the business implications of watson - ibm · pdf filethe business implications of watson ......
Post on 30-Mar-2018
220 Views
Preview:
TRANSCRIPT
IBM Research
IBM Proprietary - DO NOT DISTRIBUTE
Dr. Carl E. Abrams
Financial Services Sector Business Executive
IBM Research
The Business Implications of WATSON
IBM Research
IBM Research Hawthorne
IBM Proprietary - DO NOT DISTRIBUTE[2]
IBM RESEARCH
Outline
► IBM Research Overview
► What is Watson and Why Now?
► How it Works
► Business Implications
IBM Proprietary - DO NOT DISTRIBUTE[4]
IBM RESEARCH
IBM Research OverviewFamous for its science and vital to IBM
Innovation that Matters
FOAKFirst of a Kind
EBOEmerging
BusinessOpportunities
ResearchServices
Briefings
Business
TechnologySociety
New Insights
1970s 1980s 1990s 2000s
Collaborate on client-specific technology and business solutions
Centrally Funded
Joint Programs
Research in the Marketplace
Collaborative Innovation
� Corporate funded research agenda
� Technologytransfer
� Collaborative team
� Shared agenda
� Effectiveness
� Work on client problems
� Create business advantage for clients
� Industry-focused research
Almaden Research Center, 1955
Austin Research Laboratory, 1995
India Research Laboratory, 1998
Zurich Research Laboratory, 1956
China Research Laboratory, 1995
Haifa Research Laboratory, 1972
Tokyo Research Laboratory, 1982
Thomas J. WatsonResearch Center, 1961
ResearchPartnership
Brazil Research Laboratory, 2010
Melbourne Laboratory, 2010
IBM Proprietary - DO NOT DISTRIBUTE[5]
IBM RESEARCH
Diversity of Disciplines at IBM Research
ChemistryComputer Science
ElectricalEngineering
Materials Science Mathematical
Sciences Physics
Service Science, Management &
Engineering
Behavioral Sciences
BusinessInnovation
TechnologyInnovation
Social Innovation
Demand Innovation
Science & Engineering
Business & Management
Social & Cognitive Sciences
Economics & Markets
IBM Research
IBM Proprietary - DO NOT DISTRIBUTE[6]
What is WATSON and Why Now?What is WATSON and Why Now?
© 2010 IBM Corporation
IBM Research
7 IBM Proprietary - DO NOT DISTRIBUTEIBM Confidential
What is WATSON today?
Watson is a workload optimized system based on IBM DeepQAarchitecture running on a cluster of IBM®POWER7® processor-based servers.
Given an English language question on almost any subject, Watson can provide a set of potential answers in 3 to 5 seconds.
Watson, purpose built for Jeopardy! is powered by 90 POWER7 servers with 2880 cores and 15,000 Gigabytes of memory
It was built to do one thing extremely well – play Jeopardy!
Today Watson is a system
designed for Jeopardy!
question and answers
© 2010 IBM Corporation
IBM Research
8 IBM Proprietary - DO NOT DISTRIBUTEIBM Confidential
What is WATSON today?
� 90 x IBM Power 7501 servers
� 2880 POWER7 cores
� POWER7 3.55 GHz chip
� 500 GB per sec on-chip bandwidth
� 10 Gb Ethernet network
� 15 Terabytes of memory
� 20 Terabytes of disk, clustered
� Can operate at 80 Teraflops
� Runs IBM DeepQA software
� Scales out with and searches vast amounts of unstructured information with UIMA & Hadoop open source components
� Linux provides a scalable, open platform, optimized to exploit POWER7 performance
� 10 racks include servers, networking, shared disk system, cluster controllers
1 Note that the Power 750 featuring POWER7 is a commercially available server that runs AIX, IBM i and Linux and has been in market since Feb 2010
IBM Proprietary - DO NOT DISTRIBUTE[9]
IBM RESEARCH
The Brain vs. the Computer
► The Human Brain
■ The human brain is very slow( about a million times slower than electronic circuits)
■ Parallel architecture
♦ hundred billion neurons ( if each neuron is equated to a byte then the human brain can store about 1 petabyte)
♦ each of which has an average of a thousand connections to other neurons.
♦ hundred trillion simultaneous computations
■ Although each interneuronal connection is capable of performing only about two hundred computations each second, a hundred trillion computations being performed at the same time add up to about twenty million billion calculations per second!
