CONCEPT
The Five Tribes of Machine Learning
Pedro Domingos’s organizing framework for the field—symbolists, connectionists, evolutionaries, Bayesians, and analogizers—five tribes each right about something and wrong about everything else, five clues to one master algorithm that no single tribe has yet assembled.
The five tribes of machine learning are Pedro Domingos’s framework for making sense of a field whose apparent diversity of methods conceals a deeper question about the nature of learning itself. The symbolists, descended from the logic-and-rules tradition, believe intelligence is symbol manipulation and that learning is inverse deduction. The connectionists, inspired by the brain, believe learning is what
neural networks do when they adjust weights in response to error; their master algorithm is backpropagation, and their moment is now. The evolutionaries treat learning as natural selection, breeding solutions rather than specifying them. The Bayesians hold that all knowledge is uncertain and that learning is the disciplined updating of beliefs in light of evidence through Bayes’ theorem. The analogizers hold that we learn by recognizing similarity—that the central question of learning is how to measure it. Each tribe, Domingos argues, has discovered a genuine principle of learning, and each has hit a genuine wall. The walls, crucially,