Cognitive computing: ‘A new way to approach knowledge’
A Q&A with Judith Hurwitz, Technology Strategist, Thought Leader, and Author, Part 1
Published April 30, 2015
An emerging technology for decades, artificial intelligence is finally on the map as developers and enterprises begin to realize its potential and seek opportunities to stretch the boundaries of machine learning to drive modern-day business goals.
Judith Hurwitz, President and CEO of Hurwitz & Associates, a research and consulting firm focused on emerging technology including big data and cloud computing, anticipated this day would arrive and has spent years learning about cognitive computing and natural language processes.
“I’ve been able to understand what it takes for this most important emerging trend to go from a good idea or what people would think of as, ‘That would never happen. It’s too hard; we can’t do it,’ to … a lot of these things do happen, but it always takes longer than what people think,” she says.
In this two-part Q&A, Insights Magazine sat down with Hurwitz to learn more about this evolving technology. In this first part, Hurwitz discusses where cognitive computing is on its journey to real-world implementation, the importance of knowledge management, and she talks about some of the industries that have already started to realize the benefits of this system.
In Part 2, Hurwitz discusses the common misconceptions still made today, what businesses should know when evaluating how to approach cognitive computing, and the opportunities to unlock the value of big data.
Insights Magazine: How is cognitive computing evolving and where is it today in terms of implementation in business?
Judith Hurwitz: The idea of being able to take knowledge from lots of different sources, bring it together, and be able to really take advantage of large volumes of data and really understand it without a person having to be involved in every step in the process—that’s been something that computer scientists have been trying to figure out for decades. How can we have knowledge management? How can we take this huge amount of data, bring it all together, and make sense of it in a fast, repeatable way? That’s what people have been trying to do for a long time.
People have looked at different approaches to this, whether it was text analytics, whether it was artificial intelligence, machine learning, or database technologies. So there are a lot of technologies that people assumed would solve the problem, but what cognitive computing really represents is bringing together a lot of technological approaches and creating a new way of approaching knowledge—both knowledge aggregation and being able to learn from knowledge. That’s really what cognitive computing represents; it’s a converged platform that brings together a lot of elements.
For example, the original grand challenge in [IBM] Watson was to build a computer that competed with human champions and win the game of "Jeopardy!". IBM researchers explored how to take knowledge—just knowledge about the universe—without really knowing what’s being asked, and design a system that figures out nuance, figures out context, and then learns over time. It was a pretty big challenge. Watson won the game of "Jeopardy!" because the system was able to answer questions with speed and accuracy. At this early stage, unstructured data from Wikipedia was ingested into Watson, and then the system was trained and designed to learn how to figure out the context for data when questions were asked. It used some natural language processes; it used some artificial intelligence; it used some machine learning; it used a lot of these different pieces. It was more brute force because it wasn’t a solution, it was this grand challenge; it was an experiment to see what was possible.
What we’re seeing now is the transformation from an experiment to something that can be applied to real-world problems.
I think one of the most interesting aspects of a cognitive computing system is that it’s really designed from the data out. With a traditional computer or with software, you begin with logic: What business problem am I trying to solve? You then create some logic that says, ‘When X happens, look these up A, B, and C, follow these 12 steps, and then come up with a result.’ It works very well when you have a problem that is very specific and not going to change a lot. The reason we’ve gotten into a lot of problems with applications is that they change all the time—the logic changes, the business processes change, and then you need to re-code the logic. A cognitive system assumes that you’re going to start with the data. And by analyzing the data, understanding the data, and building a model based on the data, you are led to the logic in a sense. Cognitive computing is actually changing the model of computing—now you’re letting the data lead you to answers as opposed to the data being the last step in the process.
IM: What are the business implications to having better knowledge management?
Hurwitz: In many industries, the people that understand their area the best usually have many, many years of experience, and they’ve learned experientially. A physician with 30 years of experience in a specialized area of medicine may walk into a patient’s hospital room, ask a few questions, review medical tests, and quickly identify what’s wrong and how to solve the problem. But take a physician in training, who goes to see the same person with the same symptoms, and that person is maybe going to research what’s been published on the topic and see what other people have come up with for treatments. They may call some experts that they’ve met or that they know are premier in their field, and ask for their advice.