► The Computer
■ Essentially single computation at a time
■ Simulate two billion computations per second
■ About 2018 machines will reach the computational capacity and storage density of the human brain…however….it is the connectivity that counts as much as the speed
Ref; http://www.kurzweilai.net/meme/frame.html?main=/articles/art0267.html
IBM Proprietary - DO NOT DISTRIBUTE[10]
IBM RESEARCH
Processing Power
2010-02-16
Compute-power is abundant:Yesterdays supercomputers are today's mobile phones
Tera FLOP
Giga FLOP
Mega FLOP
Kilo FLOP
1 FLOP
0
Te
ra F
LO
Ps
Te
ra F
LO
Ps
IBM Blue Gene/P*
84% CGR
ASCI WhiteUS Dept of Energy
SETI@HomeEarth
Simulator
$1000 buys
DesktopGaming
IBM Roadrunner
http://www.top500.org
ATMs GlobalTrading Floors
Batch Processing
Calculation Machines
Web Channels
Natural Language
Early Tabulators
after Kurzweil, 1999 & Moravec, 1998
IBM Blue Gene/Q (10 Petaflops 2012)
IBM Sequoia (20 Petaflops 2013)
IBM Proprietary - DO NOT DISTRIBUTE[11]
IBM RESEARCH
What can you do with 20 Petaflops? Or 20 quadrillion mathematical processes per second?
► If each of the 6.7 billion people on earth had a hand calculator and worked together on a calculation 24 hours per day, 365 days a year, it would take 320 years to do what Sequoia will do in one hour.
►20 petaflops could offer a 50x improvement in our capability to predict earthquakes, allowing scientists to predict an earthquake's effects on a building-by-building basis across an area as large as Los Angeles County.
►20 petaflops could also provide a 40x improvement in our capability to monitor and forecast weather. This would allow forecasters to predict local weather events that affect areas 100 meters to one kilometer in size, down from their current ten-kilometer ability.
►The Sequoia will be powered by 1.6 million cores (specific 45-nanometer chips in development) and 1.6 petabytes of memory. It will be housed in 96 racks spanning roughly 3,000 square feet.
►20 petaflops coupled with a petabyte of real memory will enable discrete event simulation of the entire financial system at the transaction level
IBM Proprietary - DO NOT DISTRIBUTE[12]
IBM RESEARCH
Storage Class Memory – Timeline & Systems Impact
2010-02-16
Storage Class Memory Characteristics:
– Holding Petabytes in memory can lead to new levels of real-
time analytics for fraud, pattern detection, business
optimizations , …
ArchivalActive StorageMemoryLogic
2008
1990
2014+ RAM Storage Class Memory Disk Tape
RAMFlash
SSD Disk Tape
RAM Disk Tape
Storage Class Memory is the next major advance in memory and storage systems
SSD Improvement:•300% throughput•45%-75% IO
SCM Improvement:•100X less power•2X denser than Flash•1000X more durable•Scalable beyond 22nm•By 2020 1000X to 10000x better performance, power and floor space usage
Racetrack Memory Nears the Finish Line
Physicsworld.com (Institute of Physics Jan 10, 2011
IBM researchers have moved another step closer to commercializing "racetrack memory" – a new technology that uses magnetic nanowires as high-density data storage devices. Racetrack involves moving magnetic domain walls – the boundaries between regions of opposite magnetization – along a nanowire using small spin-polarized current pulses. It could make for a new type of magnetic memory that can store up to 100 times more data than existing random-access memories (RAMs).
IBM Proprietary - DO NOT DISTRIBUTE[13]
IBM RESEARCH
Answer please in the form of a question…
What is Watson?
Financial Services GTO 2011 - DO NOT DISTRIBUTE[14]
IBM RESEARCH
Jeopardy - The IBM Challenge
2011-02-23Financial Services GTO 2011 - DO NOT DISTRIBUTE
► 1997 - Chess
■ A finite, mathematically well-defined search space
■ Large but limited number of moves and states
■ Everything explicit, unambiguous mathematical rules
► 2011 - Human Language
■ Ambiguous, contextual and implicit
■ Grounded only in human cognition
■ Seemingly infinite number of ways to express the same meaning
[14]
© 2011 IBM Corporation1515
Broad Domain
Our Focus is on reusable NLP technology for analyzing vast volumes of as-is text. Structured sources (DBs and KBs) provide background knowledge for interpreting the text.
We do NOT attempt to anticipate all questions and build databases.
In a random sample of 20,000 questions we found2,500 distinct types*. The most frequent occurring <3% of the time.
The distribution has a very long tail.
And for each these types 1000’s of different things may be asked.
