In 2011, IBM's Watson supercomputer won the Jeopardy game show, demonstrating how cognitive computing could make...
sense of vast amounts of unstructured information to deliver straightforward answers.
Now IBM is making this same technology available to developers to bring cognitive computing to Internet applications in general. The first applications were oriented to health care, but IBM and others see opportunities in finance, customer service, telecommunications and other information-flooded industries as well.
"We are coming to an era where information overload is starting to get solved," said Jamie Popkin, a Gartner Inc. vice president. "When you can process information to understand it at a deeper level, you can understand content and derive inferences from it. With Watson, we are getting to the point of being able to process stunning volumes of content quickly so you can take all of the information we have and create action either directly or indirectly."
Other major technology companies are also working on tools and technology intended to make vast amounts of information more accessible and actionable. For example, virtual personal assistants such as Google Now and Apple Siri are enabling a fundamental shift in how users handle content by learning about users' personal preferences and anticipating their activities and behaviors.
The difference: IBM's Watson is focused on deriving deep insights on building deep domain knowledge and learning capability. Put another way, Watson is content-centric, while the others are more focused on personal assistance.
Full speed ahead
IBM sees big potential in Watson's ability to transform business. The company plans to invest more than $1 billion over the next several years and dedicate more than 2,000 employees to bring Watson services to market. IBM is also allocating one-third of its overall research efforts to Watson.
Some of this investment has already started to pay off in greater efficiency. For example, the latest implementation is 24 times faster than the version demonstrated on Jeopardy -- and required only 10% of the hardware of the Jeopardy version.
As part of its outreach to developers and businesses, IBM has announced three new offerings: IBM Watson Discovery Advisor, IBM Watson Analytics, and IBM Watson Explorer. Watson Analytics will make it possible for users to seek out the best answers based on quantitative data from databases and qualitative data from text. The Watson Discovery Advisor service will help reduce the amount of time researchers need to formulate conclusions. Meanwhile, IBM Watson Explorer will provide a unified view of enterprise information.
Early adopters of the Watson technology include these companies:
- Fluid Inc., which is using Watson to improve online shopping experiences for retail businesses to drive customer engagement and conversion.
- MD Buyline Inc., which is using Watson to improve procurement of medical equipment and supplies.
- Welltok Inc., which is developing a Watson app to generate personalized itineraries for encouraging and rewarding healthy behavior.
- Healthline, which is using Watson to help consumers navigate the wide range of disparate information on the Internet and better distinguish scientific evidence from potential "snake oil."
What is cognitive computing?
"Jeopardy demonstrated something we had never seen a computer do, which is navigating the complexity of the human language," said Steve Gold, vice president of the IBM Watson Group. In other words, Watson demonstrated the ability to read and understand vast amounts of information. On Jeopardy, Watson was able to analyze more than 300 million pages of text to derive an answer in less than three seconds.
Watson can understand unstructured information delivered in HTML, the PDF format and Microsoft Word, as well as text-based content. It can also generate hypotheses that recognize different probabilities of various outcomes. In addition, it can learn and adjust its reasoning based on experience.
Watson does have limitations: At this writing, it only understands English. It also doesn't understand images or video, and it can't come up with its own ideas -- yet.
However, Watson shines at understanding content and context. "With traditional search, you don't know why it brings back what it does. That works well for certain tasks, such as choosing a light bulb," Gold said, but added, "But if I want to explore the information and get a response, traditional search is not a vehicle."
The Watson technology is capable of learning as new information comes in. This ability differs from classical computing built on logic and designed with rules to handle structured data. As just one example, "it is physically impossible for a physician to keep pace with new journal articles," Gold said. "You want Watson to continue to learn with new information and outcomes."
Unlike many search technologies, Watson doesn't generate an index to simplify the process of finding answers. Instead, it searches through the vast database of text each time a question is asked. This approach ensures that new answers reflect changes in the accumulated knowledge. It also enables the analysis of sensitive data, such as patient information, to find answers while simultaneously protecting the governance and privacy requirements of the underlying data.
Why the cognitive cloud?
The Watson technology could be installed on a few racks in an enterprise data center. But Gold said that IBM decided to focus on making it available via the cloud for several reasons. For one, this makes the technology more accessible for a wider body of applications. "We wanted to eliminate the consideration of the hardware requirement," Gold said.
The cloud also helps improve the scale, scope and time to market of new applications. It also helps leverage the crowdsourcing of information across different uses in a particular niche or vertical industry, such as health care. If Watson were running on independent systems, each instance would learn independently. By consolidating it onto the cognitive cloud, the entire service gets the benefit of collective learning.
IBM plans on having different Watson instances target various industries differently. For example, the health care instance of Watson is being trained on different information from that used in banking.
By offering Watson services to the broader development community, IBM hopes to enable a much broader set of applications. "Developers at large conceive what they want to do with the cognitive capability. They identify a problem and associate it with an offering or service," Gold said. "IBM works with them to help understand the capabilities of Watson. It also offers a sandbox where developers can upload content and can then immediately interact with the data. The developer creates the app, and the cognitive piece becomes a capability of this."
The next step
Watson represents a major step forward in the trend of using natural language to make sense of large bodies of text and structured data. Gartner's Popkin sees a variety of ways that this basic technology could be improved going forward.
For instance, the voice recognition and interactive dialogue for speakers with different dialects can be enhanced. The problem of processing information in multiple languages needs to be solved. Developers also need to address the need to process different content types to make sense of information in pictures and video. "IBM, Google and Apple understand some of these requirements and will focus on many of these things," Popkin noted.
Other issues include addressing what happens when private enterprise data is mixed with open or publicly available data. It's also important to address what happens when information governance and security become more important aspects of corporate information architecture.
Within five years, organizations should be able to ask a mobile device natural-language questions and expect a direct answer -- or a result that leads to direct action or triggers action in an interactive dialogue, Popkin said. "Cognitive computing will become a pervasive aspect of trying to interact with intelligent machines."