Ten years ago, I began regularly ­visiting areas of the world where, at the time, practically no one had a cell phone. This year, returning to these same places, I saw that most adults—and nearly every teenager—had smartphones. The computing power in the smartphone that I hold in my hand today dwarfs the power of the computer that filled a room in my college days. Not only that, but I can carry on a conversation with my phone, asking it questions and even receiving verbal answers (well, sometimes).


In my college years, the concept of “artificial intelligence” (AI) was introduced—the idea that software could allow computers to interact and solve problems in the same way that humans can. We have since seen more and more examples of computers and software solving problems that once required a human being’s participation. IBM developed systems that could defeat human chess and Jeopardy champions. More recently, Google DeepMind developed a system called AlphaGo that defeated a human world champion Go player. Because the game of Go requires more intuition than playing chess, it’s a much more difficult problem to solve. Using neural networks and Monte Carlo simulation techniques, AlphaGo was able to train itself and improve its playing techniques by playing against itself millions of times.


Today, the term “cognitive computing” is often used to indicate systems such as Google DeepMind’s AlphaGo or IBM’s Watson that can reason and understand at a high order similar to human thinking. Generally speaking, cognitive computing is a subset of AI that uses multiple AI technologies to produce what we recognize as thinking.

Most cognitive computing architectures “learn” by establishing layers of interconnected neural networks and then running multiple Monte Carlo simulations, changing the relative weightings of the decision points in the networks as they determine the best route to the desired results. The AlphaGo system consists of two layers of neural networks: a layer that estimates the probability that a particular position on the board would produce a win and a network that proposes possible moves for the value network to evaluate. The interaction between the two layers then produces the final move to be executed.


Currently, cognitive computing technology is advancing much faster than previously anticipated. Before AlphaGo’s win over a world ­champion, many people expected that such technology was five or more years away. Beyond playing games of skill, cognitive computing systems are being applied to solving many difficult real-world problems. Medical journals publish 700,000 new articles each year. It’s impossible for a single person to keep up with all the advances in the field of medicine, but by using cognitive computing, the text from these articles and existing medical knowledge could potentially be combined with personalized patient history, genome sequencing, and available drug treatments to develop medical solutions tailored to a specific patient in a matter of minutes as compared to the weeks or months of research that this might take today.


Already there are many cognitive computing solutions in use today. In a banking organization, cognitive computing is being used to predict overdrafts a week in advance with 94% accuracy and four weeks ahead of time with 87% accuracy. In an insurance company setting, cognitive computing is being used to predict nonrenewals and cancellations in advance with 78% accuracy. In addition, it’s possible to predict financial and life events within the customer base that can be used to take actions to increase sales, conserve business, and serve customers better. Another property and casualty insurance company has created a virtual agent with the ability to converse with online customers and help them select coverages that insure their property while remaining within the customers’ budgetary ­constraints. The more data these cognitive computing systems process, the more accurate they become at predicting outcomes thanks to their built-in learning abilities.


There are several keys to cognitive computing success. First, both subject matter and cognitive computing expertise are required. Communicating the goals of the project and the potential impact on processes and employees also is important. Finally, making sure key data is accessible, accurate, and supported by a robust technical infrastructure will smooth the way for a successful implementation. Fortunately, there are many cloud-based solutions available that can provide access to cognitive computing resources at a relatively reasonable cost.

Just as the growth of mobile computing power has quickly transformed the developed and developing parts of the world, so cognitive computing has the potential to accelerate solutions to some of the world’s most difficult problems and find the answers. One day not only will you be able to ask questions of your phone, but you’ll even get the right answers.

About the Authors