All in the name of • Fuzziness of Wordplay Jeopardy - Sherlock’s Hat Evolution of • Incremental v/s Leap supercomputers A priori v/s post priori -Process Improvement Comparison with Deep -Disruptive innovation Blue • The problem of LegacyQA Algorithms: Systems1. NLP -Scientific Paper: eg. of2. TREC Google3. PIQUANT4. Semantic Search • Team Size and Location (Ontology - Swoogle) -One Room RDF/OWL
Querying Template Hardwired Steps: No deterministic Project • Problem Analysis & Objective Requirement Gathering -Win Jeopardy! • Specification Information Classification Accuracy & Quality of Data • Architectural Design -Confidence Estimation • Implementation – Narrowing of Ideas(MECE) Development Benchmarking – • Testing Humans/Machines • Standard Software -Winners’ Cloud/Bake Off Development Lifecycle Model Text v/s Knowledge Search v/s Competing Algorithms Human Mind v/s Machine Learning : Logarithmic to Exponential
Traditional System Watson System The requirements are usually • The requirement wasn’t clear. pretty straightforward. The requirements would need Traditional information to be determined by the system initiative involves program itself during its run- mostly hard coding, with very time. minimal dynamic code • predominate hard coding generation. cannot be used to address the The requirement and Jeopardy problem. It has to be constraints are mostly dynamic predetermined and factored Coding and the logic has to into the development of the be generated in real time. program.
The “problem” that Watson project was commissioned to solve was “win at Jeopardy.” Ferrucci was not addressing a deficiency in a current system or process. He and his team were instead creating a “problem” or puzzle to solve. Second, this was a “green field” implementation. There weren’t any clues on how to break down “win at Jeopardy” in smaller process steps to define visible milestones and deliverables. Due to the nature of the game, requirements would need to be found by Watson in real-time based on the clue data.
Watson was a non-traditional system which seek to go beyond the limits of traditional computing and software programs Hence, It needed to look for new innovation methods for system development to meet to desired goal It used several development approach simultaneously due to lack of any empirical evidence to suggest which SDLC process to adopt. The Watson project a hybrid of Rapid Prototyping, Incremental development approach in parallel
The Watson system utilized a different approach to its methods of analyzing, referencing, and prioritizing user input. The Watson system used an approach that included Parallelism a wide rage of question analytic methods confidence estimates balancing the strictness applied to knowledge sources know as the DeepQA approach. Using an approach know as “parallelism” this Watson would use a variety of external resources from varied resources to gather data and decompose the query and generate possible hypothetical answers.
The project first of its kind on such a large scale, hence it required a new approach A more detailed project objective with key milestones and sub-processes should have been used to measure the progress for better planning The team used rapid prototyping very late. They could have been used this method zearly in the development project to gain insights into the system