00:06.0
00:09.1
IBM's Watson is at the forefront of a new era of computing:
00:09.1
00:12.1
Cognitive computing. It's a radically new kind of
00:12.1
00:14.7
computing, very different from the programmable systems
00:14.7
00:17.6
that preceeded it. As different as those systems were from
00:17.6
00:20.0
the Tabulating Machines of a century ago.
00:20.1
00:24.1
Conventional computing solutions, based on mathematical principals
00:24.2
00:26.6
that emanate from the 1940s, are programmed
00:26.6
00:30.7
based on rules and logic, intended do derive mathematically precise answers
00:31.6
00:33.7
often following a rigid decision tree approach.
00:33.8
00:36.5
But, with today's wealth of big data, and the need
00:36.5
00:39.4
for more complex evidence-based decisions, such a
00:39.4
00:43.9
rigid approach often breaks, or fails to keep up with available information.
00:43.9
00:47.2
Cognitive computing enables people to create a profoundly new
00:47.3
00:49.5
kind of value, finding answers in insights
00:49.6
00:51.5
locked away in volumes of data.
00:51.6
00:54.4
Whether we consider a doctor diagnosing a patient,
00:54.4
00:58.0
a wealth manager advising a client on their retirement portfolio,
00:58.1
01:01.2
or even a chef creating a new recipe, they need new approaches
01:01.2
01:05.2
to put into contacts the volume of information they deal with
01:05.2
01:07.9
on a daily basis in order to derive value from it.
01:07.9
01:10.5
This process serves to enhance human expertise.
01:10.6
01:13.9
Watson and its cognitive capabilities mirror some of the key
01:13.9
01:16.5
cognitive elements of human expertise.
01:16.6
01:19.0
Systems that reason about problems like a human does
01:19.1
01:21.5
when we, as humans, seek to understand something,
01:21.5
01:25.0
and to make a decision, we go through four key steps:
01:25.0
01:28.4
first, we observe visible phenomena and bodies of evidence;
01:28.4
01:32.2
second, we draw on what we know to interpret what we're seeing,
01:32.2
01:34.8
to generate hypothesis about what it means;
01:34.8
01:38.9
third, we evaluete which hypothesis are right or wrong;
01:38.9
01:42.9
finally, we decide, choosing the option that seems best,
01:42.9
01:44.4
and acting accordingly.
01:44.4
01:48.2
Just as humans become experts by going through the process of observation,
01:48.2
01:51.8
evaluation and decision making, cognitive systems like Watson
01:51.8
01:55.5
use similar processes to reason about the information they read.
01:55.5
01:58.2
Watson can also do this at massive speed and skill.
01:58.2
02:00.4
So, how does Watson do it?
02:00.4
02:03.6
Unlike conventional approaches to computing, which can only handle
02:03.7
02:07.1
neatly organized structures data, such as what is stored in a database,
02:07.1
02:11.4
Watson can understand unstructured data, which is 80% of data today:
02:11.4
02:14.6
all of the information that is produced primarily by humans for
02:14.7
02:17.4
other humans to consume. This includes everything from
02:17.4
02:22.4
literature, articles, reasearch reports, to blogs, posts and tweets.
02:22.4
02:25.6
While structured data is governed by well-defined fields that contain
02:25.6
02:30.1
well-specified information, Watson relies on natural language, which is governed by
02:30.2
02:32.6
rules of grammar, context and culture.
02:32.7
02:37.1
It's implicit, ambiguous, complex, and a challenge to process.
02:37.1
02:39.1
While all human languages are difficult to parse,
02:39.2
02:42.1
certain idioms can be particularly challenging.
02:42.2
02:46.2
In English, for instance, we can feel blue because it's raining cats and dogs.
02:46.2
02:48.9
While, we're filling in a form, someone asks us to "fill out".
02:48.9
02:53.2
When it comes to text, Watson doesn't just look for key words matches or synonyms,
02:53.2
02:57.5
like a search engine, it actually reads and interpret texts like a person.
02:57.5
02:59.9
It does this by breaking down a sentence grammatically,
02:59.9
03:03.7
relationally, and structurally, discerning meaning from the semantics
03:03.7
03:04.6
of the written material.
03:04.7
03:08.3
Watson understand context. This is very different than simple
03:08.3
03:11.8
speech recognition, which is how a computer translates human speech
03:11.8
03:16.6
into a set of words. Watson tries to understand the real intent of the user's language,
03:16.6
03:20.8
and uses that understanding to possibly extract logical responses, and draw
03:20.8
03:24.7
inferences to potential answers through a broader ray of linguistic models
03:24.7
03:25.7
and algorithms.
03:28.9
03:32.4
When Watson goes to work in a particular field, it learns the language,
03:32.4
03:35.3
the jargon, and the mode-of-thought of that domain.
03:35.3
03:39.2
Take the term "cancer" for instance. There are many different types of cancer,
03:39.2
03:42.2
and each type has different symptoms and treatments.
03:42.2
03:46.2
However, those symptoms can also be associated with diseases
03:46.2
03:49.0
other than cancer. Treatments can have side-effects, and affect
03:49.0
03:51.4
people differently, depending on many factors.
