© International Business Machines 2015
IBM Watson
© International Business Machines 2015
Michael Karasick, PhD
Vice President, IBM Watson Innovations
msk@us.ibm.com
IBM Watson
© International Business Machines 2015
IBM Watson
Agenda
What Watson does
Watson Application Patterns
How Watson Works
Watson for Software Creators
2
© International Business Machines 2015
IBM Watson
Agenda
What Watson does
Watson Application Patterns
How Watson Works
Watson for Software Creators
3
© International Business Machines 2015
IBM Watson
You are here
New Techniques are Necessary for Data Overload
Amount of data world wide by 2020*:
– 44 Zettabytes; or
– 44 x 1021 bytes; or
– 44,000,000,000,000,000,000,000 bytes
Internet
of Things
Images &
Multimedia
Text
Enterprise
Data
*EMC Digital Universe with
Research & Analysis by IDC (April
2014)
4
© International Business Machines 2015
IBM Watson
New Computing Paradigms Address Challenges of the Day
5
1950: PROGRAMMING
• Stored data, instructions
• Languages for computing
• Metrics for computation
2011: COGNITION
• Massive data scale
• Data for training
• Real-world modalities
1900: TABULATION
• Punched card tabulation
• Scale, automation
• Seeds of future innovation
© International Business Machines 2015
IBM Watson
6
cog·ni·tion
/,käɡˈniSH(ə)n/
noun
the mental action or process of acquiring knowledge and understanding through
thought, experience, and the senses.
synonyms: perception, discernment, apprehension, learning, understanding,
comprehension, insight; reasoning, thinking, thought
• a result of this; a perception, sensation, notion, or intuition.
plural noun: cognitions
© International Business Machines 2015
IBM Watson
7
Insert Watson Video #1 Slide Here
© International Business Machines 2015
IBM Watson
Watson Finds Knowledge in Noisy Data at Enormous Scale
8
“Listens to signals” for a domain.
Obtains patterns (meaning) from the signals:
– detects and scores the strength of the signal features;
– learns which features are meaningful.
Determines confidence by analyzing supporting evidence.
Uses trained machine learning:
– previously trained against data that represents domain ground truth.
Elevates and amplifies human cognition:
– Exploration, engagement, discovery, enforcement, and decision support
© International Business Machines 2015
IBM Watson
Obtaining Insight into a Public Company from Regulatory Filings
9
Event
Company Person
SecurityLoan
Annual Report Loan Agreement
Proxy Statement Insider Transaction
Counterparty Relationships
Loan Exposure
SEC/FDIC Filings of
Financial Companies
(Forms 10-K,8-k, 10-Q, DEF 14A,
3/4/5, 13F, SC 13D SC 13 G
FDIC Call Reports)
Scattered Integrated
Discovery
%Owner Officer
Employee
Director Insider
© International Business Machines 2015
IBM Watson
Obtaining Insight into an Individual by Analyzing their Writing
Watson Personality Insights
(Signal) (Meaning)
10
© International Business Machines 2015
IBM Watson
Differences with Programmatic Computing
Learned behavior:
– rather than being instructed via program or rules; and so
– less brittle to change than programmed systems.
Behavior adapts over time according to:
– on-going experience; and
– exposure to new information.
… These systems need LOTS of training data to achieve
human proficiency
– initially and through on-going experience.
