In my previous blog, we discussed how cognitive business understands, reasons, learns and interacts. Watson is IBM’s brand for cognitive capabilities.
Watson came in lime light when it appeared as a contestant on the US game show Jeopardy! where it handsomely beat two of the show’s best ever contestants (it’s winning total was more than three times that of second placed Ken Jennings). The show poses answers, and contestants must correctly identify the question being asked. For example, one puzzle Watson faced was “Jodie Foster took this home for her role in Silence of the Lambs”. Watson correctly inferred that in this content “took this home” meant “winning an Oscar”. Sometimes “took this home” infers a cold, groceries, or any number of things. Watson’s cognitive system enabled it to behave with human-like characteristics and correctly understand the context.
How does Watson provided answers to those questions?
Watson did the following to provide the correct answer:
1. Question Analysis – In this step, Watson tries to figure out what the question is asking for, and what the answer type (should) be.
2. Hypothesis Generation – Here, Watson creates hundreds of different possible candidate answers. Later Watson will prioritize one of the answer as correct.
3. Hypothesis and Evidence Scoring – Now, Watson weighs each answer. It downgrades or upgrades answers, by looking at the evidence that does not or does support the hypothesis.
4. Final Merging and Ranking – Finally, Watson ranks all the candidate answers, and displays the top 3 answers. It gives confidence scores for each candidate answer and says out the final, first ranked answer.
In 2011 comprised what is now a single API—Q&A—built on five underlying technologies (Natural Language Processing, Machine Learning, Question Analysis, Feature Engineering, and Ontology Analysis). Since then, Watson has grown to a family of 28 APIs. By the end of 2016, there will be nearly 50 Watson APIs— with more added every year! Each API is capable of performing a different task, from recognizing bias in language to gathering information in news reports. In combination, these APIs can be adapted to solve any number of business problems or create deeply engaging experiences. And soon Watson will have ability to interpret data that human senses cannot, such as infrared and sonar.
How Watson is different from Traditional computer Systems?
Traditional computer systems depend on a knowledge base of structured information. They are limited in the kinds of information they can use, and that information must be analyzed and structured for them before they can use it. In contrast, Watson can read unstructured information and figure out its contents, giving it access to a much larger body of information and allowing it to digest that information with much less pre-processing. Because Watson is trained on a corpus of knowledge rather than being programmed, it has more flexibly and so it understands what we are looking for. And it’s ranking of answers helps humans make better decision.
I will explore some of Watson’s APIs / customer use cases in my future blogs. Stay tuned…