Machine Learning: A Guide for the Perplexed
By Jason M. Pittman, D.Sc.
Machine learning: a far-off holy grail for science fiction authors, right? Wrong. Machine learning is already present, and being applied to a variety of contexts.
If you've used location services on your smartphone, you've experienced machine learning. If you've used Uber or Lyft, you've experienced machine learning. Lastly, if you've used a plagiarism checker (e.g., Turnitin) or received an invitation to financial fortune from a Nigerian cabinet minister, you've experienced machine learning.
In all these cases, machine learning is employed to optimize solution sets to practical issues. Broadly, we can describe the practical issues as shortest path to a destination, modifying pathing dynamically given environmental factors, and differentiating valid email from nefarious scams. There are many more machine learning examples and the rate is exponentially increasing.
Machine learning, at first glance, may seem like true machine intelligence. That is, learning depends upon the ability to learn and the scale of our ability to learn depends on intelligence. Thus, if something can learn it follows that it must be intelligent. After all, machine learning is a specific subfield in the broader field of artificial intelligence.
What's more, popular culture perceives learning machines as possessing a type of uncanny consciousness whereby these systems can create knowledge for themselves. Take AlphaGo for example -- the way in which AlphaGo learned appears to be hidden from us. This is a common perception of machine learning: that the actual learning is opaque despite being explicitly programmed.
If machine learning is becoming more pervasive (and it is), and the learning mechanics are truly opaque, we have a growing problem. Accordingly, my aims in this series are to (a) illustrate the simplistic beauty of machine learning and (b) dispel the mystery of where and how machine learning is implemented within our technological society.
While I want the conversation to be accessible, I also feel that some pseudocode examples will increase our understanding. It is inevitable that we'll dig into a little mathematics as well. Don't worry, though, you won't need advanced degrees in computer science or math to make your way through the content. The opposite is my intention; I want to bring you to a place from which you can (a) understand machine learning generally and perhaps (b) pursue more advanced study if you wish knowing you have a basic foundation already.
To set the stage properly, my contention is machine learning is not machine intelligence. I don't say such to discredit machine learning in some manner. Quite the contrary, I merely want to situate machine learning into an approachable content. Further, machine learning is not some esoteric craft. Rather, machine learning is merely the instantiation of statistics through algorithms, applied over datasets. Statistics, algorithms, and data(sets) are approachable, knowable components. There is a strong undercurrent of science in machine learning as well and thus we will spend some time exploring hypothesis formation and testing.
Tune in next time when we take the first step in our machine learning journey: decision trees!