Clients Use Cognitive Computing to Solve Business Challenges
Illustration by Kai and Sunny
For example, when asking a voice-activated digital assistant to locate the nearest bookstore, you’re tapping into back-end systems powered by deep learning. Such systems have dramatically improved our ability to interface with computers.
Deep learning also addresses the challenge of image recognition and categorization. “It’s relatively easy to train a computer to visually recognize a square,” Soutter says. “But if you want to describe what the hundreds or thousands of different expressions the human face can make, from all angles, it’s a programmatic nightmare.” Deep-learning systems have begun to do this with increasing accuracy. “This is probably one of the most popular uses right now because the problem of image recognition has been so hard,” he adds.
In addition, deep learning is driving development of driverless cars. That work is far too complicated to program by writing code. Instead, vehicle developers are teaching systems the rules of safely driving on roads, Soutter says.
A case in point: IBM partner NVIDIA used video footage of a person driving 3,000 miles to develop a driverless car. The footage showed what the driver saw and how he reacted to different items, such as road signs and other vehicles, in a 3-D space. The team also recorded usage data from the car’s instruments, including the gas and brake pedals. Then, without writing a line of code, NVIDIA’s team poured this data into a deep-learning training server.
The computer created an algorithm of how to be a car in the real world, Soutter notes. “It came up with a model that didn’t run on that high-performance computer but ran on a small GPU package installed within the automobile.”
Solving Complex Problems
While few organizations develop driverless vehicles, many interact with customers worldwide and need a simpler way for both customers and partners to interact with their IT systems. International businesses often communicate in numerous languages, spurring the drive for natural language processing and real-time translation.
“These are very complex problems if you’re trying to solve them programmatically,” Soutter says. “It’s practically impossible to write in code, ‘How do I describe how to interpret English, Mandarin and Brazilian Portuguese, and understand what the user is saying?’ But deep-learning technologies help us get an accurate natural-language processing engine for an accurate real-time translation.”
Other companies incorporate cognitive computing in industry-specific ways. Financial services firms use deep learning-generated algorithms to perform risk analytics to uncover fraudulent transactions. Meanwhile, consumer-facing companies use machine-learning technologies to interpret customer patterns and behaviors.
One example: When viewers watch a TV series using a media streaming service, the streaming company can recommend another program they might like. That’s because it has incorporated into its customer-facing algorithm a deep-learning capability that seeks out patterns in the behaviors and interactions of its viewers. This kind of machine-driven insight helps businesses get close to their customers.
Machine learning is also being used to develop inspection tools for large-scale energy distribution systems. Petroleum refineries and electrical utilities maintain miles of pipelines or high-tension wires. Preventing line failure is crucial, but it’s an onerous and expensive task if done by crew members.