Skip to content
My Docs

Cognitive Computing

Source: My personal notes from a work Cognitive Computing Course from 2016, hence older terminology, services mentioned

  • Not an emerging technology. It is being used now – example Hilton hotel has a cognitive concierge
  • Self learning software

Introduction from IBM

Section titled “Introduction from IBM”

IBM Watson – quick learning, expansive. Cognitive computing era 2011->

IBM History with Cognitive Computing

Section titled “IBM History with Cognitive Computing”
  • IBM grand challenges: Chess Deep Blue 1997, Human language – Watson 2011 ○ Human + computers working together always beat a human or a computer in chess. ○ Human language is complex, not a computable problem.
  • Play Jeopardy 2009 ○ Answer in 3 seconds ○ Process natural language of questions ○ Compare against former winners

What is cognitive computing? Principles

Section titled “What is cognitive computing? Principles”
  1. Able to learn
  2. Value is expertise (like human)
  3. Evolve over time
  4. Use natural human expression

Hard and Easy to Compute

Section titled “Hard and Easy to Compute”

Computers find these things easy: ○ Math ○ Structured queries (database, search, keyword)

  • Hard to process: natural language is implicit, contextual, ambiguous, and often imprecise
  • Splits a natural language into keywords and process their meaning: “in May 1898 Portugal celebrated the 400th anniversary of this explorer’s arrival in India à celebrated, May 1898, Portugal, explorer, arrival in India

Learning

Section titled “Learning”
  • Text > Syntactic frames (subject, verb, object) > semantic frames (meaningful sentences)
  • Massive parallel probabilistic evidence based architecture
    • Process natural language
    • Assemble evidence from knowledge base
    • Rate probability of answer
    • Parallel processing – unstructured information Services
  • Watson is on Bluemix and can be used for applications
  • It has APIs
  • Watson is available through Bluemix.
    • Threat risk and privacy assessments are required for organizations using of Watson for sensitive data
  • IBM’s recommendation on starting to use the service is to “start small”
  • What do I need to use Watson?
    • Skills: application development experience, knowledge of a programming language, ability to use REST APIs
    • Process: Have an idea for business application of cognitive computing > start with something simple, storyboard it > do user testing > scale the application § Plan > Do > Check > Act

Use Cases

Section titled “Use Cases”
  1. Explore - data
  2. Engage - communications, Q&A, a fixed domain
  3. Discovery - wide domain, research and intelligence
  4. Decision - knowledge + process — medical use cases
  5. Policy

Case dimensions affect classification - size, scope, users, time, connections

Ingest information — to —> Watson –> provides consultation

Imagine You have a room full of assistants that read and remember everything and you can consult them. They provide you relevant pieces of information to your query, useful decision making information, options, or other lines of inquiry.

  • Implied decision triggers form connections
  • Next generation work:
    • Extract semantic info from pictures
    • Streaming data (e.g. internet of things IOT)
    • Predictive analytics

Information Discovery

Section titled “Information Discovery”

Decision making/Expert Q&A

  • Question
  • Check keywords
  • Find and present evidence

versus

Search engine

  • Question
  • Possible answers, can use popularity or attributes
  • Calculate confidence
  • Deliver results

Building a Watson System

Section titled “Building a Watson System”
  • Plan - Selecting and ingestion of documents to build knowledge. Selected is key otherwise garbage in, garbage out.
  • Do - Domain training for application
  • Check - User pilot, testing
  • Act - staged deployment, go to production.

Example: US Government national security intelligence system

Ingestion of documents

  • Watson avoids cognitive bias unless the data ingested is inherently biased.
  • Bias could related to the data volume and quality

Evolution of Watson in Analysis and Decision Support

Section titled “Evolution of Watson in Analysis and Decision Support”
  • Multiple language process. Don’t use straight translation
  • Text, video, image, audio ingestion
  • Decision points:
    • Data instances
    • Reports, queries, data aggregates
    • Predictive models (option 1, 2, 3…)
    • Confidence rating
    • Continuous feedback

Demonstration: Policy application

Section titled “Demonstration: Policy application”

Interface for:

  • Questions
  • Menu
  • Listing of popular questions
  • User > Java app > DB — Open data sets
  • Uses natural language processing (NLP) to display results
  • Results have 3 user interactions:
    • Helpful
    • Not helpful
    • I still need help
  • Opportunity for administration app to manage data sets (e.g. expiry, meta, classifiers)
  • App is trained with classifiers (like keyword filters and categories). For policy it could be poverty reduction, ministry names)

Demonstration: Node Red

  • Visual internet programming for IOT Bluemix has a texting service, Informix, etc.