cognitive computing: can computers learn from experience? - deloitte dbrief, nov 7, 2013

The Dbriefs Technology Executives series presents:

Cognitive Computing: Can Computers Learn from Experience?Rajeev Ronanki, Principal, Deloitte Consulting LLPDavid Steier, Director, Deloitte Consulting LLP

November 7, 2013

Copyright © 2013 Deloitte Development LLC. All rights reserved.

Cognitive computing introduction

How is it different from traditional computing?

Structured vs. unstructured data

Cognitive analytics

Case studies

Convergent solutions

Agenda

Follow @MarkAtDeloitte or #cognitivecomputing during the webcast


Are you familiar with any of these new cognitive computing approaches like IBM’s Watson, Numenta and Google Now?

• Had not heard of them prior to this webcast• Have heard of some of these in the news• Have followed them but have not thought of them as

representing a new class of computing• Have used some of these tools and look forward to more• Don’t know/Not applicable

Poll question #1


Historical timelineThe evolution of Cognitive Computing

• Turing Test published: a computer that exhibits intelligent behavior equivalent to, or indistinguishable from, that of a human.

• Scientific community focuses on machine translation

• Scientific community focuses on AI• First successful NLP systems• MYCIN diagnosed infectious blood

diseases

• Semantic classification & probabilistic parsing are combined in machine systems. Can derive rules and their probabilities

• First commercial database management system tracks huge amount of structured data for Apollo Moon Mission

• Machine learning algorithms for language processing introduced

• Judea Pearl brings probability and decision theory into AI

• Watson. Question –answering system capable of answering questions posed in natural language

• TAKMI (Text Analysis & Knowledge Mining) developed to capture and utilize knowledge embedded in text files – applied to call centers

• TAKMI provides insights on patient groups to help doctors treat groups of patients at a time

• Watson: IBM, WellPoint, Memorial Sloan Kettering use Watson to give doctors treatment options in seconds

• World’s first single molecule computer circuit

• The High Performance Computing Act of 2004 was enacted

• IBM Content, Predictive and Streaming analytics

• Streaming analytics process 5 million messages of market data per second to speed up financial trading decisions

1950’s

1960’s

1970’s

1980’s

1990’s

2000’s

1997

2001

2004

2007

2009-2010

2013

Current

Over the decades…The last 10 years…


Cognitive computing

• Emulates strengths of the human brain, including parallel processing & associative memory

• Enables natural language processing of structuredand unstructured data.

• Understand/leverage big data in real time• Use machine learning to develop context-based

hypotheses

Basics Current Investments

Cognitive Computing can push past the limitations of human cognition, and connect the dots between big data, enabling more informed decisions. A couple of industry examples include:

Academic Commercial Consumer

The development of computer systems inspired by the human brain

Financial Services Implications:Advise in trading, and help identify financial fraud cases

Healthcare Implications:Incorporate all new medical evidence, individual patient histories, and eliminate geographic constraints.

Potential applications of cognitive computing

Boltzmann Machine

Never-Ending Language Learning (NELL)

Saffron Natural

Intelligence Platform

Facebook AI Group

Kngine

IBM Watson

Numenta

Google Now

Apple Siri


How is it different?

• Sequential processing

• Driven by programming language

• Not real-time

• Predefined logic

• Static business rules

• Passive

• Defined input parameter

• Event driven

• Machine learning and natural language based

• Parallel processing different sources at the same time

• Context driven

• Dynamic learning algorithm

• Sensory & mobile based

• Continuous collection and feedback

Traditional Computing Cognitive Computing

User Interface

Application Layer

Processing Platform


Cognitive computing architecture

Analytic Solutions

Data & Analytics Platform

Content Lifecycle Services

Data Corpus Core Engine

Con

tent

Sou

rces

Extract Ingest Discover

Curation Services

Ingestion Services

Enrichment Services

Search Indices

Semantic Models

Derived Knowledge

Question Analysis

Hypothesis Generation

Evidence Scoring

Final Merge and Rank

Cognitive Analytics Applications

Natural Language Processing (NLP) Stack

Machine Learning Modules

Computing Resources (Cloud/On-Premise)

Modeling & Processing Engine


Do you see a role for adopting cognitive computing technologies into your business?