*13% are non-distinct (e.g, it, this, these or NA)
Even going for the head of the tail willbarely make a dent
We do NOT try to build a formal model of the world
Financial Services GTO 2011 - DO NOT DISTRIBUTE[16]
IBM RESEARCH
Watson’s analytics is more than search
►Web search returns a ranked list of “possible” web pages containing the requested data ■ Search engines results are based on
popularity and page ranking■ User must still analyze results – sift
through a web page – to find the best answer
►Watson’s analytics understand the structure and wording of the question asked■ Finds a specific answer■ Ranks its answer and provides a level of
“confidenc” that it is correct based on experience
►Watson answers “natural language”questions■ Can contain puns, slang, jargon and
acronyms that must all be evaluated
2011-02-23Financial Services GTO 2011 - DO NOT DISTRIBUTE[16]
© 2011 IBM Corporation17
What Computers Find HardComputer programs are natively explicit, fast and exacting in their calculation over numbers and symbols….But Natural Language is implicit, highly contextual, ambiguous and often imprecise.
�Where was X born?One day, from among his city views of Ulm, Otto chose a water color to
send to Albert Einstein as a remembrance of Einstein´s birthplace.
�X ran this?If leadership is an art then surely Jack Welch has proved himself a
master painter during his tenure at GE.
Structured
Unstructured
© 2011 IBM Corporation18
Basic Game Play
Technology Classics The Great
Outdoors
Speak of
the Dickens
Mind Your
Manners
Before and
After
$200 $200 $200 $200 $200 $200
$400 $400 $400 $400 $400 $400
$600 $600 $600 $600 $600 $600
$800 $800 $800 $800 $800 $800
$1000 $1000 $1000 $1000 $1000 $1000
6 Categories6 Categories
5 Levels of Difficulty
5 Levels of Difficulty
�1 of 3 Players Selects a Clue
�Host reads Clue out loud
ALL POLICEMEN CAN THANK STEPHANIE KWOLEK FOR HER INVENTION OF THIS POLYMER FIBER, 5 TIMES TOUGHER THAN STEEL
TECHNOLOGY
� All Players compete to answer
� 1st to buzz-in gets to answer
� IF correct
�earns $ value
�selects Next Clue
� IF wrong� loses $ value � other players buzz again (rebounds)
� Two Rounds Per Game + Final Question
� ONE Daily Double in First Round, TWO in 2nd Round
© 2011 IBM Corporation19
Different Types of Evidence: Keyword Evidence
celebrated
India
In May 1898
400th anniversary
arrival in
Portugal
India
In May
Garyexplorer
celebrated
anniversary
in Portugal
Keyword MatchingKeyword Matching
Keyword MatchingKeyword Matching
Keyword MatchingKeyword Matching
Keyword MatchingKeyword Matching
Keyword MatchingKeyword Matching
19
arrived in
In May, Gary arrived in India after he celebrated hisanniversary in Portugal.
In May 1898 Portugal celebrated the 400th anniversary of this explorer’s arrival in India.
Evidence suggests “Gary” is the answer BUT the system must learn that keyword matching may be weak relative to other types of evidence
© 2011 IBM Corporation20
On 27th May 1498, Vasco da Gama landed in Kappad Beach
On 27th May 1498, Vasco da Gama landed in Kappad Beach
celebrated
May 1898 400th anniversary
arrival in
In May 1898 Portugal celebrated the 400th anniversary of this explorer’s arrival in India.
Portugal
landed in
27th May 1498
Vasco da Gama
Temporal Reasoning
Statistical Paraphrasing
GeoSpatialReasoning
explorer
On 27th May 1498, Vasco da Gama landed in Kappad Beach
On the 27th of May 1498, Vasco daGama landed in Kappad Beach
Kappad Beach
Para-phrase
s
Geo-KB
DateMath
20
India
Stronger evidence can be much harder to find and score.
The evidence is still not 100% certain.
�Search Far and Wide
�Explore many hypotheses
�Find Judge Evidence
�Many inference algorithms
Different Types of Evidence: Deeper Evidence
Financial Services GTO 2011 - DO NOT DISTRIBUTE[21]
IBM RESEARCH
Question100s Possible
Answers
1000’s of Pieces of Evidence
Multiple Interpretations
100,000’s scores from many simultaneous Text Analysis Algorithms100s sources
. . .
