What is IBM Watson, What is Cognitive Computing, What organizations benefit from IBM Watson, and get an exclusive look into IBM Watson in an in-depth demo exploration. For more information about Watson, email Cresco at info@crescointl.com or visit http://www.crescointl.com.
3. 3
Examples of unstructured text data
• Equipment operating manuals
• Maintenance documentation
• Regulatory requirements
• Enterprise policies
• Doctor, nurse, and lab notes
• Etc.
• …….
• …….
4. 4
Observe Interpret DecideEvaluate
Most enterprises today are not effectively leveraging the
data that is not in traditional structured form
Traditionally
Structured
Data (numbers,
or small chunks
of text)
• Collected by
structured
automated methods
• Enters as structured
inputs
• Stored in relational
systems
• The structure
defines the rules
and meaning
• Accessing and
processing are very
fast
• Numerical slicing and
dicing
• Statistical and other
advance techniques
are easy to apply
• Decision rules are
easy to assign
• Predictive analytics
• Decisions are clear
from strong evidence
• Decisions support
business experts
• Analyses are fast
and accurate
5. 5
Observe Interpret DecideEvaluate
Most enterprises today are not effectively leveraging the
data that is not in traditional structured form
Traditionally
Structured
Data (numbers,
or small chunks
of text)
• Collected by
structured
automated methods
• Enters as structured
inputs
• Stored in relational
systems
• The structure
defines the rules
and meaning
• Accessing and
processing are very
fast
• Numerical slicing and
dicing
• Statistical and other
advance techniques
are easy to apply
• Decision rules are
easy to assign
• Predictive analytics
• Decisions are clear
from strong evidence
• Decisions support
business experts
• Analyses are fast
and accurate
Data with Other
Structures
(blocks of text,
images, video,
audio, sensory,
etc.)
• Collected in large
batches with
many different
formats
• Enters systems
with little structure
• Stored in massive
file repositories or
data lakes
• Machines might
derive very general
descriptions, but to
get to deeper
meaning, humans
are required
• Accessing and
processing the data
are challenges and
require expert
programmers
• Often nothing is
done
• Unless humans first
do the difficult task of
structuring the data,
machines can not do
much with it, so this is
usually done by
humans
• Analysis techniques
are not familiar and
requires expert
analysts
• Often, nothing is
done
• Decisions are often
not clear as the
supporting evidence
is often not well
defined
• Support is needed
for non-experts, but
is not human-friendly
• Interpretation and
evaluation can be
very slow and
inaccurate
• Often nothing is
done
6. 6
New techniques bring the ability to analyze ambiguously-structured data, and to
help enterprises to develop, leverage, and transfer innovative expertise.
Final
Score:$ 24,000 $ 21,600$ 77,147
IBM enters a Q&A computer called
Watson in the Jeopardy! exhibition, and
it successfully beats the best human
contestants
Extensive research has developed better technology
Grand Challenge:
Automatic Open-
Domain Question
Answering
~2008 2011
IBM Research
generates many
additional potential
offerings based on
new technologies
The Watson Brand
group is established
(SaaS solutions and
PaaS API)
2012-2013 2014
IBM Research tackles
a long-standing
Artificial Intelligence
challenge
Watson Brand
Additional
potential
offerings based
on new
technologies
Watson
Offerings
7. 7
New capabilities have started a different phase in the
history of computing: Cognitive Computing
Data with
Other
Structures
(blocks of text,
images, video,
audio,
sensory, etc.)
• Understands very
sophisticated contexts
• Finds new insights that
were not possible from
only structured data
• Can make sense of
massive volumes of data
• Automatically interprets
and evaluates quickly
and accurately
• Provides for evidence-
based decisions
• Supports non-experts
• Can be tuned by subject
matter experts instead of
programmers
• Adds rich context and derives deep insights, with new
capabilities (some examples below):
• Identify Features, Cluster
• Add Semantic and Advanced Context, Interpret, Convert
• Create Structure (index), Classify, Categorize
• Summarize
• Enable Federated Access, Find, Filter, Rank with Evidence
• Match Complex Criteria, Fit, Analyze Trade-offs
• Correlate, Show Relationships, Expand
• Orchestrate Dialog
• Create New Combinations
• Contribute to Predictive Analysis and Next Best Action
• Simplifies the processing of mass amounts of data
• Leverages machine learning to reduce the need for
programming
• Makes access, processing and interaction human-friendly
Observe Interpret DecideEvaluate
8. 8
The Association of Information and Image Management recognizes
the opportunity, along with many other thought leaders
9. 9
Cognitive Computing - Clarifications
• It is inspired by human cognition, and not attempting to replicate it
– Brains: bio-chemical and largely analog
– Computers: other materials, ones and zeros, and usually Von Neumann designs
– The objective is to automate tasks that previously required humans
• There is no specific technical definition
– the key themes are “unstructured data” and “relating in a human-like way”
• Understanding natural language is only part of it
– the goal is to reach much richer and deeper contexts
• It is not just about deep Artificial Neural Networks (ANN)
– Leverages the best choice between ANN, statistical algorithms, rules techniques, heuristic
approaches, etc.
