Cognitive Computing Explained
IBM is a thought leader in analytics and using that leadership to drive cognitive systems and cognitive business. In fact, Big Blue says that cognitive computing is the company’s moon shot and its investing billions of dollars for cognitive R&D and go to market.
Naturally, there's some healthy skepticism as to whether cognitive technology delivers upon the promise espoused by IBM or is more of a marketing effort seeking perceived differentiation in a crowded business intelligence market.
From my vantage point, cognitive computing is beyond hyperbole, grounded in science and a disruptive technology that will offer competitive advantage to early adopters. However, I also recognize that cognitive is not well understood and that cognitive messaging is slow to resonate in large part as it is often abstract and not tangible. Until business and IT leaders can understand the technology's purpose, comprehend its unique capabilities and vision it within their own industries and businesses, cognitive will not cross the chasm from early adopters to the early majority.
So to lend a hand to the messaging, I'm going to use this post to define cognitive computing, share its unique benefits, provide some interesting real world examples and highlight the approach to implement cognitive systems as part of a CRM business intelligence platform.
Cognitive Computing Defined
Cognition is all about thinking, understanding, learning and remembering. Cognitive computing is all about creating analytics technology that mimics the human brain's ability to perform these functions.
IBM positions cognitive at the top of the Business Intelligence (BI) continuum.
Big Blue doesn't provide a single definition of cognitive systems (which is unfortunate) but instead describes cognitive technology by its capabilities, which include:
- The ability to consume and understand all types of data, and especially unstructured data, via patterns (or anomalies), sensing and interactions
- The ability to reason from data-driven discovery techniques and by forming hypothesis, considering arguments and weighing recommendations
- The ability to learn from expert training, prior interactions and the consumption of more data. Cognitive systems apply machine learning principals in order to learn without being explicitly programmed. They are taught not programmed. They learn from subject matter experts, interaction and experience
- The ability to engage naturally with people by using natural language processing, machine learning algorithms and contextual communications. This technology is able to engage in dialogue with humans comfortably and understand people based on their identities, context, interactions and history
- Cognitive systems are probabilistic and deliver confidence-weighted responses with supporting evidence
These systems tackle complex, ambiguous, uncertain and even conflicting data, questions and problems particularly well. Cognitive systems are able to weigh conflicting evidence and recommend an answer that is "best" rather than "right".
80 percent of the world's data is unstructured big data. Existing analytics solutions struggle to exploit the value of big data in large part because they were designed for quantitative processes, operate within programmed parameters and defined semantics (metadata), and are unable to cope with data ambiguity and the many types of data. These systems also tend to be more suited for reporting on historical data than modeling data for future or predictive outcomes.
Cognitive systems deliver several unique benefits.
- Deeper engagement between man and machine. The technology recognizes its human counterpart's demographics and appends its understanding with context (location, time of the day, environment), behaviors (activities, prior offer-response exchanges, historical transactions) and sentiment (tone, intensity, emotional state) in order to deliver more personalized interactions. As an example, Pandora uses cognitive technology to consider more than 450 attributes to personalize its music service.
- More expertise applied to problem solving. The technology accelerates, enhances and scales human expertise. The cognitive training phase captures the expertise of subject matter experts and top performers, records their know-how, makes their methods repeatable and accelerates the learning and development of others. This can help alleviate the brain drain caused with employee turnover and retirements as well as keep up with each professions growing body of knowledge. U.S. companies spent over $160 billion on learning and development last year, yet it is estimated that 90 percent of new skills learned are lost within a year. Making these skills available on-demand and in context can lower training costs while simultaneously improving performance results.
- Cognitive systems aid in objective decision making and reducing human bias by applying evidence-based processes. This goes a very long way in achieving the elusive goal of replacing intuition, guesswork, estimations and averages with data driven, fact-based decision making backed by confidence levels.
- Continuous improvement over time. The technology learns and evolves, based on new information, results and actions.
- Better decision making as a result of more types and volumes of data to consider along with a problem solving process which includes hypotheses, reasoned arguments and evidence-based recommendations.
