10 Things You Need to Know About the Technology Behind Watson

What is so fascinating about a Computer System vs. Quiz Show?  The popularity of America’s favorite quiz show, Jeopardy!, stems from the unique challenges it poses to its contestants: the breadth of topics; the puns, metaphors, and slang in the questions; the speed it takes to buzz and answer.

These factors make Jeopardy! the perfect testing ground for Watson, the IBM computing system that can understand the complexities of human language and return a single, precise answer to a question.

Next month, IBM’s Watson will play Jeopardy! (on live network TV) with two of the all-time champions.  IBM offered a press sneak peek this week at a practice round that included Alex Trebek.  After seeing the clips, I am getting excited and am convinced this technology breakthrough is something special.  Here s what you need to know:

1.  What is Watson?

Watson is the name for IBM’s Question Answering (QA) computing system, built by a team of IBM Research scientists and university collaborators who set out to accomplish a grand challenge – to build a computing system that rivals a human’s ability to answer questions poised in natural language with speed, accuracy and confidence. It leverages Natural Language Processing (or NLP) to process extreme volumes of text.

Watson is powered by an IBM POWER7 platform to handle the massive analytics at speeds required to analyze complex language and deliver correct responses to natural language clues.  The system is a combination of current and new IBM technologies optimized to meet the specialized demands of processing an enormous amount of concurrent tasks, and content while analyzing content in real time.

2.  What is Natural Language Processing?

Natural language processing (NLP) is a field of computer science and linguistics concerned with the interactions between computers and human (natural) languages. It describes a set of linguistic, statistical, and machine learning techniques that allow text to be analyzed and key information extracted for other uses such as Question Answering or Content Analytics.

3.  What are QA and DeepQA?

Question Answering (QA) is the task of automatically answering a question posed in natural language. It involves first trying to understand the question to determine what is being asked. Then by analyzing a wide variety of disparate content mostly in the form of natural language documents to find reasoned answers. And finally, to assess based on the evidence, the relative likelihood that the found answers are correct. Collections can vary from small local document collections, to internal organization documents, to compiled newswire reports, to the World Wide Web. QA is regarded as the next step beyond current search engines.

DeepQA goes well beyond simple question reformulation or keyword analyses. Queries that include disambiguation, unfamiliar syntax, spatially or temporally constrained questions – or simply bad question framing – require a deeper level of content and text analysis.

4.  What is unique about the QA implementation for Watson?

Competing with humans on Jeopardy! poses an additional set of challenges, including, the variety of question types and styles, the broad and varied range of topics, the demand for high degrees of confidence and speed required a whole new approach to the problem.

5.  How does QA technology compare to document search?

The key difference between QA technology and document search is that document search takes a keyword query and returns a list of documents, ranked in order of relevance to the query (often based on popularity and page ranking), while QA technology takes a question expressed in natural language, seeks to understand it in much greater detail, and returns a precise answer to the question.

I touched on the frustrations of search in my previous posting Goodbye Search … It’s About Finding Answers … Enter Watson vs. Jeopardy

6.  How does Watson compare to the chess-playing system, Deep Blue?

Deep Blue demonstrated that computers can solve problems once thought the exclusive domain of human intelligence, albeit in perhaps very different ways than humans do.  Deep Blue was an amazing achievement in the application of compute power to an extraordinarily challenging but computationally well-defined and well-bounded game.  By searching and evaluating a huge space of possible chess board configurations, Deep Blue had the compute power to beat a grand master.

Watson faces a challenge that is entirely open-ended and defies the sort of well-bounded mathematical formulation that fits a game like Chess.  Watson has to operate in the near limitless, ambiguous and highly contextual domain of human language and knowledge.  Ultimately Watson’s scientific goal is to demonstrate how computers can get at the meaning behind a natural language question and infer precise answers from huge volumes of content, with justifications that ultimately make sense to humans.

Rather than challenging the human to search a vast mathematical space, the Watson project challenges the computer to operate in human terms.  Watson strives to understand and answer human questions and to know when it does and doesn’t know the answer.  The capability to assess its own knowledge and abilities, something humans find relatively easy, is exceedingly difficult for computers.