– Often, a combination of techniques is used
• Machine learning is just one aspect
– Tuning algorithms without programming
– For most situations, supervised machine learning is the best fit
• It is more about Recognition than about Prediction (Forecasting)
10. 10
Enhances
Watson enhances the cognitive
process of professionals to
strengthen decision making in the
moment
Observe
InterpretDecide
Evaluate
Observe
InterpretDecide
Evaluate
Watson: a brand cover many cognitive solutions that
can offer tremendous benefits
Watson scales expertise by
elevating the consistency and
objectivity of decision making
across an organization.
Scales
Accelerates
Watson captures the expertise of
top performers and accelerates the
development of that expertise in
others.
Master
Practice
Apprentice
Study
Traditional
Learning
Curve
Learning
Curve with
Watson
11. 11
• 198082, Jan 1, 2000, 7:00:00 AM, WHILE TRAVELING DOWN THE HIGHWAY AT APPOX. 65 MPH I BEGAN TO APPLY THE BRAKES, THE
ENTIRE VEHICLE AND STEERING WHEEL SHOOK VIOLENTLY. THIS CAME WITHOUT WARNING, AFTER A FEW GENTLE PUSHES ON THE
BRAKE IN THE NEXT FEW MILES THE SHUDDER WAS LESS VIOLENT. IT NOW FEELS AS IF THE ROTORS ARE WARPED, I HAD THE
ROTORS TURNED AT 6500 MILES AND NOW IT SEEMS AS THEY ARE WARPED AGAIN. IN RECENT DAYS I HAVE BEEN CAREFUL TO NOTE
THE VEHICLES BEHAVIOR, IT SEEMS THAT AFTER 5-6 STOPLIGHTS OR BRAKING PERIODS IN QUICK SUCCESSION THAT THE NEXT FEW
STOPS ARE ESPECIALLY VIOLENT, AND THEN A FEW STOPS LATER IT GOES AWAY. THE VIOLENT SHAKING IS STRONG ENOUGH TO
CAUSE PROBLEMS AT HIGHWAY SPEEDS, I AM NOW CONFINED TO <45 MPH UNTIL THE VEHICLE IS INSPECTED NEXT WEEK
• 198283, Jan 3, 2000, 7:00:00 AM, GOING FROM REVERSE TO DRIVE, CAR ACCELERATED AND SMASHED INTO A MOBILE HOME.(DETAILS?
RDRUNER31@AOL.COM). I WAS ATTEMPTING TO BACK OUT OF SOMEONE'S DIRT DRIVEWAY. I WAS POSITIONED AT AN ANGLE. WHEN I
BACKED OUT (GOING ABOUT 3-5 MPH), I BUMPED INTO A TREE. I THEN PUT MY FOOT ON THE BRAKES SO THAT I COULD PUT THE CAR IN
DRIVE. WHEN I PUT THE CAR IN DRIVE, THE CAR ACCELERATED AT AN EXTREME RATE AND MADE A TERRIBLY LOUD NOISE. THE CAR
THEN THRUSTED FORWARD (THE ENTIRE TIME MY FOOT WAS HEAVY ON THE BRAKE - I AM 100% SURE OF THIS! ). THE BRAKE DID NOT
STOP THE CAR. AFTER TRAVELING ABOUT 60 FEET, THE CAR THEN CAREAMED INTO A MOBILE HOME. I TURNED THE IGNITION OFF. I
BELIEVE THOUGH THAT THE CAR WAS STOPPED BY THE METAL BEAMS FROM THE MOBILE HOME. THE CAR IS NOW AT THE
DEALERSHIP
• 198488, Jan 4, 2000, 7:00:00 AM, MY WIFE REGINA WAS DRIVING THE CAR SHE HIT A CAR BROADSIDE THAT HAD RUN A RED LIGHT. SHE
WAS GOING ABOUT 20 MILES PER HOUR, AND THE OTHER CAR WAS GOING ABOUT 25 MILES PER HOUR. THE FAILURE WAS THAT THE
AIR BAGS DID NOT DEPLOY. OFFICER PETERSON FROM THE MEADVILLE CITY POLICE FILED THE ACCIDENT REPORT. THE ACCIDENT
HAPPENED AT THE INTERSECTION OF PARK AVENUE AND NORTH STREET IN THE CITY OF MEADVILLE, PA 16335.