11
© International Business Machines 2015
IBM Watson
Agenda
What Watson does
Watson Application Patterns
How Watson Works
Watson for Software Creators
12
© International Business Machines 2015
IBM Watson
Discovery: Accelerate Research and Insights
13
 Law: Precedents
 Public Safety: Anomalous relationships
 Cross Industry: Drug or material discovery
Test hypotheses
Find evidence
Discover new facts
© International Business Machines 2015
IBM Watson
14
Policy Enforcement: Validate Adherence
 Insurance: Paying a claim
 Healthcare: Qualifying a patient for a clinical trial
 Healthcare: Validating acceptability of a medication
Train on policy documents
Test situation against policy
Determine if more data needed
© International Business Machines 2015
IBM Watson
15
Decision Support
Consult an expert
Help with differential diagnosis
Evaluate actions
 Healthcare: Sharing treatment expertise
 Cross Industry: Engineering or mechanical repairs
© International Business Machines 2015
IBM Watson
16
Engagement: Transform Customer Experience
For consumers of a brand
Ask questions
Receive explanations
• Cross Industry: Process concierge
• Cross Industry: Self-help and agent-assist for call centers
© International Business Machines 2015
IBM Watson
17
Insert Watson Video #2 Slide Here
© International Business Machines 2015
IBM Watson
Agenda
What Watson does
Watson Application Patterns
How Watson Works
Watson for Software Creators
18
© International Business Machines 2015
IBM Watson
19
What is the Difference Between a Search Engine and Watson
Meaning and Natural Interaction
© International Business Machines 2015
IBM Watson
20
A 58-year-old woman presented to her primary care physician after several days of dizziness, anorexia, dry mouth, increased thirst, and
frequent urination. She had also had a fever and reported that food would “get stuck” when she was swallowing. She reported no pain
in her abdomen, back, or flank and no cough, shortness of breath, diarrhea, or dysuria. Her family history included oral and bladder
cancer in her mother, Graves' disease in two sisters, hemochromatosis in one sister, and idiopathic thrombocytopenic purpura in one
sister. Her history was notable for cutaneous lupus, hyperlipidemia, osteoporosis, frequent urinary tract infections, three
uncomplicated cesarean sections, a left oophorectomy for a benign cyst, and primary hypothyroidism, which had been diagnosed a
year earlier. Her medications were levothyroxine, hydroxychloroquine, pravastatin, and alendronate. A urine dipstick was positive for
leukocyte esterase and nitrites. The patient was given a prescription for ciprofloxacin for a urinary tract infection and was advised to
drink plenty of fluids. On a follow-up visit with her physician 3 days later, her fever had resolved, but she reported continued weakness
and dizziness despite drinking a lot of fluids. She felt better when lying down. Her supine blood pressure was 120/80 mm Hg, and her
pulse was 88 beats per minute; on standing, her systolic blood pressure was 84 mm Hg, and her pulse was 92 beats per minute. A urine
specimen obtained at her initial presentation had been cultured and grew more than 100,000 colonies of Escherichia coli, which is
sensitive to ciprofloxacin.
Questions are Nuanced and Domain Specific
Domain terminology is nuanced and variable
– Machine Learning critical for training
Evidence Profile for UTI Diagnosis
Disease named as
a symptom
© International Business Machines 2015
IBM Watson
21
…
Watson Long Tail of Questions
FrequencyQuestionisAsked
Question
Factual Questions (Jeopardy)
People: Wikipedia
Watson: Fact Pipeline
Yes/No pipeline
Equivalent Questions
People: Association
Watson: Topical Answers
Frequent Questions
People: Memorize
Watson: Predefined Answers
Explanatory Questions
People: Research
Watson: Passage Rating
…
Disambiguating
Questions
Context
Fusion
© International Business Machines 2015
IBM Watson
Agenda
What Watson is
Watson Application Patterns
How Watson Works
Watson for Software Creators
22
© International Business Machines 2015
IBM Watson
IBM Watson
23
© International Business Machines 2015
IBM Watson
IBM Watson
CogniToy
Elemental
Path
24
© International Business Machines 2015
IBM Watson
25
• Phone
• Text
• Chat
Natural Language
Classifier
Dialog
Common “Recipes” Reflect Usage
How do I reset my password?
