Machine learning systems open up numerous business opportunities -- and obstacles -- for CIOs, according to Ed Featherston, VP and principal architect at cloud computing consulting firm Cloud Technology Partners. Featherston spoke with SearchCIO at the recent Cloud Expo in New York. In this video, he explains how machine learning systems can help CIOs identify patterns in big data sets and delineates how the rail industry is saving money on maintenance costs by putting machine learning algorithms to use. He also explains how to identify what projects are right for machine learning enhancements and enumerates some of the challenges that CIOs should expect when implementing machine learning for big data analytics.
What opportunities do machine learning systems offer CIOs?
Ed Featherston: It's an interesting new world for CIOs because they've always had lots of data, have tried to analyze it in different ways and identify the patterns that are in there. What machine learning does is help them identify patterns that they may not have seen or found before and find potential new business opportunities or new ways to change things in the business as they go forward.
A great example I like to use is the rail industry. Maintenance on trains is very costly. If they have to take a train out of service, it costs them a huge amount of money not only for the repairs, but also because of service level agreements and fines that they have to pay because it's out of service. But by taking the data that they're pulling out of the sensors on all of these rail trains now, and funneling it through machine learning algorithms that look at it and analyze it, they can predict ahead of time that train XYZ's part BCD is going to fail in three weeks and that it should be taken and repaired. They can schedule it without impacting the overall systems.
It's a capability they've never had before. It's saving them a fortune in maintenance costs by being able to predict that ahead of time. This is not something an average person would have been able to do by looking at all of the data and figuring out what patterns actually tell them that, "Oh, this part is about to fail."
How can CIOs identify the right problems that might be a good candidate for machine learning capabilities?
Featherston: They have to look at it from the business perspective first. What are the biggest challenges? What's hurting their business the most or slowing them down or preventing them? And then say, 'Okay, if I was able to anticipate those problems by having enough data that could be fed in, that somebody could analyze and figure out the problems, would it provide me business value on the other side?' Like everything else, it's still coming from 'figure out the business problems.'
If somebody could figure out what that information was telling you, and that would help solve that problem, then that's probably a good candidate to be looking at from a machine learning perspective. The other thing is do they have the right data to look at? Machine learning without data is useless. It's sort of like a high-performance car with no gas in the engine.
What challenges should CIOs expect when leveraging machine learning systems?
Featherston: There are multiple challenges in it. First off is identifying the machine learning algorithms or capabilities that might help solve the problem, and there are various vendors out there that are offering that piece of it. IBM Watson is probably the most famous one. What makes machine learning important isn't the algorithm, it's the data that feeds that algorithm. CIOs used to have to actually hire a machine learning, AI specialist to come up with an algorithm that will help them crunch the data and give them the answer they wanted. That's not necessarily the case anymore. What they still do need are the data people and the data scientists that can look at the data sets that they have and say, 'What data do I have available? What data is going to be useful to help me achieve that business goal I'm looking for, that I'm going to feed into the engine?'
Another part of the challenge that they also find is that they might not get the results they're expecting. The machine learning engine may tell them something completely different and completely new that they didn't expect out of the data. They can't go in with a preconceived notion of, 'I'm going to go put this data in and it's going to tell me A, B, C.' If you already knew it was going to tell you A, B, C, you don't need to spend all that money on machine learning and all that data. It's important to be prepared for the fact that they might find out new and interesting things that could open up other opportunities.