7.  How would this QA technology be used in a business setting?

DeepQA technology provides humans with a powerful tool for their information gathering and decision support.  One of many possible scenarios could be for the end user to enter their question in natural language form, much as if they were asking another person, and for the system to sift through vast amounts of potential evidence to return a ranked list of the most compelling, precise answers along with links to supporting or refuting evidence.  Other important scenarios will use DeepQA to analyze a collection of content and data representing a problem, for example a technical support problem or a medical case.  DeepQA will start to search for solution gathering and assessing evidence from many disparate data sources engaging human users to help provide the missing pieces of information that can help arrive at a solution or for example a differential diagnosis, in the case of medicine.

In addition, these answers would include summaries of their justifying or supporting evidence, allowing the user to quickly assess the evidence and select the correct answer.

Business applications include Customer Relationship Management, Regulatory Compliance, Contact Centers, Help Desks, Web Self-Service, Business Intelligence and more.  These applications will demand a deep understanding of users’ questions and analysis of huge volumes of natural language, structured and semi-structured content to rapidly deliver and justify precise, succinct, and high-confidence answers.

8.  What is the role of Unstructured Information Management Architecture (UIMA) in DeepQA and the Watson project?

Unstructured Information Management Architecture (UIMA) is the IBM developed open-source framework for analysis of unstructured content, such as natural language text, speech, images and video, which Watson uses to integrate and deploy a broad collection of deep analysis algorithms over vast amounts of content.

A number IBM ECM products are based on and leverage UIMA today. IBM Content Analytics, IBM OmniFind Enterprise Edition, IBM eDiscovery Analyzer and IBM Classification Module all are powered by, or benefit from, natural language processing and UIMA.

9.  Are any Enterprise Content Management (ECM) technologies actually part of Watson?

Yes, IBM Content Analytics is part of Watson.  After the question is asked, the text needs to be processed using natural language processing.  IBM Content Analytics (LanguageWare) and other techniques (secret sauce) are used to process the text, and understand the question, as part of the complex processing required to fully answer questions with confidence.  I will tackle this issue in more detail in my next blog posting.

IBM Content Analytics (or ICA) is a content analysis platform used to derive rapid insight from content and data.  It can transform raw information into business insight quickly without building models or deploying complex systems enabling businesses to derive insight in hours or days … not weeks or months.  It’s easy to use and designed for any knowledge worker who needs to search and explore content.  ICA can be extended for deeper insights by integrating to Cognos, SPSS, InfoSphere, Netezza and other Business Intelligence, Analytics and Data Warehouse systems. 

The ICA product itself includes tooling (LanguageWare) which is used to customize NLP processing and build industry or customer specific models and solutions.  This capability is at the core of natural language processing and is the very same ICA capability that is used in Watson.

10.  Who is going to win on February 14-16th?

My prediction … Watson is.

I watched the video yesterday of the practice rounds and Watson is impressive.  Watson performed impressively against Ken Jennings and Brad Rutter (the two contestants and the all-time champions).

So … who won the Jeopardy! practice round?

Watson won handily …


Watson’s score was $4,400, beating Jennings by $1,000 and nearly quadrupling Rutter’s score.

IBM will donate 100% of Watson’s winnings to charity, while Rutter and Jennings said they will each donate 50% of their prizes. 

I am going to host a viewing party for colleagues, friends and family.  This is going to be exciting and fun … I can’t wait.  As an IBMer, I’ll be rooting for Watson to win but not for the obvious reason.  My rooting is really about my passion for the amazing technology breakthrough and the power of content analytics. 

Who do you think will win?  Leave me your thoughts below.

41 thoughts on “10 Things You Need to Know About the Technology Behind Watson

  1. Very timely piece following a weekend of football and IBM Watson commercials. Since this sounds like a learning application, does it need to run 7/24/365 over the content for the Q/A and does it need massive computing resources to operate. Is this considered an AI application and does IBM see this as a cloud based service ?

    Thank you.

    1. Here is a video that goes into more detail about the details of how it works and relationship of QA to AI. It’s not AI … at least as defined by most people.


      Yes it requires alot of hardware to run. Watson is powered by off-the-shelf IBM POWER7 platforms to handle the massive analytics at speeds required to analyze complex language and deliver correct responses to natural language clues. The system a combination of current and new IBM technologies optimized to meet the specialized demands of processing an enormous amount of concurrent tasks, and content while analyzing content in real time.

  2. Watson will win. It will be interesting to see how the Watson technology is deployed as it goes from displaying to doing. I could see a Watson-type system tasked solely with learning from social media 24/7, and then pooling that information somewhere to bring about the ability for another Watson-based system to provide SaaS research and information aggregation. And if the customers already had the IBM ECM pieces, especially full-text-based search within their documents, the information could then be scoped and filtered appropriately, all within the context of a real-time query. “Watson, should I set the upgrade cycle for our new app to be 12 months or 18 months?”