– If speech, convert to Text
– Context = “Online Banking”
Watson identifies intent
– Intent = “Password Reset”
– Confidence – 0.876655900
Watson dialog codifies implementation
– Intent=“Password Reset”
– Context = “Online Banking”
– Eventually invoke database system
© International Business Machines 2015
IBM Watson
26
Text to Speech
Concept Expansion
Personality Insights
Tone Analyzer
Language Identification
Machine Translation
Entity Extraction
Sentiment Analysis
Message Resonance
Question and Answer
Relationshiop Extraction
Visualization Rendering
Concept Insights
Data News
Speech to Text
Watson Services are Cloud-Delivered
Bluemix
Tradeoff Analytics
Visual Recognition
Language Detection
Text Extraction
Microformat Parsing
Feed Detection
Keyword Extraction
Linked Data Support
Image Link Extraction
Image Tagging
Face Recognition
Classification
Author Extraction
Taxonomy
© International Business Machines 2015
IBM Watson
Personality Insights
Enables deeper understanding of people's personality
characteristics, needs, and values to help engage users
on their own terms
The IBM Watson Personality Insights service uses linguistic analytics to infer cognitive
and social characteristics, including Big Five, Values, and Needs, from communications
that the user makes available, such as email, text messages, tweets, forum posts, and
more. By deriving cognitive and social preferences, the service helps users to
understand, connect to, and communicate with other people on a more personalized
level.
“My words” show my personality.
“Big 5” Traits:
– openness;
– conscientiousness;
– extraversion;
Calibrated with standard tests.
– agreeableness;
– neuroticism; and
– 47 others.
www.ibm.com/WatsonDeveloperCloud
27
© International Business Machines 2015
IBM Watson
Obtaining Insight into an Individual by Analyzing their Writing
Watson Personality Insights
(Signal) (Meaning)
28
© International Business Machines 2015
IBM Watson
29
© International Business Machines 2015
IBM Watson
Concept Insights
Explores information based on the concepts behind your
input, rather than limiting investigation to findings based
on traditional text matching
The Concept Insights service maps user-input words to the underlying concepts of those
words based on training on English Wikipedia data. Doing so can broaden the user's
investigation beyond the actual words used in an inquiry. Two types of associations are
identified: explicit links when an input document directly mentions a concept, and implicit
links which connect the input documents to relevant concepts that are not directly mentioned
in them. Users of this service can also search for documents that are relevant to a concept
or collection of concepts by exploring the explicit and implicit links.
www.ibm.com/WatsonDeveloperCloud
Link words between documents and concepts
Explicit linkage when a document names a concept
Implicit linkage when concept indirectly mentioned
Trained on English Wikipedia
30
© International Business Machines 2015
IBM Watson
31
Natural Language
Classifier
Interpret natural language and classify it with confidence
The IBM Watson Natural Language Classifier service enables developers without a background
in machine learning or statistical algorithms to create machine-learning, natural language
interfaces for their applications. The service interprets input text (questions or other) and returns
a corresponding classification with associated confidence levels. The return value can then be
used to trigger a corresponding action, such as redirecting the request or answering the
question.
BETA
www.ibm.com/WatsonDeveloperCloud
• Classifies text against predefined categories
– Classification (with confidence)
– Then answer question or trigger an action
• Easy to train
– List of [text, class] pairs
© International Business Machines 2015
IBM Watson
32
© International Business Machines 2015
IBM Watson
© 2015 International Business Machines Corporation
IBM Watson
2015
Australian
Open
#
# EXTRACT THE ANSWERS FROM A WATSON QUERY
#
watsonQuery = JSON.parse(...)["question"]
answers = watsonQuery["evidencelist"]
answer = answers[@answerIndex]
@answerConfidence = answer[“value”]
@answerText = answer[“text”]
@answerValue = answer[“value”]
33
© International Business Machines 2015
IBM Watson
34
“ a”
By 2018 half of all consumers will regularly
interact with services based on cognitive
- IDC FutureScape
© International Business Machines 2015
IBM Watson
www.ibm.com/
watson
Watsonjobs
developer.ibm.com/watson
msk@us.ibm.com35

IBM Watson

  • 1.