• Still unclear as to how these fit into our strategy• We are tracking these developments but no plans yet• Beginning to develop a roadmap in some areas• Have active plans or projects beginning now• Don’t know/Not applicable

Poll question #2


Not all data is created equal

Structured data is being rapidly augmented by unstructured data

Structured Data

Email & blog contentVideo & social contentPatient records10-Ks & public filingsIndustry reports & research journals

Unstructured Data

Transaction & CRM dataResearch & market dataMainframe dataPoint-of-sale data

AnalysisAnalysisHypothesis

Generation & Scoring

Hypothesis Generation &

Scoring

Final Evidence & Scoring

Final Evidence & Scoring

Initial Question

Initial Question

Final Insights

Final Insights

Advanced processing capabilities such as Natural Language Processing analyze disparate data to yield valuable insights

40-50% annual growth

in digital data volume1

~8xof unstructured data vs. structured data by 20203

1https://www-950.ibm.com/events/wwe/nedc/scesfall12.nsf/RodAdkins.pdf, page 52HP Autonomy Whitepaper: Transitioning to a New Era of Human Information, page 33https://www-950.ibm.com/events/wwe/nedc/scesfall12.nsf/RodAdkins.pdf, page 2

62%annual growth

in unstructured data2

https://www-950.ibm.com/events/wwe/nedc/scesfall12.nsf/RodAdkins.pdf

https://www-950.ibm.com/events/wwe/nedc/scesfall12.nsf/RodAdkins.pdf


Cognitive analytics

Cognitive Computing

Industry Application

Healthcare: A global cognitive analytics provider and major cancer medical center creating a cognitive system that uses cancer patient treatment data to assist oncologists to diagnose and treat patients based on the most current available data.

Retail: Cognitive computers serving as customer service lines, in-store kiosks, or digital store clerks providing answers to customers’ questions around products, trends, recommendations, etc. pulled from millions of data points and structured / unstructured data.

Financial Services: Narrative Science arming investment managers and financial advisors with customized portfolio intelligence, and clients with regular, mobile-friendly account performance summaries, updates, imbalance alerts, changes in risk, etc.

Basic Application

• Drive insights with rich context/unstructured data• Form hypotheses and predictions based on

machine learning to aid real time decision making

• Self-correcting and evolving algorithm that emulates human cognition

• Big data processing

• Platform for machine based learning

• Processing of unstructured data

• Natural language processing

• Processing power and flexibility

Real time decision making Context rich insights/data

Multiple Industries

Cognitive Analytics


Case study: Health Plan improves patient care

ü Combine claims and provider data to gain insight into the true health of the patient

ü Use Cognitive Analytics to identify patients most at risk for hospital re-admissions and high-cost events

ü Refine and enrich the solution as the dataset evolves

ü Augments scarce skillset, and leverages subject matter experts to review/refine instead of performing the primary analysis

Ø Proactively identify at-risk patients and prevent disease before it occurs or gets worse

Ø Over 100 terabytes of claims data, with 200+ points of correlation

Ø Petabytes of medical notes, physical exams, test results, etc. from providers

Ø Ability to process a billion new claims each year is constrained by clinical subject matter expert

High-Level Process

Combine with related

population data

Combine with related

population data

Identify key at-risk patients

Identify key at-risk patients

Refine based on human input

Refine based on human input

Ingest large volumes of claims &

provider data

Ingest large volumes of claims &

provider data

Final candidate list of at-risk

patients

Final candidate list of at-risk

patients

The problem The solution


Case study: Financial Services firm improves customer service

ü Use Cognitive Analytics to detect customer micro segments

ü Track customers with high-value and high attrition risk, and predict future high-value customers

ü Develop personalized marketing strategies to maximize responsiveness and create promotions appropriate for each client

Ø Firm has vast amounts of transactional data, but is light on data scientists and missing the opportunity to see what drives customer behavior

Ø Over 500 terabytes of transactional data and multiple 3rd party data sources

High-Level Process

Integrate with 3rd

party sourcesIntegrate with 3rd

party sources

Create simulations /

generate models

Create simulations /

generate models

Ingest large volumes of

transactional data

Ingest large volumes of

transactional data

Targeted marketing for

high value customers

Targeted marketing for

high value customers

Identify target candidates / promotions

Identify target candidates / promotions

The problem The solution


What barriers to adoption of these technologies do you see in business?

• Too new, we do not have a clear enough understanding of them

• There remain doubts about their effectiveness• We do not have a clear vision of how to adapt them to our

needs given the small number of examples that exist now• It may just take time to build the business case• Building the business case and measuring value from the

solution• Don’t know/Not applicable

Poll question #3


Case study: Unlocking the full picture of health

• Patient has type 2 diabetes• Regular checkup that included a blood sugar test

• Reasons for visit: Hypertension; medication and treatment plan non-compliance

• Recommendations for medication and lifestyle changes to manage stress, alcohol use. and begin smoking cessation

• Perspective into the patient’s lifestyle and reason for medication non-adherence (e.g., does the patient have underlying issues with depression?)

• How long has the patient been having symptoms of hypertension and type 2 diabetes?

• What triggered this occurrence? Does patient have underlying depression?

• What were the final diagnoses?• What were all the costs of the outpatient visit?

