By Tom Palmieri, Director of Operational Intelligence and Service Management
Machines -- like people -- learn best from great teachers
Machine learning is all about pattern recognition. Computers don’t learn like we do; it's more like "psuedo-learning." But what machines and people do have in common is the need for a capable teacher.
Determining patterns and exceptions is important for a wide variety of disciplines and businesses. Let’s say we collect a field of data over time that sets a baseline or expectation, which determines "the usual" behaviors. Then, by using machine learning techniques, computers can mathematically detect when a particular field of data is markedly different from what is expected. Computers are very good at using data driven analytics for determining that something is different from expected and at determining what is causing that change.
Must have a code that you can live by
Of course, learning techniques can be made more useful if there are more concepts of what is considered usual. For example, "usual" for a retailer around Christmas is different from "usual" during other times of the year, just as a Monday shopping experience is unlike one on a Sunday. Understanding that a usual Sunday brings in $10,000 in sales, for example, means that when a Sunday’s receipts significantly exceed (or shrink from) what is reasonably expected, the business can further research reasons for the change. Identifying those reasons, and then acting upon them, separates growing businesses from stagnant ones.
Businesses can optimize and discover statistical patterns which form the backbone of predictive analytics through the use of machine learning and artificial intelligence algorithms. Machine learning is based around the idea of providing machines access to data and letting them learn for themselves. Machine learning through AI, therefore, helps you find outliers or an abnormal combination of circumstances by analyzing multiple performance indicators that, taken in concert, seem unusual. In addition to its use in retail, you can see how it would also be useful in healthcare to provide differential diagnosis.
Many companies have been using predictive analytics, which takes statistical modeling and machine learning techniques to analyze past data and predict future outcomes. New to the fray is prescriptive analytics, which combines business rules, machine learning and computational modeling in order to recommend the next steps to reach a pre-specified goal.
Feed them on your dreams
It’s important to remember that humans teach the machines. Good information into the system generally results in reliable information coming out. An AI approach to find correlations is a great start, but the hallmark isn’t just constant repetition. The system needs to interact with a human for guidance. Data points carry certain weights and humans are responsible for judging the importance of particular data points. That’s how the AI learns, and is the idea behind neural networks. The computers learn from stimulus and recognize patterns, using the results to change actions and make decisions in a humanlike way.
A basic example of how this plays out is to think about email. If an email comes to your inbox and you send it to your spam folder, the AI learns that those types of emails may consistently be spam. That’s great, because it keeps your email box from becoming cluttered with unwanted communication. However, what if, during an email spring cleaning, you accidentally swept an email of interest into the spam folder? The AI would have learned something incorrectly. And that’s where trouble can start.
Machine learning is an attempt to make life more efficient. Ideally, AI will help us enjoy more success, be more responsive, spend less time on low-value work, make more money and feel less angst. IT has always been a human driven endeavor, working alongside typical laborers such as movers or typists. As we replace those types of jobs with automation, we make new jobs available and improve existing ones. Pessimists might look at this and focus on the loss of jobs, but if we can sharpen the tools needed by certain professions, the benefits become clear. For instance, you probably wouldn’t care about the jobs possibly impacted by the advancement of AI that allowed a loved one’s cancer to be caught three months early.
Context is complex. Adding and subtracting data makes a difference. And, at the end of the day, there may not be a sweet spot identified: Your mileage may vary. Teaching AI to intuit relationships is difficult. In the end, the quality of the teacher matters most.
Want to discuss how to incorporate Machine Learning in your organization?
Contact a Melillo Consultant Today!