    © International BusinessMachines 2015 IBM Watson © International Business Machines 2015 Michael Karasick, PhD Vice President, IBM Watson Innovations msk@us.ibm.com IBM Watson
  • 2.
    © International BusinessMachines 2015 IBM Watson Agenda What Watson does Watson Application Patterns How Watson Works Watson for Software Creators 2
  • 3.
    © International BusinessMachines 2015 IBM Watson Agenda What Watson does Watson Application Patterns How Watson Works Watson for Software Creators 3
  • 4.
    © International BusinessMachines 2015 IBM Watson You are here New Techniques are Necessary for Data Overload Amount of data world wide by 2020*: – 44 Zettabytes; or – 44 x 1021 bytes; or – 44,000,000,000,000,000,000,000 bytes Internet of Things Images & Multimedia Text Enterprise Data *EMC Digital Universe with Research & Analysis by IDC (April 2014) 4
  • 5.
    © International BusinessMachines 2015 IBM Watson New Computing Paradigms Address Challenges of the Day 5 1950: PROGRAMMING • Stored data, instructions • Languages for computing • Metrics for computation 2011: COGNITION • Massive data scale • Data for training • Real-world modalities 1900: TABULATION • Punched card tabulation • Scale, automation • Seeds of future innovation
  • 6.
    © International BusinessMachines 2015 IBM Watson 6 cog·ni·tion /,käɡˈniSH(ə)n/ noun the mental action or process of acquiring knowledge and understanding through thought, experience, and the senses. synonyms: perception, discernment, apprehension, learning, understanding, comprehension, insight; reasoning, thinking, thought • a result of this; a perception, sensation, notion, or intuition. plural noun: cognitions
  • 7.
    © International BusinessMachines 2015 IBM Watson 7 Insert Watson Video #1 Slide Here
  • 8.
    © International BusinessMachines 2015 IBM Watson Watson Finds Knowledge in Noisy Data at Enormous Scale 8 “Listens to signals” for a domain. Obtains patterns (meaning) from the signals: – detects and scores the strength of the signal features; – learns which features are meaningful. Determines confidence by analyzing supporting evidence. Uses trained machine learning: – previously trained against data that represents domain ground truth. Elevates and amplifies human cognition: – Exploration, engagement, discovery, enforcement, and decision support
  • 9.
    © International BusinessMachines 2015 IBM Watson Obtaining Insight into a Public Company from Regulatory Filings 9 Event Company Person SecurityLoan Annual Report Loan Agreement Proxy Statement Insider Transaction Counterparty Relationships Loan Exposure SEC/FDIC Filings of Financial Companies (Forms 10-K,8-k, 10-Q, DEF 14A, 3/4/5, 13F, SC 13D SC 13 G FDIC Call Reports) Scattered Integrated Discovery %Owner Officer Employee Director Insider
  • 10.
    © International BusinessMachines 2015 IBM Watson Obtaining Insight into an Individual by Analyzing their Writing Watson Personality Insights (Signal) (Meaning) 10
  • 11.
    © International BusinessMachines 2015 IBM Watson Differences with Programmatic Computing Learned behavior: – rather than being instructed via program or rules; and so – less brittle to change than programmed systems. Behavior adapts over time according to: – on-going experience; and – exposure to new information. … These systems need LOTS of training data to achieve human proficiency – initially and through on-going experience. 11
  • 12.
    © International BusinessMachines 2015 IBM Watson Agenda What Watson does Watson Application Patterns How Watson Works Watson for Software Creators 12
  • 13.
    © International BusinessMachines 2015 IBM Watson Discovery: Accelerate Research and Insights 13  Law: Precedents  Public Safety: Anomalous relationships  Cross Industry: Drug or material discovery Test hypotheses Find evidence Discover new facts
  • 14.
    © International BusinessMachines 2015 IBM Watson 14 Policy Enforcement: Validate Adherence  Insurance: Paying a claim  Healthcare: Qualifying a patient for a clinical trial  Healthcare: Validating acceptability of a medication Train on policy documents Test situation against policy Determine if more data needed
  • 15.
