The Need to Understand Machine LearningBy
AI entered an age of implementation in 2020, and although progress slowed a bit, it remains as inevitable as it was before the pandemic. Computer scientist and AI researcher Pedro Domingos explains why: “The Industrial Revolution automated manual work, and the Information Revolution did the same for mental work, but [AI’s] machine learning automates automation itself.”
Domingos, a professor of computer science at the University of Washington, wrote The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World in 2015. He begins the book with a strong case for the need to understand how computers learn. “To use a technology, we don’t need to master its inner workings, but we do need to have a good conceptual model of it.” And he uses his own metaphor to dismiss the “impenetrable black box” excuse: “The algorithms we drive when we use Google, Facebook, or the latest analytics suite are a bit like a black limo with tinted windows that mysteriously shows up at our door one night: Should we get in? Where will it take us? It’s time to get in the driver’s seat.”
AN OPAQUE SUBJECT
That’s all well and good, but for those without an advanced degree in computer or information science, where do you even begin? The key is in the already mentioned “good conceptual model.” You don’t need a detailed programmer’s guide to algorithms or the mathematical bases for the general theories involved. In fact, Domingos promises at the beginning of his book that “All of the important ideas in machine learning can be expressed math-free.” And he keeps that promise. He also doesn’t pack chapters with coding. He starts with simple principles and then proceeds to explain how some machines can learn and even write their own programs, comparing that process to how we learn. There might even be machines that develop into “universal learners” by applying something he calls the master algorithm. He notes that some machine learners learn knowledge, and some learn skills.
With machine learning (ML) as the primary focus of the book, Domingos clarifies the distinction between AI and the learning functions. “Machine learning is sometimes confused with artificial intelligence. Technically, machine learning is a subfield of AI, but it’s grown so large and successful that it now eclipses its proud parent. The goal of AI is to teach computers to do what humans currently do better, and learning is arguably the most important of those things: without it, no computer can keep up with a human for long; with it, the rest follows.”
Even though Domingos avoids the formulas and coding, the book’s content does require some work on the part of the reader. When you get to the heart of the book, you encounter Domingos’s five tribes of ML, the rival schools of thought within ML:
- Symbolists assert that all intelligence can be reduced to manipulating symbols.
- Connectionists believe that learning is what the brain does, and we need to reverse engineer those connections.
- Evolutionaries argue that the mother of all learning is biological (natural selection).
- Bayesians claim that all learned knowledge is uncertain, with a solution found in probabilistic inference (Bayes’ Theorem).
- Analogizers suggest that the key to learning is recognizing similarities between situations.
Curiously, though each of the learning theories is different and biased, they all contribute to ML in practical ways.
A MASTER ALGORITHM
Adding to the uniqueness of the book is its narrative center—the quest for a master algorithm. This algorithmic solution would allow “All knowledge—past, present, and future—[to] be derived from data by a single, universal learning algorithm.” It doesn’t exist yet, but the possibility of such an algorithm, Domingos writes, “would be one of the greatest scientific achievements of all time.”
Domingos offers arguments from six disciplines that support the possibility of developing this kind of tool. He begins by reminding us of the astonishing breadth of ML applications and the fact that the same algorithms are doing so many different things. What he looks forward to is an algorithm that could be universally applied, bringing this kind of learning together in the way physicists have pursued a unifying theory to join all their efforts.
The book is an adventure and will turn on quite a few lights within the interior of the dark limousine. And for those using AI, Domingos reminds us, “Before we can discover the deep truths with machine learning, we have to discover deep truths about machine learning.”