The less experienced physician doesn’t have the same level of intuitive knowledge as the seasoned professional. It takes years of training and experience to be able to take in new data, analyze that data, look for patterns in the data, and then draw conclusions based on having the right level of context.
In a sense, cognitive computing applies a similar style of learning to a machine. A cognitive system begins by acquiring the right data—and we call that ingestion—so it ingests the right data from the right sources. It then curates that data and makes sure that data is actually useful and meaningful, and then it creates what we call a ‘corpus,’ a sort of standing set of information that addresses a problem. One of the issues in medicine is that on any given day there could be a thousand new journal articles that come out that address new discoveries in an area of medical research and new treatments that have just been discovered to be effective. It has become impossible for most physicians to keep up with all the new research. With a cognitive system focused on medical diagnosis, a doctor has additional tools to help understand the information out there and make use of it.
Cognitive systems demand a lot of collaboration between humans and the machine. You don’t just ingest data into a cognitive system and expect to get the accurate answers to your questions. You actually have to take that data, analyze that data, and make sure that the way the machine is interpreting that data is accurate. So there’s a whole training process that goes on. In medicine, there are some cognitive applications that are being developed to train the next generation of physicians. I think that that’s going to be a really interesting area for a cognitive system because people learn in different ways. If you are teaching a specific subject, and you see a pattern of how somebody is answering questions or reacting or commenting, you may be able to go back, and change the way you are teaching that person based on their learning style.
IM: Are there other industries besides healthcare that can realize the benefits of cognitive computing?
Hurwitz: The travel industry is really ripe for this, and there are definitely some applications being built. If you think about the average traveler, there’s no problem if you really know what you’re looking for, if you know where you want to go or when you want to go. If, on the other hand, if you want more guidance, and you can say, ‘I can go any place that’s warm within a 6-month period, and these are my other criteria,’ over time you can do enough searching and enough analysis to get answers to all of those. Basically you narrow it down: ‘Okay, I’ll pick this place, now let me see what’s available.’
But think about how much more effective your travel search experience could be if you could name all of your criteria and then you let the system continue to ask you questions like, ‘Is this what you’re looking for?’, and if there’s enough data there about enough places and enough options, then the system can really begin to provide you the capabilities of the most experienced travel agent who knows you really well—which isn’t very common anymore. A cognitive travel application could help you to quickly find your perfect vacation.
Another area would be in what we call Smarter Cities. For example, you can use cognitive computing in traffic management. Sensors can be used to generate data about traffic patterns and weather. You can combine data from sensors with knowledge about important events that may impact traffic. A system like this would allow you to automatically change traffic signals to keep traffic moving when there is a big construction project or some need to reroute traffic because of a visit from an important official.
Retail is another great example. Retailers want to offer more personalized customer service. You want to really understand changing trends and use all the data that you have about customers and about their buying patterns so that you make your store is in sync with customer expectations; to be able to match the weather conditions, knowing that there is a big rain storm coming and that people will want to buy umbrellas so to make sure that those are in stock.
Security and security assistance are other areas. To be able to look at patterns—is someone trying 17 different ways to access my systems? If you can look for patterns of violators that are well known, you can see within seconds that somebody is trying to do something that is unlawful. Being able to take action before they’ve had a chance to figure out the seventh way to get into the system, because they’re going to keep trying until they get in, but if you can notice it before it becomes a problem using machine learning, you have a much better chance of avoiding catastrophe.
So it will be able to be applied to a lot of interesting areas over time, but we’re so new at that that I think as this matures and as we see more solutions what you’ll see is that companies looking at this area will say, ‘Oh, I see what I can do here,’ and they’ll start to create brand new solutions that nobody has even thought about yet.
Click here to read Part 2 of this interview with Judith Hurwitz.