The Master Algorithm

✍️ Pedro Domingos

Tags: pedro-domingos , data-science , lang-en

With “The Master Algorithm” Pedro Domingos reveals himself as compelling storyteller trying to disseminate Machine Learning and Artificial Intelligence concepts to the general population. He is passionate and optimistic about the subject (sometimes a bit too much).

Domingos defines the Master Algorithm as a universal learning algorithm from which all knowledge—past, present, and future—can be derived from. The Master Algorithm is the Machine Learning equivalent to a grand unified model in Physics. Throughout the book chapters, Domingos visits ‘five tribes’ of Machine Learning what they believe are the master algorithm:

1. Symbolists, which has roots in Logic and philosophy that have inverse deduction as their master algorithm.
2. Connectionists try to reverse engineer the brain using backpropagation as tool.
3. Evolutionaries with genetic programming try to mimic the evolution (with mutations, crossovers and survival of the fitness).
4. The Bayesians see learning as a problem of probabilistic inference.
5. Analogizers learn by extrapolating from similarity judgments and are influenced by psychology

He uses a very challenging task, searching for the cure of cancer, to show the shortcomings of each of the five tribes “master algorithms” in isolation. Domingos then proposes how to leverage the strengths of each tribe as a starting point to the ultimate Master Algorithm.

Finally the book envisions how life would be with a Master Algorithm around. In this part I had mixed feelings. It seems a bit of a stretch to assume that people will get used to a new 100% digital and data-driven lifestyle as easily as Domingos proposes.

Caveat: Although this book is touted as a Machine Learning 101 guide, it is not a very easy read for a person who had not any previous exposition to the field.