Douglas Lenat, Who Tried to Make A.I. More Human, Dies at 72
Douglas Lenat, an artificial intelligence researcher who spent nearly 40 years trying to build common sense into computers, recreating human judgment one logical rule at a time, died on Thursday in Austin, Texas. He was 72.
His wife, Mary Shepherd, said the cause was bile duct cancer.
In the late 1970s, as a professor of computer science at Stanford University, Dr. Lenat developed an A.I. system he called Eurisko — a Greek word meaning “I discover.” It was designed to automate the discovery of new scientific concepts, methods and laws by analyzing data.
In 1981, he used this system to analyze the rules of an exceedingly complex role-playing game called Traveller Trillion Credit Squadron, in which players used a trillion-dollar budget to design and deploy a fleet of warships. Much like chess, Go and Jeopardy! in later years, the game was an ideal proving ground for the latest A.I. technology.
Each night, after combing through the many volumes of the Traveller rule book, Eurisko identified new ways of winning the game. Some were ridiculous — at one point, it suggested that the best way of winning was to change the rules — but others were promising.
Each morning, Dr. Lenat would adjust the system, pushing it away from the ridiculous and toward the practical. Guided by his common sense, Eurisko eventually found an unorthodox but powerful strategy. Rather than spending the trillion-dollar budget on large, mobile, well-protected warships — as other players did — it suggested building hundreds of tiny ships that barely moved and were not well protected but carried enormous firepower.
Over the July 4 weekend, Dr. Lenat entered a Traveller tournament in nearby San Mateo, Calif., competing with several hundred other players. Using Eurisko’s strategy, he won the tournament. The next year, the event’s organizers changed the rules so that the strategy would no longer work. But after working with Eurisko to discover a new approach, Dr. Lenat won the tournament again.
The experience inspired a new project that would consume him for the next four decades.
Running across dozens of computers, Eurisko could discover possibilities that Dr. Lenat — and other humans — had not. But it needed help from human judgment. Machines could not be truly intelligent, he realized, unless they, too, had common sense.
The project was called Cyc. He set out to define the fundamental but largely unspoken laws that outline how the world works, including everything from “you can’t be in two places at the same time” to “when drinking a cup of coffee, you hold the open end up.” He knew it could take decades — perhaps centuries — to complete the project. But he was determined to try.
In recent years, the Cyc project — and the rule-based approach to A.I. research it represented — has fallen out of favor among leading A.I. researchers. Rather than defining intelligence rule by rule, line of code by line of code, the giants of the tech industry are now focused on systems that learn skills by analyzing massive amounts of digital data. This is how they build popular chatbots like ChatGPT.
Many researchers now believe that this kind of sweeping data analysis will eventually reproduce common sense and reasoning. But as today’s computers struggle with even simple tasks and play fast and loose with the truth, others believe that the industry can learn from Dr. Lenat and his never-ending struggle to build common sense by hand.
“These chatbots think that when you hammer a nail into the wall, it should be vertical,” said Ken Forbus, a Northwestern University professor and A.I. researcher. “They can be very useful. But they don’t understand the world.”
Douglas Bruce Lenat was born on Sept. 13, 1950, in Philadelphia, a son of Nathan and Gertrude (Cohen) Lenat. When he was 5, he and his family moved to Wilmington, Del., where his father, a trained chemist, owned a bottling company called London Dry.
After his father’s death in 1963, he returned to the Greater Philadelphia area alongside his mother and older brother, Ronald. As a high schooler in Wyncote, Pa., his after-school job involved cleaning goose pens and rat cages. To find a better life, he learned to program computers.
At the University of Pennsylvania, he completed three degrees in four years — bachelor’s degrees in math and physics and a master’s in applied mathematics — before moving to the West Coast for his doctorate. He enrolled at Stanford to study artificial intelligence. His thesis committee included three of the researchers who had founded the field in the late 1950s.
It was a fallow period for artificial intelligence research — what was later called an A.I. Winter. But Dr. Lenat was among a new generation of researchers who revived interest in what had become a decades-long struggle to create machines that could mimic the brain.
In the early 1980s, several of the country’s leading tech companies helped create a corporation meant to keep the United States at the forefront of technological research: the Microelectronics and Computer Technology Corporation, or M.C.C. Led by Admiral Bobby Ray Inman — former director of naval intelligence, former director of the N.S.A. and former deputy director of the C.I.A. — the corporation hired Dr. Lenat as its chief scientist in 1984. From the company’s new headquarters in Austin, Texas, he began work on his common sense engine.
Two years later, he told Time magazine that the project would require 350 human years of work to even approach success. It was a vast underestimate.
In 1994, as another A.I winter arrived, Dr. Lenat spun the project out into a new company called Cycorp. Financed by various government organizations and private companies, he continued building his common sense engine until his death. He and his collaborators wound up spending more than 2,000 human years on the project, writing more than 25 million rules.
In addition to his wife and his brother, Dr. Lenat is survived by a daughter, Nicole Danielle Hermanson, from his first marriage, which ended in divorce; and two granddaughters.
An inveterate traveler, he visited more than 100 countries and all seven continents. After cremation, Ms. Shepherd said, she plans to arrange for some of his remains to be scattered on the moon.
In the fall, as ChatGPT captured the public imagination, Dr. Lenat and the cognitive scientist Gary Marcus began a new paper meant to show the new generation of researchers what they could learn from his nearly 40 years of work on Cyc. In the midst of the project he had a recurrence of cancer that had first appeared in 2021.
In July, Dr. Lenat urged Dr. Marcus to help him finish the paper. A shortened version was published a month before he died. “He took on the project that no one else had the guts to take on,” Dr. Marcus said. “He never completely succeeded. But he showed us at least part of the way.”