Cognitive computing will be the starting point for business applications in the future, consultant Judith Hurwitz...
said, but companies probably won't go out and buy full-scale systems. Instead, they'll be available as a set of cloud services.
She envisions the technology as being part of the services-based IT model organizations are moving toward.
Cognitive computing systems -- or cognitive systems, as they're sometimes known -- approximate human brain function. And through collaboration between human beings and machines, Hurwitz said, they can be used to analyze and correlate huge amounts of data and build applications based on that data, rather than on business logic, or programming, like traditional apps.
More than an algorithm
A cognitive system is not just a machine-learning algorithm, though training it to see patterns in data and understand it in context is a big part of it, Hurwitz said. These are huge, complex systems comprising many parts, all resting on the foundation of a public cloud, private cloud or on-premises infrastructure. Layered on top of that are internal and external data sources -- unstructured data, such as text, video and images, and structured data, such as database records -- and data access and management services.
Judith HurwitzIT consultant
Another key component of cognitive systems is ontologies. These are databases of knowledge on specific topics the systems churn through and learn from. And at the very top are visualization services on which new applications can be built.
Big research universities may build such systems from scratch to do experimenting, but a marketplace of ontologies and pretested sets of data geared toward specific industries will most likely serve the majority of organizations looking to build data-based applications.
Once the tools are at their disposal, organizations can get started, but they "don't start with the world; they select a domain, and usually they start with a specific problem that they're grappling with," Hurwitz said.
'Lifecycle of knowledge'
For many business problems, a type of machine learning called supervised learning will work, she said. The algorithm is taught to detect or match patterns in specific types of data -- for example, the number of items sold on particular days of the year -- and can be used to make predictions about, say, what effect a certain marketing campaign will have.
There are other models. Reinforcement-learning algorithms develop strategies based on performance feedback, such as the results from playing a game. Unsupervised learning looks for correlations hidden in large swaths of data; it works when an organization doesn't know exactly what it's looking for.
Then comes a lot of trial and error: forming a hypothesis, identifying the right data sources, feeding the system the data and seeing what happens.
"You operate it, you see how it works -- and you start all over again," Hurwitz said. "This is really a lifecycle of knowledge that we're going to get to."
The future, technically
The field is just beginning to burgeon, with IBM's Watson being used for healthcare research and prepped for future use in fields such as telecommunications and financial services. Google and Amazon have their own cognitive computing approaches, and more tech companies are sure to follow. The result of these efforts, Hurwitz said, will be a whole new way of using technology.
"This is really what digital disruption is all about," said Hurwitz, referring to the change ushered into businesses and industries by digital technologies. "It's not about creating faster websites or being able to automate a process. It's really about transforming the way we think about data and the way we think about logic."
Consultant Judith Hurwitz discusses how cognitive systems build applications in part one of this two-part report.
Cognitive computing not just a thought experiment
IBM Watson aims for prime time
World of change on view at Cloud Expo