Judea Pearl

Overview

Judea Pearl is a computer scientist and statistician who turned reasoning under uncertainty into a formal, computable subject. He invented Bayesian networks — a mathematical formalism for defining complex probability models — together with the principal algorithms for reasoning with them^[1]. He later moved on to causation, setting out a framework for building causal models and analyzing causal relationships from data^[1]. In 2011 he received the ACM Turing Award^[1].

Education and early research

Pearl was born on September 4, 1936, in Tel Aviv, then under the British Mandate for Palestine, and grew up in Bnei Brak^[1]. He studied electrical engineering at the Technion, receiving a B.S. in 1960, and continued to a Ph.D. in electrical engineering at the Polytechnic Institute of Brooklyn, completed in 1965^[1]. His dissertation was titled "Vortex Theory of Superconductive Memories"^[1] — a study of memory devices that connected his doctoral training to the hardware work he would do in industry over the following years.

Industry and UCLA

From 1961 to 1965 Pearl was a Member of Technical Staff at RCA Research Laboratories in Princeton, New Jersey^[1]. From 1966 to 1969 he was Director of Advanced Memory Devices at Electronic Memories, Inc. in Hawthorne, California^[1]. In 1969 he joined UCLA as Assistant Professor of Engineering Systems, and became Professor of Computer Science in 1976^[1]. He founded and directed the Cognitive Systems Laboratory there from 1978, and became Professor of Statistics in 1996^[1].

Bayesian networks and belief propagation

A Bayesian network represents a set of variables as nodes of a directed graph: an arrow from one variable to another says that the second depends directly on the first, and a compact list of conditional probabilities for each variable given its parents then specifies the entire joint distribution. Pearl invented this formalism and the main algorithms for computing with it^[1].

His paper "Reverend Bayes on Inference Engines: A Distributed Hierarchical Approach" introduced belief propagation: an exact, distributed, asynchronous, linear-time inference algorithm for trees^[1]. In a tree-shaped network, inference can be carried out by having neighboring nodes exchange messages until the beliefs settle, so the computation is spread across the network rather than done in one central place. The algorithm later became the basis for turbocodes^[1].

Search, probabilistic reasoning, and causality

His 1984 book *Heuristics: Intelligent Search Strategies for Computer Problem Solving* covered A* and other search algorithms, along with the idea of deriving admissible heuristics from relaxed problem definitions^[1]. A* explores possible solution paths while using a heuristic to estimate the remaining cost; if that estimate never overstates the true cost, the search remains guaranteed to find an optimal solution. Relaxing a problem — dropping some of its constraints — is one systematic way to obtain such estimates.

His 1988 book *Probabilistic Reasoning in Intelligent Systems* sparked a revolution in the field of artificial intelligence^[1], and by 2012 about 50,000 publications had Bayesian networks as their main subject^[1].

His 2000 book *Causality* presented do-calculus, a complete mathematical framework for building causal models and analyzing causal relationships from data^[1]. The framework separates seeing from doing: noticing that a barometer reading has dropped is not the same as reaching in and setting the instrument, and only the second kind of intervention tells us what causes what.

Recognition

Pearl received the IJCAI Research Excellence Award in 1999, the Lakatos Award of the London School of Economics in 2001, and the ACM Allen Newell Award in 2003^[1]. He received the ACM Turing Award in 2011, and announced that he would donate a major portion of the Turing Prize money to the Daniel Pearl Foundation^[1].