    © International BusinessMachines 2015 IBM Watson 15 Decision Support Consult an expert Help with differential diagnosis Evaluate actions  Healthcare: Sharing treatment expertise  Cross Industry: Engineering or mechanical repairs
  • 16.
    © International BusinessMachines 2015 IBM Watson 16 Engagement: Transform Customer Experience For consumers of a brand Ask questions Receive explanations • Cross Industry: Process concierge • Cross Industry: Self-help and agent-assist for call centers
  • 17.
    © International BusinessMachines 2015 IBM Watson 17 Insert Watson Video #2 Slide Here
  • 18.
    © International BusinessMachines 2015 IBM Watson Agenda What Watson does Watson Application Patterns How Watson Works Watson for Software Creators 18
  • 19.
    © International BusinessMachines 2015 IBM Watson 19 What is the Difference Between a Search Engine and Watson Meaning and Natural Interaction
  • 20.
    © International BusinessMachines 2015 IBM Watson 20 A 58-year-old woman presented to her primary care physician after several days of dizziness, anorexia, dry mouth, increased thirst, and frequent urination. She had also had a fever and reported that food would “get stuck” when she was swallowing. She reported no pain in her abdomen, back, or flank and no cough, shortness of breath, diarrhea, or dysuria. Her family history included oral and bladder cancer in her mother, Graves' disease in two sisters, hemochromatosis in one sister, and idiopathic thrombocytopenic purpura in one sister. Her history was notable for cutaneous lupus, hyperlipidemia, osteoporosis, frequent urinary tract infections, three uncomplicated cesarean sections, a left oophorectomy for a benign cyst, and primary hypothyroidism, which had been diagnosed a year earlier. Her medications were levothyroxine, hydroxychloroquine, pravastatin, and alendronate. A urine dipstick was positive for leukocyte esterase and nitrites. The patient was given a prescription for ciprofloxacin for a urinary tract infection and was advised to drink plenty of fluids. On a follow-up visit with her physician 3 days later, her fever had resolved, but she reported continued weakness and dizziness despite drinking a lot of fluids. She felt better when lying down. Her supine blood pressure was 120/80 mm Hg, and her pulse was 88 beats per minute; on standing, her systolic blood pressure was 84 mm Hg, and her pulse was 92 beats per minute. A urine specimen obtained at her initial presentation had been cultured and grew more than 100,000 colonies of Escherichia coli, which is sensitive to ciprofloxacin. Questions are Nuanced and Domain Specific Domain terminology is nuanced and variable – Machine Learning critical for training Evidence Profile for UTI Diagnosis Disease named as a symptom
  • 21.
    © International BusinessMachines 2015 IBM Watson 21 … Watson Long Tail of Questions FrequencyQuestionisAsked Question Factual Questions (Jeopardy) People: Wikipedia Watson: Fact Pipeline Yes/No pipeline Equivalent Questions People: Association Watson: Topical Answers Frequent Questions People: Memorize Watson: Predefined Answers Explanatory Questions People: Research Watson: Passage Rating … Disambiguating Questions Context Fusion
  • 22.
    © International BusinessMachines 2015 IBM Watson Agenda What Watson is Watson Application Patterns How Watson Works Watson for Software Creators 22
  • 23.
    © International BusinessMachines 2015 IBM Watson IBM Watson 23
  • 24.
    © International BusinessMachines 2015 IBM Watson IBM Watson CogniToy Elemental Path 24
  • 25.
    © International BusinessMachines 2015 IBM Watson 25 • Phone • Text • Chat Natural Language Classifier Dialog Common “Recipes” Reflect Usage How do I reset my password? – If speech, convert to Text – Context = “Online Banking” Watson identifies intent – Intent = “Password Reset” – Confidence – 0.876655900 Watson dialog codifies implementation – Intent=“Password Reset” – Context = “Online Banking” – Eventually invoke database system
  • 26.