Cognitive Examples
The first cognitive system was Watson. Most of the world was introduced to Watson in 2011 when Watson appeared on the television show Jeopardy! and beat the show's two greatest champions. Watson applied Natural Language Processing (NLP) with a large dataset of mostly unstructured data. Watson was without human contact or an Internet connection during the matches. It applied NLP, machine learning and statistical analysis to interpret the clues in the questions, compare multiple answers by confidence levels and respond with the correct answer before its human competitors.
Newer generations of Watson are being trained in for-profit and not-for-profit objectives.
In the healthcare industry, IBM and Memorial Sloan Kettering have created Watson for Oncology in order to help oncologists treat cancer patients with individualized, evidence-based treatment options. Watson analyzes and compares patient data with thousands of historical cases in order to help doctors narrow the options and pick the best treatments for their patients. The doctor is still very much in charge, but Watson assists by making sense of larger data sets which leads to better and more timely results.
In business, Watson is providing answers to improve the complex processes related to mergers and acquisitions. Understanding what business to acquire, how to value it and how to integrate it after the acquisition are multifaceted questions with large financial risks. Companies are today using Watson to assess target markets for the best acquisition candidates, measure synergies and efficiencies, assess value, visualize trade-offs, explore various mixes of purchase consideration (cash, notes, stock, etc.) and identify integration risks.
A retailer is using cognitive computing to improve demand forecasting and inventory allocation. By applying predictive analytics to in-house structured data and eight types of external unstructured data, including Twitter trends, weather patterns and local events, the retailer is able to reduce demand forecasting errors by about 50% and better route inventory to the stores that will move that inventory the fastest.
Cognitive Deployment Approach
The below approach highlights the process to implement cognitive systems.
- Identify the opportunities. Begin by identifying problems to be solved or opportunities to be capitalized on. These focus areas may be related to growth strategies, products, services, operations or business processes. Opportunities should be formed as business case hypothesizes, stack ranked by forecasted payback or other measure important to the organization, and assessed based on the unique capabilities of cognitive systems, such as:
- Questions or decisions that consider several data sources and consume a lot of time;
- Self-service or other inquiries which use a question and answer (Q&A) dialogue that would benefit from Natural Language Processing;
- Questions which may have more than one right answer and therefore benefit from confidence weighted responses.
- Prepare the foundation. Identify the platform, data, training experts and budgeting that will be needed to achieve the desired objectives.
- Begin with human trainers. Cognitive solutions are trained, not programmed, and learn from experts, interactions, outcomes, experience and new information. The training process is often called supervised learning. Cognitive systems depend on humans with domain expertise to train them and define Q&A pairs for the system to learn.
- Build a corpus. A data corpus will likely include structured and unstructured data from many sources. Identifying data sources is a critical task. A common reason many businesses struggle with analytics is that they do not have the right data available to make the right decisions.
- Experiment quickly. In my view, cognitive deployment is a lot like innovation and will benefit from short iterations that include fast moving steps such as ideate, test, measure, adjust, retest, measure, adjust, succeed or move on.
- Manage the change. New business processes supported by new ways of thinking will represent change to the business and thereby incur resistance to change. An organizational change management (OCM) program will be required.
A cognitive strategy may begin with a stepping stone approach of infusing cognition into tightly bounded undertakings and then leveraging those successes in a way that ultimately leads to repeatable processes and pervasive business transformation. Plan to start small, iterate and stay agile in order to support the inevitable learning and adjustments that will occur along the journey.
IBM offers a low barrier entry program for cognitive adopters called the Watson Platform which includes a library of cognitive technologies and APIs.
The Point is This
The promise of cognitive systems is to apply natural language processing and machine learning in order to scale human expertise and improve decision making. These systems help people extend their knowledge and make complex decisions with high volumes of fast moving big data at the speed of thought.
Many industries are incurring accelerated change which is re-ordering the competitive landscape more frequently. The paradox of being data rich and information poor will accelerate this change.
Successful businesses will be defined by their ability to collect and curate the right data and apply analytics in order to make insights actionable at the point of decision, and make better decisions. It's a complex undertaking which is why those who succeed will achieve competitive advantage over those who don't.