Critique
(in about 642 words)
How Machines Learn
No matter how people feel about them, algorithms are all around us. From setting the price of your train tickets to deciding what videos you would like to watch, these little bots are an inseparable part of our modern lives. But how do they actually work? How did we achieve learning using programming? This video by YouTuber CGP Grey aims to explain these mysterious bots that govern our lives in a way that even people with no background in computer science can understand.
Contrary to traditional algorithms, machine learning algorithms are not explicitly programmed to solve a particular problem. Instead, they are simply given thousands of examples of the problem (e.g. recognizing traffic lights) called training data, and improve upon themselves through training. A machine learning model usually starts with an arbitrary model that simply makes random decisions. Then, through millions of iterations of training and making small improvements, the model can achieve abilities that even surpass humans.
CGP Grey demonstrated the process of training a machine learning model by dividing it into 3 parts: the student bot, i.e. our prediction model; the building bot, i.e. the optimizer; and the teacher (testing) bot. He essentially humanized a machine learning model into different characters. Then, through the interactions of those characters, he explained the process of making a machine learning model in a way that everyone can understand.
In his video, he also addressed some key issues in modern machine learning, one of which is the black box problem. We humans are able to understand traditional algorithms, however complex they are. This is by design -- after all, how do people design an algorithm if they don’t first understand it? This is a little different in the world of machine learning. As we mentioned earlier, a machine learning model does not have prior knowledge on the problem it’s trying to solve. We just know that given inputs in the correct form, a machine learning model will give us an often accurate output. However, as these models became more complex, we find it almost impossible to understand how they came to that decision. Just like our own brain, “While an individual neuron can be understood, and clusters of neurons’ general purpose vaguely grasped, the whole is beyond.” One of the most significant implications for this problem is that the decisions made by a machine are no longer appealable. Got rejected for a credit card? Sorry, we don’t know how our computers came to that decision as well, better luck next time! I really think that this problem deserves more attention from the general public, and CGP Grey’s video does exactly that.
The animation style also deserves special attention. Rather than using accurate visualizations of machine learning models, CGP grey chose to only use stick figures and cartoon characters to illustrate a complex process. Not once in the video did we see a mathematical formula or a line of code. While this approach left out the technical details on machine learning, it also makes the video easy to understand. CGP Grey is very aware of the fact that his goal is to educate the general public on machine learning, not to give a lecture on it.
All things considered, CGP Grey’s How Machines Learn is a masterpiece that is not only informative but also fun to watch. When algorithmic bots inevitably govern more of our daily lives, the only thing we could do is to become more educated on how they work and their implications.