    © International BusinessMachines 2015 IBM Watson 26 Text to Speech Concept Expansion Personality Insights Tone Analyzer Language Identification Machine Translation Entity Extraction Sentiment Analysis Message Resonance Question and Answer Relationshiop Extraction Visualization Rendering Concept Insights Data News Speech to Text Watson Services are Cloud-Delivered Bluemix Tradeoff Analytics Visual Recognition Language Detection Text Extraction Microformat Parsing Feed Detection Keyword Extraction Linked Data Support Image Link Extraction Image Tagging Face Recognition Classification Author Extraction Taxonomy
  • 27.
    © International BusinessMachines 2015 IBM Watson Personality Insights Enables deeper understanding of people's personality characteristics, needs, and values to help engage users on their own terms The IBM Watson Personality Insights service uses linguistic analytics to infer cognitive and social characteristics, including Big Five, Values, and Needs, from communications that the user makes available, such as email, text messages, tweets, forum posts, and more. By deriving cognitive and social preferences, the service helps users to understand, connect to, and communicate with other people on a more personalized level. “My words” show my personality. “Big 5” Traits: – openness; – conscientiousness; – extraversion; Calibrated with standard tests. – agreeableness; – neuroticism; and – 47 others. www.ibm.com/WatsonDeveloperCloud 27
  • 28.
    © International BusinessMachines 2015 IBM Watson Obtaining Insight into an Individual by Analyzing their Writing Watson Personality Insights (Signal) (Meaning) 28
  • 29.
    © International BusinessMachines 2015 IBM Watson 29
  • 30.
    © International BusinessMachines 2015 IBM Watson Concept Insights Explores information based on the concepts behind your input, rather than limiting investigation to findings based on traditional text matching The Concept Insights service maps user-input words to the underlying concepts of those words based on training on English Wikipedia data. Doing so can broaden the user's investigation beyond the actual words used in an inquiry. Two types of associations are identified: explicit links when an input document directly mentions a concept, and implicit links which connect the input documents to relevant concepts that are not directly mentioned in them. Users of this service can also search for documents that are relevant to a concept or collection of concepts by exploring the explicit and implicit links. www.ibm.com/WatsonDeveloperCloud Link words between documents and concepts Explicit linkage when a document names a concept Implicit linkage when concept indirectly mentioned Trained on English Wikipedia 30
  • 31.
    © International BusinessMachines 2015 IBM Watson 31 Natural Language Classifier Interpret natural language and classify it with confidence The IBM Watson Natural Language Classifier service enables developers without a background in machine learning or statistical algorithms to create machine-learning, natural language interfaces for their applications. The service interprets input text (questions or other) and returns a corresponding classification with associated confidence levels. The return value can then be used to trigger a corresponding action, such as redirecting the request or answering the question. BETA www.ibm.com/WatsonDeveloperCloud • Classifies text against predefined categories – Classification (with confidence) – Then answer question or trigger an action • Easy to train – List of [text, class] pairs
  • 32.
    © International BusinessMachines 2015 IBM Watson 32
  • 33.
    © International BusinessMachines 2015 IBM Watson © 2015 International Business Machines Corporation IBM Watson 2015 Australian Open # # EXTRACT THE ANSWERS FROM A WATSON QUERY # watsonQuery = JSON.parse(...)["question"] answers = watsonQuery["evidencelist"] answer = answers[@answerIndex] @answerConfidence = answer[“value”] @answerText = answer[“text”] @answerValue = answer[“value”] 33
  • 34.
    © International BusinessMachines 2015 IBM Watson 34 “ a” By 2018 half of all consumers will regularly interact with services based on cognitive - IDC FutureScape
  • 35.
    © International BusinessMachines 2015 IBM Watson www.ibm.com/ watson Watsonjobs developer.ibm.com/watson msk@us.ibm.com35