SF Technotes

A Cheat Sheet for Understanding Watson

By Michael Castelluccio
April 13, 2017


If someone were to ask you to name a powerful AI-enabled supercomputer, you might immediately begin to mentally sort the names of companies like Google or Oracle or Amazon. Or, you might just remember the moniker for the one with the highest public profile—the one that defeated those Jeopardy champions on television—Watson.









Named for the historic CEO of the company, Thomas J. Watson, IBM’s wunderkind might be the only supercomputer whose dramatic televised debut endowed it with a presence and the beginning of a public personality. Watson doesn’t have a company logo. It has, like so many of the rest of us who join groups online, a self-professed avatar, or symbolic ego. This is Watson.




Watson is conversational by nature and design. Wikipedia defines this AI machine as “a question answering (QA) computing system that IBM built to apply advanced natural language processing, information retrieval, knowledge representation, automated reasoning, and machine learning technologies to the field of open domain question answering.” Watching or listening to Watson work, you might think it’s more personal tutor than search engine. That’s due to its QA technology—an engine that processes spoken or written queries or problems by trying to understand each question far beyond the keyword matching that you see in search functions, and by producing answers that are complex, detailed, and nuanced.


You might recall that one of the first commercial uses for Watson’s software was at Memorial Sloan Kettering Cancer Center in New York, where the computer assisted in utilization management decisions for lung cancer patients. More recently, Watson was enlisted to help create conversation for children’s toys through Chatterbot, and in February of this year, H&R Block announced it would begin using a Watson-based program to find tax deductions.




At IBM’s InterConnect 2017 conference (March 19–23), Watson was given the task of directing the visitors to the sessions offered.


Query page for InterConncect_web



Type in the question, “What’s a Watson?” and you’d get more than 40 relevant sessions returned. One of the more interesting was simply titled “Mythbusters—how Watson works.”


The session was presented by Rob High Jr., IBM fellow, vice president, and CTO for Watson. High began with what IBM understands AI to be. He offered five essential characteristics:


  1. “Cognitive systems understand human expressions—textual, verbal, visual—
  2. By reasoning about the actual intention or problem being addressed.
  3. They learn how to recognize patterns of meaning through examples and feedback,
  4. And they interact with humans on their own terms, and in a way that inspires people.
  5. And they do it at scale!”


Then he addressed five myths about Watson that need correction.


1. Watson is only used by super-large companies that have super-sized computers.


This definitely is true in some instances, but Watson is used by many other organizations. It can be delivered on premises, but it’s also available in the cloud, or you can engage both systems. The data for the system can be “public, private or even crowdsourced—opening the use of Watson to significant communities.”


2. Only super-Ph.D. types can build applications using Watson.


Watson’s built out of a number of APIs (application program interfaces), which makes it open to the four general categories of language, speech, vision, and empathy. Language offers a set of possibilities due to the computer’s ability to understand text and copy “such as taxonomy, keywords, and sentiment analysis.” There’s text-to-speech and speech-to-text. Due to the vision APIs, Watson can recognize faces and classify images. And maybe strangest of all, “There’s empathy where Watson has tone analysis, emotion analysis, and can provide personality insights.” High offers the example of customer feedback analysis in which Watson’s tone analyzer can apply psycholinguistics, emotion analysis, and language to assess the customers’ tone—anger, joy, disgust, fear, sadness, and so on. The Personality Insights function can even provide “how likely a customer is to click on an ad, follow on social media, or buy eco-friendly (among other actions).”


3. Watson can only be trained by mad scientists.


Watson’s CTO explains the falsity of this assumption by reminding us of the AI principle of how those systems learn and recognize patterns through examples and feedback. In order to learn, Watson has a Knowledge Studio where subject matter experts and developers expose the computer to the “linguistic nuances of industries and knowledge domains.” Those teachers can be anyone with domain knowledge.


4. Watson wants to take over the world.


To counter any robot Armageddon scenarious, High offers the machine’s purpose: “Watson wants to have a conversation.” As a virtual agent, the computer has been equipped to recognize and respond to the users’ emotion. It has the four elements that make for a successful virtual agent or chat-bot: “Engage the user; focus on the users’ broader concern; build on an idea; leave the user inspired and satisfied.”


5. Who is the voice of Watson?


It’s not clear why Rob High couched this as a “myth,” but his description of the “talent search” is very interesting. The process sought a long list of essential talents:


“1,000’s of people tried out for the voice of Watson. Gone are the monotone voices of old. Now Expressive SSML [Speech Synthesis Markup Language] and Voice Transformation SSML bring life and a human lilt to computed voices. Expressive voices can be directed to sound apologetic, uncertain and sympathetic. Empathic understanding can help guide a response in line with the customers’ emotions, that is in line with that feeling. Glottal tension, timbre, breathiness and pitch are among many elements that can be uniquely configured to provide just the right sounding response to any situation. Some examples of just some of the Watson-enabled voices that you may have interacted with: tax advice with H&R Block, personalised weather with the Weather Channel, even wedding planning with Meeka (from the Mecasei company).”


To sum up, IBM’s Watson is an AI super-computer smart enough to best the three all-time Jeopardy champions, empathic enough to sense anger or sadness in your voice, and flexible enough to use whatever type and whatever size of database you would like it to work with. For those old enough to remember the first IBM mainframes installed at work, picture a side-by-side comparison of those early, barely capable giants whose only responses came back on wide sheets of green-striped paper and Watson. Hardly seems appropriate to gather both in the same generic category—computers.




For those curious about Watson’s technical specs, the computer is a neural net that uses a number of open source programs.



Watson is a clustered network of 90 IBM Power 750 servers. (Cost: about $3 million.) These can process at a rate of 80 teraflops (trillion floating-point operations) per second. The combined data store of the servers is more than 200 million pages of information, which it processes against six million logic rules.It has 2,880 processor cores and 15TB of RAM. And the computer and its data can be set up in a space that would accommodate 10 refrigerators.



IBM’s DeepQA software includes natural language processing and machine learning. Other essential components include Apache UIMA (Unstructured Information Management Architecture), Apache’s Hadoop programming framework, and SUSE Enterprise Linux Server 11.





Michael Castelluccio has been the Technology Editor for Strategic Finance for 23 years. His SF TECHNOTES blog is in its 20th year. You can contact Mike at mcastelluccio@imanet.org.

0 No Comments