In a recent IBM Research industry analyst briefing, “Research Perspectives on Cognitive Computing”, IBM provided details regarding the $1B investment in cognitive computing they will make over the next several years and revealed that 30% of the IBM Research budget and 2000 people will be devoted to the effort.
The initiative has its roots in IBM Watson, but will broaden what Watson means by exploring new levels of cognitive computing, pushing beyond open domain Q&A to work in new ways. For example, “Watson Jr.” is planned as an intelligent personal concierge, a personal assistant that knows your preferences based on your digital life and connects you to things of interest around you. “Chez Watson” will use creative cognitive thinking to generate recipes based on food science, taste and preferences as well as skill level and available ingredients (these are examples of potential applications and actual product names may differ). Many of these new applications will be more focused toward the individual.
In this article I will look at the four levels of cognitive computing, and each successive level‘s more sophisticated capabilities; how IBM cognitive computing functionality will be extended to address these new levels; and other emerging cognitive computing technologies that will be the basis for future commercialized products available in the next one to six years.
The 4 Levels of Cognitive Computing
IBM Research has identified 4 levels of cognitive computing:
- Assistance – Self-service tools where users leverage specific domain knowledge by asking a question and get a range of output including text, voice, graphs and pictures to answer a particular question. The example provided was “How much will it cost to send my daughter to college?” The Watson Engagement Advisor (an announced product) is an example of a cognitive computing tool at this level.
- Understanding – Mapping patterns and connections for better understanding of a situation. These capabilities have been used by utility companies to predict power outages and manage a planned response.
- Decisions – A collaborative approach that uses a decision tree and looks at specific domain knowledge, conflicting points of view, evidence and relationships between factors to provide a reasoned answer to a question. The medical domain is driving this application to be used as a collaborative learning tool. Medical students at the Cleveland Clinic Lerner College of Medicine are using Watson for an interactive, problem-based learning approach to assist in patient diagnosis. An example of decision-oriented application is the MD Anderson (Cancer Center) Oncology Expert Advisor that integrates the knowledge of MD Anderson’s clinicians and researchers to help clinicians develop, observe and fine-tune treatment plans for patients, while helping them predict adverse events that may occur during care.
- Discovery – These kinds of tools will be targeted to subject matter experts (SME’s) in specific domains for high value use cases such as drug discovery or for providing a new application for an existing drug. Knowledge is collected and used along with experimentation and mathematical simulation so that SME’s can understand the domain in a deep way to derive new insights and see new relationships. IBM described it as “an advisor on an SME’s shoulder. These tools are also ideal for legal materials science and any application that involves huge amounts of intellectual property that can help us infer from what we know. The recently announced Watson Discovery Advisor is an example of a tool that operates at the discovery level, with more advanced discovery features being added over time.
Watson’s new capabilities
In order to exploit these new levels of cognitive computing, Watson and/or other IBM cognitive computing offerings will be given new and enhanced powers:
- The power to “reason” using a complicated decision tree that could will enable Watson to look for evidence to support hypotheses to form strategic arguments
- The power to “see” where an IBM cognitive offering could look at related images, analyze them and look for anomalies using reasoning in a complex domain. One application of this would be to help radiologists and cardiologists (only accurate 57% of the time) analyze medical images to identify problems.
- The power to “empathize” could give Watson the ability to look at an individual’s linguistic footprint (emails, tweets, facebook posts etc.) to create a personality profile for yourself and others, allowing users to customize their interactions with that person – called “personality analytics”. IBM has dubbed this application, “Know thyself”.
Emerging technologies in Cognitive Computing
Farther out on the horizon are the next generation of cognitive computing technologies that IBM Research is exploring—technologies that will find practical applications and be commercialized in the longer term. Those projects include:
- SyNAPSE Neurosynaptic Systems – these systems will replace current architectures that are not optimized to handle new computing paradigms and data types (sensor data for example) for the kind of learning developing in cognitive computing. These systems are “brain-inspired” with the goal of reducing the power and time required to compute on very large data sets.
- Mockingbird – a cognitive learning and messaging system based around a community and identifying those ideas that are relevant to that community – particularly through social media such as twitter and on-line discussion forums
- Glimpse – Uses contextual learning and expansion of concepts to help the SME collect and share data and use evidence and analytics to make new discoveries ( will operate at the discovery level of cognitive computing).
- Piazza – A deep contextual search that uses not only keywords but is able to identify relevant information in a search based on a relationship to the search query rather than just the keywords themselves. Use cases include an intellectual property search or discovery of evidence in legal investigations.
- MOOV- is a tool that enables people to make complex decisions effectively, taking into account trade-offs and combining visualizations with analytics. For example, a retailer could use MOOV to design an optimal promotion plan that could maximize sales volume while maintaining desired margin and revenue targets
Clabby Analytics has been following Watson since before the Jeopardy challenge. At first, it was difficult to see where the practical applications for Watson would be. Medical diagnosis seemed like a good fit, to help enhance a physician’s ability to analyze and correlate a patient’s medical history and symptoms with vast volumes of related research, but other applications didn’t immediately come to mind. I like the idea of cognitive computing and its positioning, “cognitive computing is intelligence augmentation where the human is at the center and cognitive systems assist.” Watson is a “smart advisor” that works with humans to improve productivity and efficiency. IBM made this distinction when comparing cognitive computing to artificial intelligence (AI) where the machine is at the center – and this is undoubtedly one of the reasons why AI never gained much traction.
By understanding the cognitive levels, it is easy to see the future applications of Watson beyond just the medical diagnosis field – in drug discovery and legal applications and business scenarios with multiple factors and tradeoffs. To me, the most exciting are the directions and future product ideas that cater to the individual. Who can argue with the idea of a personal assistant? And who among us doesn’t struggle with personal relationships on some level? How best to relate to a colleague or manager? A spouse? A teenager? I like the direction Watson is taking and I am looking forward to following that journey.