HypothesisGeneration
Hypothesis and Evidence Scoring
Final Confidence Merging & Ranking
SynthesisQuestion &
Topic Analysis
QuestionDecomposition
HypothesisGeneration
Hypothesis and Evidence Scoring
Answer & Confidence
DeepQA: Massively Parallel ProbabilisticEvidence-Based Architecture
2011-02-23Financial Services GTO 2011 - DO NOT DISTRIBUTE[21]
Financial Services GTO 2011 - DO NOT DISTRIBUTE[22]
IBM RESEARCH
Watson Today: Processes Unstructured Text & 200 Hypothesis/3 seconds
Statistical Ensembleof 600 to 800
Scoring Engines
~30 Machine Learning Models Weigh Scores, Produce
Confidence for Each Question 0<P<1
Hypothetical Question With Greatest Confidence is Chosen
Evidence-Based
Decision Support System
Evidence-Based
Decision Support System
S1S1 S2S2 S3S3 SNSN. . .
StaticData
Corpus
StaticData
Corpus
Answer: A large country in the Western Hemisphere whose capital has a similar name.
Hypothesis Generated from “Answer”Guess Questions Q1, Q2 … Qi
Hypothesis Generated from “Answer”Guess Questions Q1, Q2 … Qi
Question: What is Brazil?
Element Refresh Time
DataCorpus
2 Weeks
Hypothesis Engines
Weeks toMonths
Scoring Engines
Weeks toMonths
Decision Support Engine
4 Days
Watson
3,000 cores;100 TFlops2 TB memory
~ 200 KW
2011-02-23Financial Services GTO 2011 - DO NOT DISTRIBUTE[22]
Financial Services GTO 2011 - DO NOT DISTRIBUTE[23]
IBM RESEARCH
Building on Watson to Expand Domains and Autonomy
Evidence-Based
Decision Support System
Evidence-Based
Decision Support System
S1
S1
S2
S2
S3
S3 SiSi. . .
DynamicData
Corpus
Hypothesis GeneratorHypothesis Generator
Sj
Sj
1
2
3
4
1. Dynamic Data Corpus: constantly mine
data sources (including crowd-sourcing) to
build and update libraries of background
knowledge across domains.
2. Expand Hypothesis Generation to different
domains; leverage crowd-sourcing
3. Add Scorers for Different Input
Modalities: images, video, voice,
environmental, biological…
4. Deeper Reasoning: Improve learning
algorithms to make them real-time and on-
line, and to allow higher-levels of semantic
abstraction.
5. Domain Adaptation Tools: Develop
software tools to help automate domain
adaptation.
Autonomously & accurately identify essential features across multiple domains.Learning systems must understand context to disambiguate.
2011-02-23Financial Services GTO 2011 - DO NOT DISTRIBUTE[23]
Financial Services GTO 2011 - DO NOT DISTRIBUTE[24]
IBM RESEARCH
WatsonIt’s about much more than a quiz show: What it means for business
►A new paradigm in IT■ The computing paradigm for business has changed – traditional computing models are
being replaced by systems that underlie every business process, and the hardware and software performance in those systems is closely tied to actual business performance
■ Watson harnesses IBM’s commercially available, workload optimized POWER7 system which can process thousands of simultaneous tasks at rapid speeds – ideal for complex analytics workloads.
►The future of business is in the data■ Watson-like analytics applied to business can provide answers with a confidence ranking
that can be gleaned from both structured and unstructured data by running hundreds of different kinds of analytical queries across all different kinds of information
■ Applying those innovations from Watson to an organization can help transform business models – meaningful insights from information can help anticipate and shape better business outcomes, improve business operations and boost service to customers
►A “smarter” way to get things done■ Beyond Jeopardy!, the technology behind Watson can be adapted to help solve
business and societal problems – diagnose disease, handle online technical support, parse vast tracts of legal documents
■ Watson-like capabilities applied to healthcare, government, transportation, and other industries can enable Smarter Planet transformations
2011-02-23Financial Services GTO 2011 - DO NOT DISTRIBUTE[24]
[25]
IBM RESEARCH
Training Watson
Baseline 12/06
IBM Watson playing in
the "Winners Cloud"
25
Watson will lose badly!
…. Watson will perform admirably against humans
…Watson can win!
Watson requires training based on a history of right / wrong (or good / bad) answers to questions – just like any expert
Source: IBM Research, MI, SCIP, BCG analysis
IBM Research
© 2006 IBM Corporation26 IBM Proprietary - DO NOT DISTRIBUTE
What makes a good domain for Watson?
1. A large corpus of unstructured information (text at present)
2. Dynamic refresh of the corpus (static at present)
3. Corpus internal to the firm and external to the firm
IBM Research
© 2006 IBM Corporation27 IBM Proprietary - DO NOT DISTRIBUTE
Industry Focus for Watson
� HealthCare
� Financial Services
IBM Research
© 2006 IBM Corporation28 IBM Proprietary - DO NOT DISTRIBUTE
Financial Services
� Supplement to trading information
� Investment banking
� Affluent/Private Banking
� Knowledge management
top related