Introduction
Who teaches the algorithms choosing for you?
Every click helps a machine build its picture of you. Pedro Domingos opens up the learning machinery behind those judgments, then pursues a harder question: could one algorithm learn everything the others can?
The machine learning revolution
Software can grow from examples
Machine learning changes what programmers have to specify. Traditional software takes instructions and data and produces an answer. A learner takes examples, often paired with desired answers, and builds a model that can handle new cases. Recognizing handwriting becomes possible without an engineer spelling out every shape a handwritten letter can take.
Domingos compares this to farming: the learning algorithm is the seed, the data is the soil, and the learned program is the crop. Humans still choose what to plant and judge the results. Their work shifts away from constructing every detail by hand.
The robot scientist Adam makes this shift concrete. Given knowledge of yeast metabolism, it proposes hypotheses and carries out experiments to test them. For businesses, the same principle creates a feedback loop: more customers produce more data, which can improve predictions and attract more customers. That advantage depends on useful data and successful learning, rather than collection alone.