Exploring the robo-witch: Is machine learning magic? July 11, 2017 - Blogs on Text Analytics
Throughout history, we have been interested in seeing the future, either in the form of checking the arrival time of the prince on a white horse or predicting stock prices to get rich! What caused this curse?! It’s probably part of human nature but we can also trace fortune-telling back to the advent of astrology, tarot, and numerology. I knew it, physicists! If we are looking for educated villains, they are the culprits!!! Now, getting back to the modern era, we know that there are nice tools and methods out there that allow us to be able to make accurate predictions… Yes, that’s correct, machine learning is one of them! AHA, we’ve found physicists’ accomplices in the 21st century, computer geeks!
Although it seems as if machine learning techniques can be easily applied to different problems, the question is, can we use them to solve any and all problems? Well, I am not an unfair dude, let’s assume we provide the algorithm with megatons of data to peruse! Okay, we have a computer buddy, a nice, fancy algorithm, and a lot of data to crunch. It sounds as if we are all set to discover the meaning of life. Let’s just feed in the data, step back, and watch our robo-witch fit an equation to the data… Ooops, we forgot that machine learning can only solve “solvable problems”! So, the algorithm isn’t useful if your problem is not solvable with the data that you have. Or if the algorithm isn’t designed to properly process the data. Not convinced yet? OK, let’s assume you want to predict the next recession, and your data set is based on the number of pockets in everyone’s pants and shorts! I am pretty confident that if you build a model it won’t work because there is no relation between pockets and recession. Don’t be too hard on me, it’s just an example! Hopefully, there is no relation even though we are told that these days everything is related to everything else!!
Machine learning is more than five degrees of separation. Machine learning is all about math, there is no magic involved. I know it might be a tough thing to swallow but you can’t cut people in half and put them back together again nor make the Statue of Liberty disappear. If a human expert can’t find a solution for a given problem, not necessarily an optimal one, a computer can’t either. But if you have a solvable problem and suspect some hidden relations might be part of the solution, then machine learning will search for patterns in data and find you a wonderful solution faster than you can. Alright, enough for this post… I’m late for work, where is my magic carpet?!?