03:51.4
03:55.1
Watson evaluates standard of care practices, and thousands of pages
03:55.2
03:58.4
of literature that capture the best science in the field,
03:58.4
04:01.8
and from all of that, Watson identifies the therapies that offer the best
04:01.8
04:05.1
choices for the doctor to consider in their treatment of the patient.
04:05.1
04:08.4
With the guidance of human experts, Watson collects the knowledge
04:08.4
04:12.9
required to have literacy in a particular domain, what's called a Corpus of knowledge.
04:12.9
04:15.7
Collecting a Corpus starts with loading the relevant body
04:15.7
04:19.3
of literature into Watson. Building the Corpus also requires
04:19.3
04:22.3
some human interventation to go through the information and
04:22.4
04:26.5
discard anything that is out of date, poorly regarded or immaterial to
04:26.5
04:30.1
the problem domain. We refer to this as Curating the content.
04:30.1
04:33.8
Next, the data is pre-processed by Watson, building indices
04:33.8
04:37.6
and other metadata that make working with that content more efficient.
04:37.6
04:41.8
This is known as Ingestion. At this time, Watson may also create a
04:41.8
04:45.1
knowledge graph to assist in answering more precise questions.
04:47.1
04:50.5
Now that Watson has ingested the corpus, it needs to be trained
04:50.5
04:53.7
by a human expert to learn how to interpret the information.
04:53.7
04:57.6
To learn the best possible responses and acquire the ability to
04:57.6
05:01.5
find patterns, Watson partners with experts, who trained in using
05:01.5
05:04.3
an approach called Machine learning. An expert will upload training
05:04.4
05:07.8
data into Watson in the form of question/answer pairs,
05:07.9
05:11.0
that service ground truth. This doesn't give Watson explicit
05:11.0
05:15.3
answers for every question it will receive, but rather teaches it the linguistic patterns
05:15.4
05:19.5
of meaning in the domain. Once Watson has been trained on QA pairs,
05:19.5
05:22.2
it continues to learn through on-going interaction.
05:22.2
05:25.5
Interactions between users and Watson are periodically reviewed by
05:25.5
05:28.9
experts and fed back into the system to help Watson better
05:28.9
05:32.5
interpret information. Likewise, as new information is published,
05:32.5
05:36.4
Watson is updated so that it's constantly adapting to shifts in knowledge
05:36.4
05:39.6
and linguistic interpretation in any given field.
05:39.6
05:42.7
Watson is now ready to respond to questions about highly
05:42.7
05:46.8
complex situations, and quickly provide a range of potential responses
05:46.8
05:49.6
and recommendations that are backed by evidence.
05:49.6
05:53.7
It's also prepared to identify new insights or patterns locked away
05:53.7
05:54.5
in information.
05:58.8
06:00.7
From metallurgists looking for new alloys,
06:00.7
06:03.5
to researchers looking to develop more effective drugs,
06:03.5
06:07.2
human experts are using Watson to uncover new possibilities
06:07.2
06:10.3
in data, and make better evidence-based decisions.
06:10.3
06:14.3
Across all of these different applications, there is a common approach
06:14.3
06:17.3
that Watson follows. After identifying parts of speech
06:17.3
06:20.2
in a question on inquiry, it generates hypothesis.
06:20.2
06:24.3
Watson then looks for evidence to support or refute the hypothesis.
06:24.3
06:28.6
It scores each passage based on statistical modeling for each piace of evidence
06:28.6
06:31.0
known as weighted evidence scores.
06:31.0
06:34.1
Watson has to meet its confidence based on how high the responses
06:34.1
06:36.9
rated during evidence scoring and ranking.
06:36.9
06:40.4
In essence, Watson is able to run analytics against a body of data
06:40.4
06:43.0
to clean insights, which Watson can turn into inspirations,
06:43.0
06:46.8
allowing human experts to make better and more informed decisions.
06:46.8
06:51.1
Across an organization, Watson scales and democratises expertise
06:51.1
06:55.2
by surfacing accurate responses and answers to an inquiry or question.
06:55.2
06:59.2
Watson also accelerates expertise by surfacing a set of possibilities
06:59.2
07:02.4
from a large body of data, saving valuable time.
07:02.4
07:05.6
Today, Watson is revolutionizing the way we make decisions,
07:05.6
07:09.3
become experts and share expertise in fields as diverse as
07:09.3
07:12.5
law, medicine, and even cooking.
07:12.5
07:16.0
Further, Watson is discovering and offering answers and patterns
07:16.1
07:19.4
we hadn't known existed faster than any person or group of people
07:19.4
07:23.5
ever could, in ways that make a material difference everyday.
07:23.5
07:28.3
Most important of all, Watson learns, adapts, and keeps getting smarter.
07:28.3
07:32.5
It actually gains value with age by learning from its interactions with us
07:32.5
07:36.0
and from its own successes and failures, just like we do.
07:36.0
07:39.5
So, now that you know how it works, how do these ideas
07:39.5
07:43.3
inspire how you work? How can Watson make you a better expert?
07:43.3
07:44.7
What will you do with Watson?
