We’re always talking here about artificial intelligence (AI) and machine learning (ML). Every now and then, it’s good to circle back to underlying concepts at the core of our powerful industrial applications like SparkCognition Oil and Gas Maintenance Advisor, Deep NLP, Renewable Suite, and more. Today we’re giving you a pop quiz. What is machine learning, and how does it work? Don’t worry if you don’t have a perfect answer. We’re going to try to break it down for you.
Traditional programming vs. machine learning
When humans learn something, it’s often happening in a classroom or by reading a book, but of course, we also learn through daily experience, e.g., discovering why you shouldn’t touch a hot stovetop (the hard way). Keep in mind—the same goes for computers.
In the early days of computing (and still today), humans wrote intricate programs for their computers that provided explicit instructions for responding to particular scenarios. This traditional programming approach required complex, hard-wired logic trees that endeavored to foresee every possible case the program might encounter. It was the only way to do it, like teaching computers how to work by teaching them step by step from kindergarten to college, so to speak. Eventually, given data and a program, the computer could reliably return an output, like a good student can. But it took a long time and great effort, and scalability was limited. And when a circumstance took place that the programmers hadn’t anticipated, “does not compute” was often the result.
With the arrival of machine learning, computer scientists vastly expanded computing capabilities by enabling computers to learn from examples and their iterative experience of trying to interpret them—like trial and error—rather than through the constricted IF/THEN approach of traditional programming. This approach turned traditional programming on its head. In machine learning, given data and output (plus a pattern-recognizing algorithm), the computer returns a program. It’s as if the computers skipped school and went directly to work, figuring out what skills they need to acquire by comparing and contrasting skills demonstrated by humans. This shift is extremely important since it is responsible for the existence of systems that can now routinely understand the context and meaning of a visual image, interpret written text, or recommend intelligent actions in the physical world—things we take for granted today.
Just remember: Data plus output equals program
Ask someone to define machine learning for you, and you’re likely to get various answers, but the most common response will have something to do with teaching a computer to solve problems the way humans do. That’s attractive in its simplicity but somewhat less than complete. Given the importance of ML in driving so much of not only the business applications mentioned at the outset but also our everyday lives through things like chatbots, voice transcription, movie recommendations, and GPS navigation, it’s important that we agree on just what it is we’re talking about.
A good explanation comes from the folks at MIT, who offered this terse assessment in a recent paper: “Machine learning is the subfield of artificial intelligence that gives computers the ability to learn without explicitly being programmed.”
You can also remember the essential difference between traditional programming and machine learning:
- Traditional programming: Give your computer data and a program. Get an output.
- Machine learning: Give your computer data and output. Get a program.
So machine learning begins with data and output—the more of it, the better to learn. This data could be numbers, photos/videos, words of text, or numerous other types of information. In addition, the data can be structured (i.e., stored in spreadsheets or databases) or unstructured (e.g., free-form text documents like emails and social media posts or hand-written maintenance documents).
Let’s say you want to generate a program to match people’s faces to their names using machine learning. You’ll need data and output (pictures of faces plus the names that go along with the faces) and a machine learning algorithm that can sort out the patterns between the data and output and learn to induce, or guess, what the program would be that would be able to identify names to faces. It automatically generates your program, which is called a model. So the next time you have a set of faces, you can reliably conclude which names match which faces using that model. In traditional programming, building a program that could solve a problem like that would take enormous resources. In machine learning, you just need a training data set and the right algorithm, and you’re on your way to solving the problem. A machine learning practitioner works to choose the best model produced by the algorithm that solves the problem. Of course, in the early going, the model’s conclusions will often be relatively poor, but with the completion of enough iterations (typically thousands or even millions), and tweaking by data scientists, the outcomes become uncannily accurate, frequently even insightful.
Worth noting is the fact that we are not talking here about technology that marginally makes more efficient something humans could achieve on their own. Rather, the ability of ML to quickly and scalably reduce millions of data points to actionable recommendations and insights transcends anything humans could remotely accomplish, even if they had limitless time. This point was succinctly acknowledged by computer science professor Aleksander Madry, director of the MIT Center for Deployable Machine Learning: “Machine learning is changing, or will change, every industry, and leaders need to understand the basic principles, the potential, and the limitations.”
In summary, machine learning is a powerful set of artificial intelligence capabilities that is revolutionizing not only many aspects of professional and industrial life but our personal lives as well. A recent Deloitte survey found that 67% of companies are already using machine learning in some capacity and that 97% are planning to use it in the coming year. Every movie recommendation you get from Netflix, every product suggestion from Amazon, and every text message that gets transcribed on your cell phone happens because of machine learning (to name but a small fraction of the now-countless examples of ways this technology touches our everyday lives).
To learn more about how SparkCognition uses machine learning to create industry-leading applications and how ML can help your company reach the next level of operational and financial performance, contact us to set up an exploratory conversation or see a demo of our solutions.