Structured Data | Claims• Visit Type: Outpatient• Primary Diagnosis: Type 2 diabetes, hypertension• Lab Result: Elevated random blood sugar levels of

240mg/dl

Unstructured Data | Medical Records & Notes• Reason for visit: Dizziness and blurred vision for 10

days (B/P 170/96)• Patient Background: o 6 month history of drinking a six pack & smoking a

pack dailyo Sleeps over 10 hours a dayo Recent weight gain of 30 lbs.o Has not refilled prescriptions for 5 months

Ove

rvie

wC

oncl

usio

ns a

nd

Hyp

othe

ses

Una

nsw

ered

Q

uest

ions

Through the convergence of structured and unstructured data, we can get the full view of the patient, and make recommendations to improve his or her well-being


WatsonPaths explores complex scenarios and draws conclusions much like people do in real life. When presented with a medical case, it extracts statements based on the knowledge it has learned from medical doctors and medical literature. As medical experts interact with WatsonPaths, the system will use machine-learning to improve and scale the ingestion of medical information. Through this collaboration, WatsonPaths compares its actions with that of the medical expert so the system can get “smarter”.

Case study: IBM WatsonPaths

• A result of a year-long research collaboration with faculty, physicians and students at Cleveland Clinic Lerner College of Medicine of Case Western Reserve University

• Expected to help physicians make more informed and accurate decisions faster and to cull new insights from electronic medical records (EMR)

• Planned to be used by the Cleveland Clinic faculty and students as part of their problem-based learning curriculum and in clinical lab simulations

Source: http://www.research.ibm.com/cognitive-computing/watson/watsonpaths.shtml

http://www.research.ibm.com/cognitive-computing/watson/watsonpaths.shtml


How do these technologies need to develop to become useful tools for your organization?

• Provide clarity on business perspective purpose and effectiveness

• Showcase more wins in areas that matter to us – stress results not science

• Develop packaged offerings for industries that make them easier to adopt

• It will take time, we are not early adopters• Don’t know/Not applicable

Poll question #4


Cognitive Analytics: In summary

How convergence will impact solutions

Making relevant context-based

suggestions and recommendations

Ability to make quicker, more

informed decisions

Access to the right data at the

right time

Ability to capture and process

larger amounts of data

Machine Learning

Artificial Intelligence Statistics &

Decision Science

Natural Language Processing

Distributed ComputingCloud

Computing

Database Technology

Analytics and

Business Intelligence

Visualization

Various Data

Collection Channels

Structured and

Unstructured Data

Real Time Decision Making

Cognitive Analytics

Question and answer

Join us December 5 at 2 PM ET as our Technology Executives series presents:

Cloudy With a Chance of Core: Managing Integration in an Increasingly Complex World


Eligible viewers may now download CPE certificates.

Click the CPE icon in the dock at the bottom of your screen.


Rajeev RonankiPrincipal, Deloitte Consulting [email protected]

David SteierDirector, Deloitte Consulting [email protected]

Contact info

mailto:[email protected]

mailto:[email protected]


Jim Zhu, Deloitte Consulting LLPJeff DeLisio, Deloitte Consulting LLPRui He, Deloitte Consulting LLPFatema Samiwala, Deloitte Consulting LLPRich Carelli, Deloitte Consulting LLPSteven Truong, Deloitte Consulting LLPRiddhi Roy, Deloitte Consulting LLP

Research Team:Ashish Kumar, Deloitte Consulting LLPMarjorie Galban, Deloitte Consulting LLPWilliam Shepherdson, Deloitte Consulting LLPLindsey Tsuya, Deloitte Consulting LLPEugene Chou, Deloitte Consulting LLP

Thanks to our Dbriefs team


• Application Programming Interface (API)• Customer Relationship Management (CRM)• Enterprise Resource Planning (ERP)• Natural Language Processing (NLP)

Acronyms used in presentation


This presentation contains general information only and Deloitte is not, by means of this presentation, rendering accounting, business, financial, investment, legal, tax, or other professional advice or services. This presentation is not a substitute for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your business. Before making any decision or taking any action that may affect your business, you should consult a qualified professional advisor. Deloitte shall not be responsible for any loss sustained by any person who relies on this presentation.

About DeloitteDeloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its network of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about for a detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Please see www.deloitte.com/us/aboutfor a detailed description of the legal structure of Deloitte LLP and its subsidiaries. Certain services may not be available to attest clients under the rules and regulations of public accounting.

Copyright © 2013 Deloitte Development LLC. All rights reserved.Member of Deloitte Touche Tohmatsu Limited

www.deloitte.com/aboutfor

www.deloitte.com/us/about

cognitive computing: can computers learn from experience? - deloitte dbrief, nov 7, 2013

Business

ingest large

scientific

deloitte consulting

create simulations

final candidate

natural language

final evidence

machine learning