dcsimg
 

Machine Learning: Exciting Technology, Exciting (Mostly) Applications

Tuesday Jul 31st 2018 by Carl Weinschenk

Many of the applications based on machine learning and the artificial intelligence platforms into which they fit can be as dull as more efficiently running the accounts payable department. Dull is in the eye of the beholder, however.

A machine teaching itself to do new things sounds exciting. Therefore, it figures that the applications that these more capable devices will carry out would be exciting, as well. Robots failing in their first attack on earth, figuring out what they did wrong, and coming back for another try, that sort of thing.

It's not necessarily so. In fact, many of the applications based on machine learning and the artificial intelligence platforms into which they fit can be as dull as more efficiently running the accounts payable department. Dull is in the eye of the beholder, however. What may be dull for a casual observer may be quite exciting for the CFO, CEO and other executives.

Modern machine learning can be understood as a spectrum in which statistics are on one end and computer science on the other, said Ken Sanford, Ph.D., the lead analytics architect at Dataiku. The advances being made today push the needle more toward the computer scientist. This makes these tools more practical and easier to implement.

Sanford described a Dataiku client that provides recommendation engines to about 4,000 B-to-B companies. Machine learning is playing a key role. "The sequence from when something is learned to the time it is used [now is] almost simultaneous," Sanford said. "On top of that, every week, these companies may get new information. Because of the speed of translation, they are able to update and deploy the model with minimal translation."

Though machine learning and AI are deeply related, it is possible for one to operate without the other. For instance, a machine learning application that scours emails for certain words or phrases associated with information theft could improve its performance over time as new data is assessed. All it would do, however, is find suspect words or phrases more effectively.

The AI's intelligence capabilities would take it further. For instance, a combined package could conclude that there is a 70 percent chance that emails containing suspect data are emanating from a specific machine or physical location. Though it is possible to distinguish between the jobs done by the machine learning and AI platforms, in most cases they are combined -- along with computer vision and natural language processing – in the same platform.

Better, Not Just Faster, Intelligence

The first step for machine learning and AI is to do things as well as they are done by humans, far more quickly and efficiently. The next step is to do these things qualitatively better. Humans can do just as good a job of searching for words and phrases in emails as a machine, but simply orders of magnitude more slowly. The next step for a machine learning platform is to do something like predict whether a fuzzy area on a chest image indicates cancer better than a human. "This is moving up the chain a little bit," said Kimberly Nevala, the director of Business Strategy for SAS.

Boring or not, there are many uses of machine learning today, Nevala said. She pointed to banking applications that use robotic process automation (RPA) to track automated fund transfers, track failed trades, and respond to routine data requests. In the retail sector, AI can track customer sentiment and identify emerging areas of interest. AI and machine learning can be useful in understanding unstructured data such as emails.

Dr. Michael Wu, Ph.D., the chief AI Strategist for PROs, a company that uses AI and other tools to provide dynamic pricing to its clients, said that the four big early uses of AI are human/computer interfaces such as Siri, Alexa and Cortana; recommendation engines; autonomous vehicles and automation and business decisions.

It is important for organizations to prepare for AI and its machine learning component. The common sense steps include creating environments in which security is emphasized and data is as "clean" as possible. It is also important to train those who will use the systems on the specifics of their roles and clarify to them the idea that AI and machine learning can help them do their jobs more effectively, reduce rote drudgery, and set them up for more exciting, creative – and perhaps higher paying – employment.

"A mistake a lot of technologists make is that they immediately dive into tools and techniques without thinking of methodology, processes, and the organizational alignment required for that kind of machine learning or analytic transformation," said Vince Jeffs, the senior director of Product Strategy at Pega Systems.

Machine learning and AI in the contact center is a good illustration of how to handle the powerful new tools, Jeffs said. If the new system is foisted upon the agent without his or her training or buy-in, the results likely will be neutral or negative. At best, the technology will be accepted slowly. In a telecommunications company setting, machine learning can help predict if a subscriber is likely to churn. Doing this effectively can enable steps to address that subscriber's concerns. "We call that proactive retention," he said. "It is a move from reactive to proactive."

Machine Learning: A Big World Getting Bigger

Pega Systems offers adaptive churn reduction and text analytics tools, Jeffs said. The analytics tools search through emails – such as those sent to help desks – to assess the mood of the writer. The platform learns which products or services are being referred to, the writer's general sentiment, and what action is being contemplated. The system learns where best to route messages, perhaps based on the level of urgency. In the future, Jeffs said that systems will be able to detect mood and sentiment in still images and videos.

The world of AI is big today and will grow significantly over the next few years. Machine learning is central to this vision.

"We've really taken statistics and blended it with computer science," said Dataiku's Sanford. "The math has been around for a very long time. The basic fundamentals of the algorithms have been around for a very long time. Computers have gotten a lot better, storage a lot better, and [platforms on which to use it] have gotten a lot better."

Carl Weinschenk covers telecom for IT Business Edge. He writes about wireless technology, disaster recovery/business continuity, cellular services, the Internet of Things, machine-to-machine communications and other emerging technologies and platforms. He also covers net neutrality and related regulatory issues. Weinschenk has written about the phone companies, cable operators and related companies for decades and is senior editor of Broadband Technology Report. He can be reached at cweinsch@optonline.net and via twitter at @DailyMusicBrk.

 

Home
Mobile Site | Full Site