The Master Algorithm

The Philosophy of the Algorithm

Artificial intelligence, given a set of rules, can infer and deduce high-probability conclusions. But it’s a closed system; info in, info out –there’s no evolution. Machine learning, like AI, is an algorithm, but its capable of learning other algorithms. It needs only one set of rules capable of pattern recognition, applying empirical observations to scientific models, formulating them into theories, and adapting theories upon new knowledge discovery. Machine learning needs only an initial all-encompassing algorithm and then can be left to consume other algorithms, adopting those that prove robust, discarding the rest. Pedro Domingos, in the Master Algorithm, describes statistical algorithms as predators, while machine learners are super predators. Domingos’ ecosystem also includes herbivores like databases, crawlers, and indexes, which, without, the predators could not exist.

The importance of machine learning applies not only to technology but also science. Scientific advances require “data analysis commensurate with the phenomenon studied.” Most scientific fields use outdated statistical models like linear regression, where data is forced to fit a straight line. But we live in nonlinear world. Domingos’ breadth of knowledge is near Faustian. He references philosophers: Descartes, Hume, Kant; writers: Hemingway, Tolstoy, Borges; and modern thinkers: Nassim Taleb, Noam Chomsky, and Daniel Kahneman. He adeptly relays their contributions, weaving in and out of literary, philosophical, and socio-anthropological evidence and adds his own original insight.