AI and machine learning are forcing CIOs to rethink IT strategies
Machine learning and AI are becoming necessary in today’s complicated IT environment, causing CIOs to figure out how to use them so IT pros and business both benefit.
By Zeus Kerravala
Machine learning and artificial intelligence (AI) are changing the world around us faster than ever before. We’re closer to having cars that drive themselves, natural language processing and computers that play chess with Grand Masters. As widespread as AI and machine learning have been, they have yet to impact corporate IT in any significant way.
Recently, the IT service management firm ServiceNow conducted a survey of over 500 CIOs in 11 countries and 25 industries about the state of AI in corporate IT — The Global CIO Point of View (pdf). To gain an understanding of the data collected and what it means, I recently talked with Dave Wright, chief innovation officer of ServiceNow.
How has the role of the CIO changed?
Dave Wright: The new CIO agenda is markedly different than it was just a few years ago. Yesterday’s CIO was tasked on maintaining the technical infrastructure at the company they work for. Today’s CIO is a partner with the company leaders and is tasked with finding ways of using technology to put the business in a position to lead their industry. This includes expanding the skills of the workforce, redesigning business processes and driving digital transformation efforts.
Many IT pros have looked at machine learning and AI-based automation as something negative and threatening — primarily because they believe they threaten their job. Do you believe that, and what can CIOs do to overcome this hurdle?
This is perhaps the biggest misconception. Machine learning and AI do not take jobs away; they augment the skills of IT professionals. The fact is that today’s environment is far more complicated than ever before, and it’s impossible for IT operations to manage the all of the different moving parts. Machine learning can be the IT pro’s best friend; they just need to realize how it can be used to make their jobs easier. The best way to accomplish this and to get buy-off from IT pros is to involve them in the design process. This way they have some sort of say as to how it gets deployed and what it’s used for.
It’s important for everyone involved in machine learning to understand that humans understand process flow much better than machines because they eat, live and breathe it every day. Humans should be superior at digitizing that process and then letting the machines take it from there.
The survey shows that 89 percent of respondents said they are using machine learning somewhere in their organization. That seems high to me. Did it surprise you?
We didn’t think the number would be that high, and it did surprise us. However, a deeper look at the data shows of that, only 3 percent are using machine learning across the company and another 20 percent are using it in some areas of the business. Another 26 percent are piloting machine learning and the vast majority, 40 percent, are in the research and planning phase.
This makes sense because the use of machine learning will be a “crawl-walk-run” for most organizations, as they will apply it in phases. The first phase will be using it to describe something. It analyzes the data and helps interpret it. The next phase is more cognitive where the AI can start to solve problems. The third phase will see the technology start to predict things. For example, it could perhaps predict that a security breach is going to occur based on other data.
The last phase, and we are years away from this, is prescriptive where the AI is able to predict things and then take action to remediate the action. In the previous example, it could not only predict a breach, but it could then take the necessary steps to ensure it doesn’t happen. For this to occur, the AI would use itself in an iterative manner.
The data shows that cybersecurity has the highest use of complete automation today at 24 percent. But it also has by far the largest expected increase by 2020 — when 70 percent of the decisions will be fully automated. Were you expecting this?
This completely makes sense. When you look at how automation was done in the past, it was a decision tree of hard coded “if this, then that” type of rules. Now, because of complexity, we are getting away from hard-coded things, and the rules need to be rewritten on the fly. This is particularly true for security where the stakes are incredibly high. Machines can process data sets and rewrite rules much faster than people. Think of it this way: The bad guys are all using machine learning to create malware, so it makes sense to use machine learning to combat it.
A whopping 47 percent of respondents said lack of skills is a barrier to adopting machine learning. This certainly seems like good news for engineers looking for new opportunities. What are the skills most lacking?
It certainly is good news, as there is no lack of opportunity for IT pros. The areas that are most lacking are data scientists and machine learning specialists. Most of the jobs that are open didn’t even exist a couple of years ago. I think a significant challenge for the industry is that it’s hard to find places to get trained in these areas. But I’ve seen more schools and training institutions offering courses, so that gap should close quickly.
Any final advice for the readers?
A couple of key points. The first is to clean up the sources of data because bad data leads to bad inferences. Most businesses have massive amounts of data, much of which could be classified as bad data — that is data with erroneous or poor information. The data fed into the machine learning systems must be great quality for the best decisions.
Also, companies need to re-evaluate the key performance indicators they track. For example, if predictive tools are used to solve failures, then the right metric is mean time between failure instead of mean time to repair. With security, the measurement should be incidents avoided instead of the length of time it takes to find a breach. Machine learning changes things, and the way success is measured needs to change as well.