• 198518, Jan 4, 2000, 7:00:00 AM, WHILE DRIVING CONSUMER STEPPED ON THE BRAKE PEDAL TO STOP VEHICLE, BUT BRAKES DID NOT
RESPOND. CONSUMER TRIED TO AVOID REAR ENDING ANOTHER VEHICLE BY DRIVING VEHICLE OFF THE ROAD. BUT WAS INVOLVED IN
A ROLLOVER. UPON IMPACT, AIR BAGS DID NOT DEPLOY.
• 198612, Jan 5, 2000, 7:00:00 AM, ON THE 21ST OF DEC 99, MY WIFE WAS INVOLVED IN AN ACCIDENT IN OUR WINDSTAR. THIS ACCIDENT
COULD HAVE BEEN AVOIDED IF THE HORN WERE USEABLE. THE WAY THE HORN BUTTON IS NOW YOU WILL HAVE A DIFFICULT TIME
TRYING TO FIND THE EXACT SPOT TO PUSH TO GET THE HORN TO SOUND. WHEN TIME IS CRITICAL IN THE OUTCOME, SEARCHING FOR
THAT EXACT SPOT ISN'T A PLAYER. THIS WAS THE CASE ON THE 21ST. WHEN AN INDIVIDUAL TRIED TO CROSS THE HIGHWAY HE HIT
OUR VAN ON THE RIGHT SIDE CAUSING $6500 IN DAMAGE. PRIOR TO THE IMPACT SHE TRIED TO HIT THE HORN BUT COULD NOT FIND IT
WHEN IT WAS NEEDED MOST. WE ARE ALMOST 100% POSITIVE THAT IF SHE COULD HAVE FOUND THE HORN SHE COULD HAVE
SOUNDED IT, AND THE OTHER DRIVER WOULD HAVE SEEN HER COMING. LUCKILY MY 18 MONTH OLD AND 5 YR OLD DAUGHTERS WERE
NOT WITH HER AT THE TIME.
Examples of unstructured records in the NHTSA database – Watson
can make sense out of millions of these!
15. 15
Watson can identify objects and people in visual data, and derive
speech from audio data
16. 16
Personality portraits cover a large number of aspects
Watson can recognize personality characteristics, emotions,
and tone
Watson helps people to
detect communication styles:
• Social
• Emotional
• Writing
Tone Analyzer recognizes a spectrum of
emotional tones:
21. 21
Bluemix Services (REST API’s as PaaS)
• Natural Language Classifier
• Dialog services
• Retrieve and Rank
• Alchemy (Language, Vision, News)
• Document Conversion
• Personality Insights
• Tone Analyzer
• Relationship Extraction
• Concept Expansion
Watson Solutions and Watson Developer Cloud Services (via Bluemix)
Watson Solutions
On-Premise
• Watson Explorer – Advanced Edition
SaaS
• Watson Analytics
• Watson Engagement Advisor
• Watson Discovery Advisor for Life Sciences
• Watson Discovery Advisor (coming soon)
• Watson for Wealth Management
• Watson Company Advisor
• Watson Oncology Advisor
• Watson for Clinical Trials Management
• Chef Watson
Can be combined to
create compound
solutions
• Concept Insights
• Cognitive Commerce
• Cognitive Graph
• Speech to Text
• Text to Speech
• Language Identification
• Language Translation
• Tradeoff Analytics
• Visualization Rendering
• Visual Recognition
These are typical
starting points
22. 22
Watson’s points of differentiation
Creates knowledge graph, indexing, faceting, metadata, etc.