    “Your past history shows that shorter time frames lead to larger chances of bugs. Additionally, the last five upgrades in your market segment showed an 83% negative reaction among Tweets from users in the segment. There is also a correlation between a three-month marketing push for upgrades and positive reactions, so my recomendation is to go with an 18-month cycle with a planned three-month public relations ramp-up.”

    This technology makes so many potential applications possible it’s hard to rein them in.

  3. Is neural network processing used to achive solutions (answers)?
    If so, how is it combined with the UIM technology to arrive at an
    acceptable response ??

    1. Watson fundamentally works like this … (1) The question being asked is read as text … just like the humans can do. (2) The text is processed using IBM Content Analytics (LanguageWare) and other proprietary techniques (secret sauce) to understand the question. UIMA is used for this part as well. (3) Watson examines it’s known knowledge sources to formulate possible answers (also secret sauce). Watson is not connected to the Internet. (4) Watson uses a variety of algorithms and techniques to determine confidence levels of potential answers (more secret souce). (5) Based on confidence level, an answer is selected and a decision made whether to answer or not. (6) If yes, a mechanical device is triggered to push the answer controller in the same way Jennings and Rutter have to. All of this is done in 3 seconds or less … which is truly amazing.

    1. You know Craig, you are now responsible for keeping us posted of the winnings in February. This will be very exciting to watch. I’d also bet on Watson.

      Cylon’s may well be on their way after this performance.

      1. Ilona – I am completely fascinated by how this is evolving. The shows are scheduled to be aired on February 14-16th and anyone can watch of course. I plan another blog posting or two before the event.

  4. What Jeopardy and Watson are truly representing is a competition between human ability (human contestants) versus human endeavor (Watson).
    Will IBM ever allow other technologies/companies to also compete along the side of Watson to cater human endeavor any further?

  5. Craig, as always, you explain complex technology in a precise but understandable way. Thanks for sharing your insights. With best regards from Lise Neely

  6. What are the chances that this technology could actually pass the Turing Test? Will we see personal versions of this in the next 10 years? Can the system actually learn while it is in operation?
    If the answer to all three questions is “yes,” I can see this becoming the “must have” accessory for the 2020s.

    1. In my opinion this a major advance forward but I don’t think we’ll be passing the Turning test quite yet. How the technology can be commercialized will likely be driven by market need going forward. Personal versions … sure why not 🙂 There is a pre-game learning component to Watson but I don’t know if Watson can also learn while in operation.

      1. Watson does “learn” a bit while playing the game. Jeopardy category titles give a clue as to how to answer correctly. Players – both human and Watson – have an opportunity to figure out that clue with the easier, lower value answers before taking on the more challenging ones.

      2. After looking into this a little deeper, I am changing my point of view on this. I supposed it depends on your understanding of the Turing test but I do think Jeopardy! is a proof point of a machine’s ability to pass it.

  7. I’m routing for Watson, not just because I’m an IBMer, but because I want to see people (not in the technology industry) embrace what technology is really capable of providing. I’m surprised to hear so many not wanting to see him win as they think it will “replace” humans. We need to show how technology of this great magnatude can do more good in the world. Can’t wait to see it tonight. Is anyone pitching the story to Oprah, I’d love to see her interview Watson! She’ll embrace the value this brings to everyone!

  8. To Beth Rieser- Aside from the fact that Watson only proves IBM’s work, not other technologies, is important to understand and sympathize with the idea that the future of machine intelligence will directly compete with human performance. I do share your excitement, but I the same time, we must all be aware and approach the inevitable future with caution and respect to others. Simply because in the same fashion that a medical assistance can be replaced with AI, you and I will be too. Question: do you embrace a human-like intellect or maybe a superior intellect. I personally think, that human-like intellect is not a good idea (too many risks involved), but intellect that is many times superior maybe what humanity needs.

  9. Being in software business development the quantifiable measures of a client purchasing an application have been, show me how I will achieve my ROI and how will it reduce headcount. So should this type of linquistic enabled, content plus context analyzing system send a wake up call to the call center nations, yes.

    However, used wisely and delivered at the right price point can help to level the economic playing field for many people wanting access to hard to find informational services across a variety of industries (e.g. legal, health, finance, education, government services etc).

Leave a Reply

%d bloggers like this: