In 2011, IBM's (IBM -0.35%) Watson artificial intelligence (AI) supercomputer beat 74-time-straight Jeopardy! champion Ken Jennings in a man-vs.-machine showdown on primetime television. "Winning" $77,147 to Jennings' $24,000, Watson arguably proved itself three times as intelligent as his human opponent. In so doing, IBM ushered in the new age of AI, in which humans no longer work for food, and computers do all of our thinking for us.
Oh, wait. That actually didn't happen.
Image source: Getty Images.
But in the six years since Watson beat Jennings on Jeopardy!, IBM's AI wunderkind has been in the news plenty. Among other projects, Watson has been tasked with tackling two of humanity's biggest challenges: helping Pfizer figure out a cure for cancer and navigating the U.S. tax code for H&R Block. Yet what exactly Watson is remains something of a mystery to many. In an attempt to get a better handle on Watson, I posed some questions via email to IBM Vice President Ed Harbour, head of the IBM Watson project.
Here's what he had to say. (The interview has been edited to aid clarity and brevity).
Rich Smith: What is Watson, exactly? Is it code? Is it servers? What is the clearest way a layman can envision Watson?
Ed Harbour: Watson is the AI platform for business. It is not one thing, but rather a collection of services and capabilities that include machine learning, reasoning, and decision technologies, as well as language, speech, and vision technologies. These capabilities are designed to learn at scale, reason with purpose, and interact with humans naturally to solve a wide range of practical problems, boost productivity, and foster new discoveries across many industries.
Watson can turn business data -- even data that is unstructured -- into actionable insights that enhance decision-making. Watson can take many forms, from virtual assistant to care manager, research module to customer service agent. Watson leverages the IBM Cloud, offering access to an unprecedented set of enterprise-grade cloud services that can further enhance its function to meet various business needs.
Smith: How does Watson differ from more familiar forms of AI such as Apple's Siri and Amazon.com's Alexa?
Harbour: Watson and Apple's Siri or Amazon's Alexa do totally different things. Their foundation is quite basic and just speech to text to search. Watson delivers a conversation; it delivers answers, alternatives, and evidence-based recommendations with confidence. It learns from interactions and expert training, grows and develops over time. In addition to answers, it offers alternatives and background on why it made the decisions it recommends to users. Watson retains conversations and can understand context and domain. Watson is an AI platform for the enterprise.
Other forms of AI, like Siri or Alexa, rely solely on Q&A and speech to text. They use pre-compiled, human-curated databases or FAQs to look up keywords, as do search engines like Google. Consumer data is not where most of the value is. Eighty percent of the world's data is not on the Web, but rather embedded in businesses and industries, such as client data, financial data, and medical data. Our approach is to build cognitive solutions to help specific industries and businesses tap this data. A key aspect is Watson protects clients' data and any business insights. This is not shared. Watson is also deployed on the IBM Cloud and can be scaled to meet any enterprise needs.
For example, in the case of IBM client H&R Block, Watson was enlisted to understand the "language" of tax to provide the most personalized tax-preparation experience. Watson, with the Tax Professional, helped ensure that consumers were getting the best possible tax outcome and also made the entire tax return process a more collaborative, transparent experience. The main differentiator with Watson is the actionable capability and insights it brings to enterprise clients.
Image source: Getty Images.
Smith: What would you say is the public's biggest misperception about Watson? What is one thing that folks might think that Watson can do that it actually cannot?
Harbour: The biggest misperception about Watson is that it's meant to replace humans. Watson works with humans to enhance the abilities of professionals at every level, from highly specialized surgeons to oil drillers, and automates many basic tasks. However, no matter how advanced the technology, some jobs -- specifically, those that rely heavily on empathy, ethical judgment, and social interaction -- will always be performed better by humans.
Cognitive computing introduces a new level of collaboration between man and machine. It will augment and expand human intelligence, not replace it.
Smith: Can you name one concrete example of something you would like Watson to be able to do that it cannot quite manage yet?
Harbour: The future of technology is rooted in artificial intelligence. In the next three to five years, you'll see advancements that crack the uniquely human nature of communication. For example, Watson has started to be able to detect facial expressions, to combine words, voice, and visual interfaces and form a complete understanding of a conversation. IBM is also further developing Watson's ability to understand different vocalizations of words and how that reinforces a person's emotional meaning.
The true value of Watson and cognitive systems is how it can augment and amplify human abilities. To help us think and perform our jobs better, faster, not to do it for us.
Smith: How human is Watson right now? Would it be possible for someone to interview Watson about Watson?
Harbour: Watson is trained on specific data sets to unearth insights into different industries, tasks, and specialties. Once trained, with its speech, language, and intelligence capabilities, Watson can learn and understand the intention behind a specific command and provide a refined answer for the specific industry or profession it was trained to help.
While it's technically possible to train Watson to answer questions about its own technology, unlike humans, Watson does not have a personality, so it may not make for the most engaging interview subject.
Smith: How does Watson get smarter? By adding data? By refining algorithms? Both?
Harbour: Watson technologies are trained by humans to understand information specific to different industries, specialties, and languages -- in other words, Watson learns in an expert way, not just a general way. This involves training the system to recognize patterns by feeding it large amounts of labeled data and then working with human experts to refine the answers. Through successive rounds of input and feedback from subject-matter experts, Watson's understanding and responses improve.
Looking at the implementation of specific Watson technologies -- Natural Language Understanding and Tone Analyzer are designed to allow developers to quickly bake these functionalities into their apps and see value, without a consultant. Watson Conversation Service enables powerful engagement, and Watson Discovery Service allows for unearthing powerful insights often distributed over vast amounts of documents. Even more robust services like Watson Virtual Agent -- a customer-service chatbot -- come pre-trained with over 105 intents and 35,000 utterances and ready for domains such as general customer service, telco, and retail banking. We also built Watson Knowledge Studio, which allows clients to train Watson on the language of their industry, profession, and domain -- easily. We are continually improving and refining the various underlying algorithms that power the Watson platform.
Smith: How expensive is it for a client to use Watson? How is it billed?
Harbour: Our approach to delivering Watson is flexible and tailored to the clients with both industry solutions and APIs available on the IBM Cloud.
We have several business models for Watson. The overarching theme is value-based. It's really focused on driving scale.
- Data Access/Pay Per Insights: The first business model is Pay Per Insight, so every time you touch that data, or you get an insight ... we are paid for the capabilities or insight.
- Subscription: Some companies want to have this capability. A pharma company wants to have this capability for their entire research team, and so they will pay us a subscription for each researcher.
- Share-Value With Partners: So, where our intellectual property, our data comes together with our key partners' data or insight or domain expertise and we create new value to take cost out of the system or improve capability, we share in that value that we create going forward.
- Licensing: A licensing and partnering model around the world.
Smith: Can you quantify how important Watson is to IBM's business today, in terms of the revenue it produces?
Harbour: Watson is part of our analytics business in our cognitive-solutions segment. In 2Q17, we reported that revenues in this segment grew to $4.6 billion. Over the last 12 months, our strategic imperatives, which includes analytics, has delivered $34.1 billion in revenue, 43% of IBM's total revenue.
Another good indicator of where we are in the cognitive journey is to look at scaling, where we started and where we are now. Watson was just a natural-language machine six years ago. If you want to think of it in terms of a human, it could only "hear" or "read" and only basic information at that.
Today, Watson has grown from that one natural-language QA API into a multitude of services. It has gone from simply speaking English to understanding nine languages. We've added more capabilities and have built extensive data sets industry by industry to train Watson to solve complex industry problems.
It not only can reason over simple trivia; it can reason through complex industry-specific data, like cybersecurity data or cancer data. And it not only "reads." It can "see." Watson can look at medical images and flag ones for radiologists that are unusual -- and it can understand emotion and tone, too.
Image source: Getty Images.
These capabilities are being deployed in enterprises faster. For example, you mentioned cancer research. Watson took a couple of years to learn about oncology after its Jeopardy! win in 2011. But look where we are today. We've entered clinical use in 12 countries and are expanding to more countries later this year. Watson so far has been trained on six types of cancer, with plans to add several more this year.
On taxes, we started with H&R Block last summer. It took us a few months to build a solution for them, and this year somewhere in the ballpark of 11 million people did their taxes at an H&R Block office, powered by Watson.
Another metric of success is embedding Watson across the IBM portfolio. That's our mission. We've been describing Watson as a sliver thread, weaving through multiple areas and segments. We're helping colleagues across our portfolio embed Watson into existing offerings. We've been successful in security, commerce, technology services, and systems, to name a few.
Smith: In addition to cancer and taxes, you are also using Watson to provide individual health and fitness insights to more than 190 million users of Under Armour's connected-fitness platform, while simultaneously providing traffic information to Chevy, Buick, and GMC drivers through General Motors' OnStar service. Is there a limit on the amount of juggling Watson can do, working for so many customers simultaneously?
Harbour: Each client gets their own "instance" of Watson to train with their data, to meet their needs. We can scale to any number by deploying on the IBM Cloud. The applications for this technology are limitless, and we expect to help more than 1 billion people this year.
We are continuously working to advance this adoption -- drawing more developers to use cognitive engines like Watson to build their own apps, working with more businesses and industries so they can incorporate cognitive solutions into their workflows.
Smith: Put Watson in context versus other AIs. How does it stack up against the competition today?
Harbour: IBM has a significant lead in the industry in applying AI technology, having been researching, developing, and investing in AI technology for more than 50 years. We also have the largest industrial research organization in the world and in 2016 led the industry in AI-related patents held.
Unlike other technologies in the market today that are in the experimental or elementary stage, Watson is a mature platform making a real impact on the industries it touches. Watson solutions are now being built, used, and deployed in more than 45 countries and across 20 different industries, solving big, complex societal problems, like cancer and cybersecurity. IBM has also made meaningful progress in other industries, ranging from education to commerce.
Smith: And now look 10 years into the future. What are the chances that in 2027, the average consumer will be able to use Watson in daily life?
Harbour: There's a chance that you use Watson already. For example, in addition to people taking their W-2s to H&R Block for cognitive interviews, people may be using Watson when ordering office supplies from Staples, interacting with a gift concierge via 1-800-Flowers' Gwyn system, checking auto recall and warranty information via Honda's "Ask Dave" interface, or interacting with Harman/IBM-powered smart rooms.
In the healthcare space, Watson is about more than just cancer treatment. It's available to more than 200 million patients globally, collaborating with doctors, helping improve treatment recommendations and helping deliver more efficient care.
AI is playing a bigger and bigger role in consumers' day-to-day lives. By the end of this year, IBM Watson will be available to more than a billion consumers of all kinds, helping them discover the right insurance options, make travel reservations, troubleshoot their IT, answer weather-related questions, get faster service from their bank, and more.