Specialized analytics or processing
Sophisticated human-friendly inbound and outbound interaction
Offers easy and fast tooling
Builds for enterprise size,
strength, and security
Facilitates extension
Deploys as SaaS
Extraordinarily
Deep Context
Enables integration
Accessible via many types of
user devices
Inbound and Outbound Interaction
Analytics and Processing
Contextualizing
Knowledge Graph, Indexing, Faceting, Metadata, etc
Data Sources
Context
Platform
Core Capabilities
Connects to and crawls the data sources intelligently
Curated Data
Parses, evaluates, and adds context
23. 23
IBM offerings that easily integrate with Watson
Watson also easily integrates with
many solutions:
• Intelligence (i2)
• Advanced Care Insights
(Smarter Healthcare)
• Epic (healthcare EMR)
• Curam (healthcare case mgmt)
• Emptoris (purchasing)
• Genesys (contact center)
…and more solutions
…and more products
24. 24
Key challenges for Cognitive Computing
• Understanding what the technology can and can not do, and how to apply it
• Defining the high value use cases
• Accounting for the benefits and ROI
• Changing the enterprise for adoption of the technology and solutions
• Finding data sources with superior value
• Accessing and converting data
• Training routines
• Addressing perceived risks
• Data privacy risks
• Cloud risks
• Artificial Intelligence risks
25. 25
Because this domain’s technologies and techniques are changing so rapidly, the
maturing of offerings is not always smooth
Watson’s offering development routine
Long-term
roadmap – v1
Beta
Offering
Development
Research
Long-term
roadmap – v2
Happy Path
Withdrawn at
Beta
Withdrawn at
early version
Extraordinary
new version
27. 27
Watson Developer Cloud is a platform that
provides developers easy access to expertise
via a collection of REST APIs & SDKs
WDC services are accessed via Bluemix, an open-
standards, cloud-based platform for building, running, and
managing applications
https://www.ibm.com/smarterplanet/us/en/ibmwatson/
developercloud/services-catalog.html
28. 28
Example: Agent Assisted Insurance claims CRM
Q&A
Direct responses to user
inquiries fueled by primary
document sources
Relationship
Extraction
Intelligently finds
relationships
between sentence
components Concept Insights
Explores information based
on the ideas, rather than
traditional text matching
Personality
Insights
Deeper understanding of
people's personality
characteristics, and
values
Watson Explorer
Build a 360 view of all
your information
AlchemyVision
Imagine recognition
29. 29
Watson APIs
• Natural Language Classifier – determines the essential “intent” of questions or statements, according to
classifications for which it can be trained
• Dialog Service – orchestrates a natural language dialog interaction
• Retrieve and Rank – performs an indexed search, and has a trainable ranking function to determine the best
evidence-based responses
• Concept Insights – automatically tags content in relation to a concept graph that is based on content ingested
from the English language Wikipedia (can ingest test and/or a collection)
• Language Translation – provides translation for a number of languages, and language identification for a large
number of languages
• Speech to Text – converts speech into text
• Text to Speech – synthesizes speech audio from text with either male or female voices
• Document Conversion - converts a single HTML, PDF, or Microsoft Word™ document into a normalized formats
(e.g. HTML, plain text, or JSON)
• Personality Insights – recognizes 52 personality characteristics from human text compositions
• Tone Analyzer (Beta) – classifies text as to emotional state (e.g. anger, fear, joy, sadness, and disgust. )
• Relation Extraction (Beta) - identifies Subject-Action-Object relations within text according to predefined rules
• Concept Expansion (Withdrawn) – identifies contextually related words: The Big Apple refers to NY City
30. 30
Watson APIs (cont.)
AlchemyLanguage (various pre-trained text analytics functions)
• Entity Extraction –extracts entities like people, locations and organizations for 23 languages
• Sentiment Analysis - analyzes words and phrases to categorize as to sentiment
• Keyword Extraction – analyzes text data to extract keywords that can be used to index content, generate tag clouds, and
more
• Concept Tagging – analyzes text to tagging according to desired class or type ("My favorite brands are BMW, Ferrari, and
Porsche." = "Automotive Industry")
• Taxonomy Classification - analyzes text to classify by topic (baseball, mobile phones, etc.)
• Author Extraction - If a news article or blog post specifies an author, AlchemyAPI will attempt to extract it automatically
• Language Detection - identifies more languages (95+) than any other text analysis service, at extremely high rates of
accuracy
• Text Extraction - extracts only important text and title information from any web page
• Feed Detection - automatically discover syndicated content feeds associated with specific web sites or individual web
pages
• Relationship Extraction - enables you to extract useful information from input text, such as entities and the relationships
that exist among them
31. 31
Vision
• AlchemyVision – analyzes images with a pre-trained classifier to create metadata about the features found within
(focuses on people, faces, gender, age, celebrity ID, and text)
• Visual Insights – analyzes an images, or collections of images, with a pre-trained classifier to create metadata about the
features found within (focuses on general activities, places, interests and people)
• Visual Recognition (Beta) – analyzes images to classify features, with a sophisticated trainable classifier
Data Insights
• Tradeoff Analysis – enables decisions for situations with multiple variables or requirements by allowing the selection of
specific weights to be applied to the different variables or requirements
• AlchemyData News – provides searching for news articles according to key topic for 60 days of history across 75,000
unique news sources (250,000 new articles each day) that have been analyzed via a pre-trained news-oriented classifier
Watson APIs (cont.)