This is an article from the archives. Links and some facts and findings may be outdated.

Judea Pearl's work on artificial intelligence impacts modern world

Although there’s evidence that the idea of artificial intelligence (AI) goes back centuries to antiquity, with myths of artificial beings bestowed with intelligence or consciousness, the actual research field of AI began at a conference on the campus of Dartmouth College in the summer of 1956.
Many who attended that conference predicted that a machine as intelligent as a human being would exist in a generation. Millions of dollars were invested to make this dream a reality. Despite considerable challenges — it’s taken more than a generation to create such a machine — progress in the last couple decades has been astounding.
040224 Pearl JOne scientist being recognized recently for his pioneering work in making that vision a reality and opening the doors to a new era for AI is Judea Pearl, professor of computer science at the UCLA Henry Samueli School of Engineering and Applied Science.
For his achievements, the Association for Computing Machinery, the world’s largest educational and scientific computing society, recently named Pearl the winner of the 2011 A.M. Turing Award, often considered computer science’s equivalent to the Nobel Prize. Pearl developed the theoretical foundations for reasoning under uncertain conditions as well as the graphical methods and symbolic calculus that enable machines to reason about actions and observations and to assess cause-effect relationships.
Although it took years for his concepts to propagate because they went against traditional approaches, the number of applications that incorporate his work today are ubiquitous.
His approach to reasoning has helped machines solve complex problems in many different areas: Given certain symptoms, what is the most likely medical diagnosis? If a person exhibits a particular pattern of behavior, how likely is she or he to be a security risk? If someone possesses a certain set of genotypes and phenotype, how likely is it that he or she is a carrier of a certain disease?
Pearl’s work is the foundation for Google searches, credit card fraud detection systems and automated speech recognition systems. Even IBM’s Watson, the computer that triumphed over some of the best players on "Jeopardy", owes its superior reasoning power to Pearl’s concepts.
"I was gratified to see that the community at large has found my work to be both useful and impactful," said Pearl after winning the highest honor in computer science. "As a result, I came to believe in it, too, but this time from the perspective of an outside observer, rather than a passionate researcher. Most importantly, it gave me hope that the extra credibility garnered by this recognition would help me enlighten the few remaining ‘islands’ of skepticism about my theories."
While skeptics may remain, eminent researchers in the field have no doubts about how Pearl’s work transformed AI fundamentally.
"Before Pearl, most AI systems reasoned with Boolean logic – they understood true or false, but had a hard time with ‘maybe,’" said Alfred Spector, vice president of research and special initiatives at Google. "This meant that early AI systems tended to have more success in domains where things are black and white, like chess, for example,"
One of the central challenges for researchers in the field of AI is how to get computers to represent knowledge and reason about it.
The working assumption in the early years of AI was that knowledge should be represented using logic, and that reasoning should be done using logical deduction, said Adnan Darwiche, a professor of computer science at UCLA Engineering. For almost two decades, this working assumption amounted to representing, and reasoning about, what’s true and what’s false. Then computer scientists realized that logic was not sufficient for this purpose.
"When we talk about human reasoning — commonsense reasoning in particular — that simply doesn’t work," said Darwiche, also an expert on probabilistic and logical reasoning. "Humans make assumptions, which are not necessarily true. They also make decisions even when they are not completely sure about the truth of underlying assumptions. This is simply beyond classical logic as AI researchers later discovered.
"In the 80s, there was a crisis in the field resulting from a debate between two different approaches," Darwiche explained. "One was to extend classical logic so it can handle these limitations; the other was to use probability theory to represent knowledge and model how people think. The latter approach was the one advocated by Professor Pearl’s work, and this approach currently dominates the world."
Pearl promoted the idea of letting computers calculate probable outcomes and answers. But to drastically alter thinking that prevailed for more than 20 years was a huge hurdle to overcome. Even though probability had been around for a very long time, people lacked the tools to represent multi-variable probabilities reliably and parsimoniously.
"In a sense, there was a battle on the philosophical and cognitive side to show that probability theory was the appropriate foundation for modeling human reasoning," Darwiche said. There was also another battle on the pragmatic and practical side "to show that one can develop efficient computer algorithms for manipulating probabilities."
Pearl and his wife, Ruth, before a portrait of their son, Daniel, a reporter for the Wall Street Journal who was killed 10 years ago by terrorists in Pakistan.
The impact of Pearl’s work extended well beyond computer science.
Pearl’s work was not just about representing and calculating probabilities efficiently — largely a concern for computer scientists. But it also led to greater insights into human intelligence, the realm of philosophers and cognitive scientists.
His "Baysian Network," a term he coined in 1985 named for 18thcentury English mathematician Thomas Bayes, established a dialogue between humans and machines. The network is a general and flexible modeling tool that mimics the neural activities of the human brain and constantly exchanges messages. Prior to this, the AI community had ignored the problem of uncertainty. This development was a critical step toward achieving human-level AI that can interact with the physical world.
"Pearl’s work on reasoning with uncertainty as well as his game-changing contributions to machine reasoning about causality have had a pervasive influence not only on machine learning but on natural language processing, computer vision, robotics, computational biology, econometrics, cognitive science and statistics," said Vinton Cerf, vice president and chief Internet evangelist for Google, who is also a UCLA Engineering alumnus.
"The applications are everywhere, because uncertainty is everywhere," Pearl told the New York Times recently.
In addition to its impact on probabilistic reasoning, the Bayesian Network has changed the way causality (cause-effect) is handled by the empirical sciences, which are based on combining experiments and observations. Pearl’s contributions to causal reasoning have had a major impact on the way causality is understood and measured in many scientific disciplines, including philosophy, psychology, statistics, econometrics, epidemiology and social science.
"Judea Pearl has made an important contribution by applying the tools and methods of econometric modeling in a wider setting and thereby expanding the circle of scholars familiar with the proper tools of causal inference," said James Heckman, professor of economics at the University of Chicago and 2000 Nobel Laureate in Economics. "He is an honest scholar in his frequent mention of his debt to econometrics."
For the future, Pearl said he hopes "artificial intelligence will evolve in a way that will enable us to understand ourselves better, because the ability to emulate is crucial for the understanding. When we learn to build robots that struggle with emotions and questions of responsibility and morality, we will be able to understand the architecture of those attributes in ourselves."
The A.M. Turing Award, which carries a monetary prize of $250,000 will be presented to Pearl on June 16 in San Francisco.
Media Contact