50,510 --> 00:04:53,700
by a human expert to learn how to interpret 
the information.

89
00:04:53,710 --> 00:04:57,040
To learn the best possible responses and 
acquire the ability to

90
00:04:57,040 --> 00:05:01,500
find patterns, Watson partners with experts, 
who trained in using

91
00:05:01,510 --> 00:05:04,960
an approach called Machine learning. An expert 
will upload training

92
00:05:04,960 --> 00:05:07,960
data into Watson in the form of question/answer 
pairs,

93
00:05:07,960 --> 00:05:11,000
that service ground truth. This doesn't give 
Watson explicit

94
00:05:11,010 --> 00:05:15,300
answers for every question it will receive, 
but rather teaches it the linguistic patterns

95
00:05:15,400 --> 00:05:19,500
of meaning in the domain. Once Watson has 
been trained on QA pairs,

96
00:05:19,510 --> 00:05:22,200
it continues to learn through on-going interaction.

97
00:05:22,210 --> 00:05:25,500
Interactions between users and Watson are 
periodically reviewed by

98
00:05:25,510 --> 00:05:28,900
experts and fed back into the system to help 
Watson better

99
00:05:28,910 --> 00:05:32,500
interpret information. Likewise, as new information 
is published,

100
00:05:32,510 --> 00:05:36,400
Watson is updated so that it's constantly 
adapting to shifts in knowledge

101
00:05:36,410 --> 00:05:39,600
and linguistic interpretation in any given 
field.

102
00:05:39,610 --> 00:05:42,700
Watson is now ready to respond to questions 
about highly

103
00:05:42,710 --> 00:05:46,800
complex situations, and quickly provide a 
range of potential responses

104
00:05:46,810 --> 00:05:49,600
and recommendations that are backed by evidence.

105
00:05:49,610 --> 00:05:53,700
It's also prepared to identify new insights 
or patterns locked away

106
00:05:53,710 --> 00:05:54,500
in information.

107
00:05:54,510 --> 00:05:58,800


108
00:05:58,810 --> 00:06:00,700
From metallurgists looking for new alloys,

109
00:06:00,710 --> 00:06:03,500
to researchers looking to develop more effective 
drugs,

110
00:06:03,510 --> 00:06:07,200
human experts are using Watson to uncover 
new possibilities

111
00:06:07,210 --> 00:06:10,300
in data, and make better evidence-based decisions.

112
00:06:10,310 --> 00:06:14,300
Across all of these different applications, 
there is a common approach

113
00:06:14,310 --> 00:06:17,300
that Watson follows. After identifying parts 
of speech

114
00:06:17,310 --> 00:06:20,200
in a question on inquiry, it generates hypothesis.

115
00:06:20,210 --> 00:06:24,300
Watson then looks for evidence to support 
or refute the hypothesis.

116
00:06:24,310 --> 00:06:28,600
It scores each passage based on statistical 
modeling for each piace of evidence

117
00:06:28,610 --> 00:06:31,000
known as weighted evidence scores.

118
00:06:31,010 --> 00:06:34,100
Watson has to meet its confidence based on 
how high the responses

119
00:06:34,110 --> 00:06:36,900
rated during evidence scoring and ranking.

120
00:06:36,910 --> 00:06:40,400
In essence, Watson is able to run analytics 
against a body of data

121
00:06:40,410 --> 00:06:43,000
to clean insights, which Watson can turn 
into inspirations,

122
00:06:43,010 --> 00:06:46,800
allowing human experts to make better and 
more informed decisions.

123
00:06:46,810 --> 00:06:51,100
Across an organization, Watson scales and 
democratises expertise

124
00:06:51,110 --> 00:06:55,200
by surfacing accurate responses and answers 
to an inquiry or question.

125
00:06:55,210 --> 00:06:59,200
Watson also accelerates expertise by surfacing 
a set of possibilities

126
00:06:59,210 --> 00:07:02,400
from a large body of data, saving valuable 
time.

127
00:07:02,410 --> 00:07:05,600
Today, Watson is revolutionizing the way 
we make decisions,

128
00:07:05,610 --> 00:07:09,300
become experts and share expertise in fields 
as diverse as

129
00:07:09,310 --> 00:07:12,500
law, medicine, and even cooking.

130
00:07:12,510 --> 00:07:16,000
Further, Watson is discovering and offering 
answers and patterns

131
00:07:16,100 --> 00:07:19,400
we hadn't known existed faster than any person 
or group of people

132
00:07:19,410 --> 00:07:23,500
ever could, in ways that make a material 
difference everyday.

133
00:07:23,510 --> 00:07:28,300
Most important of all, Watson learns, adapts, 
and keeps getting smarter.

134
00:07:28,310 --> 00:07:32,500
It actually gains value with age by learning 
from its interactions with us

135
00:07:32,510 --> 00:07:36,000
and from its own successes and failures, 
just like we do.

136
00:07:36,010 --> 00:07:39,500
So, now that you know how it works, how do 
these ideas

137
00:07:39,510 --> 00:07:43,300
inspire how you work? How can Watson make 
you a better expert?

138
00:07:43,310 --> 00:07:44,700
What will you do with Watson?

139
00:07:44,710 --> 00:07:53,000