Artificial intelligence has made remarkable progress in the last 10 years. So much so that, oddly enough, engineers often don’t know how the algorithms in some of these systems (the deep learning or deep learning). The BBVA Foundation Frontiers of Knowledge award In the category of Information and Communication Technologies, this Wednesday recognized the work of engineer Judea Pearl, one of the people who has contributed the most to the traceability of artificial intelligence systems.
East 85 year old American professor Born in Tel Aviv, world-renowned in his discipline, he developed a mathematical language “first for the concepts of probabilistic reasoning and later for inferring causal relationships through observed and ordered data”, the jury points out, thus providing “a modern basis for the artificial intelligence”.
One of the big questions facing scientists in the 1980s was how their models could deal with uncertainty. Pearl’s answer was statistics. In a world in which everything is interconnected, it is necessary to guide machines so that they do not start from scratch every time they face a problem, in the same way that we humans develop prejudices or a priori that guide us in our actions. If the ground is wet and we have a flat sole, it is likely, we automatically think, that we can slip if we are not careful. If behind a hedge we hear a roar, perhaps there is a tiger.
Pearl took this reasoning to the machines of the so-called Bayesian networks. “The variables are represented in a graph by nodes, so that only the arcs that actually have some relevance in probabilistic terms appear,” explains Pedro Larrañaga, professor of Artificial Intelligence at the Polytechnic University of Madrid (UPM). That is, you eliminate from the equation a whole set of situations that are considered very unlikely to happen (for example, that the roar comes from a snail).
“The system proposed by Pearl is capable of elaborating from data that structure of conditional interdependence between the nodes. That is to say, you stay with the essence of the problem”, continues Larrañaga. Pearl created a school of researchers in the 1980s that quickly spread throughout the world and that in the 1990s collaborated extensively with Microsoft, the most advanced company of the time. The model of it was taken over by science and integrated into the main research projects of the discipline.
Tear down the black boxes
Research related to artificial intelligence took a major turn in 2015. Since that year, neural networks, which use deep learning algorithms (deep learning), dominate the main advances produced in the discipline. These techniques do not consist of programming the computer with exactly what we want it to do, but showing it the rules of the game and letting the system train itself with a series of databases so that it can design its own strategies to meet certain objectives.
The problem with this approach is that nobody knows how the algorithms reason (if that term can be used): it is impossible to interpret how they have solved the problem, we only see that they have. “Compared to this is the line that considers that interpretability should be at the base of artificial intelligence, and that is where those of us who work with Pearl’s Bayesian networks place ourselves,” explains Pedro Larrañaga, professor of Artificial Intelligence at the Polytechnic University of Madrid (UPM).
Pearl’s methods are taught today in all computer science faculties and his books “have inspired momentous advances in the understanding of reasoning and thought,” the jury notes. Its “broad and deep impact” can be seen in many areas and applications, such as “in the development of unbiased and effective medical clinical trials, in psychology, robotics and biology,” she adds.
The jury for this category was chaired by Joos Vandewalle, honorary president of the Royal Flemish Academy of Sciences and Arts of Belgium and professor emeritus of the Department of Electrical Engineering at the Catholic University of Louvain (Belgium); and has had as secretary Ron Ho, director of Silicon Engineering in Meta (United States). The members have been Regina Barzilay, Distinguished Professor of Artificial Intelligence and Health at the School of Engineering of the Massachusetts Institute of Technology (United States); Georg Gottlob, Professor of Computer Science at the University of Oxford (United Kingdom) and at the Vienna University of Technology (Austria); Oussama Khatib, Professor of Computer Science and Director of the Robotics Laboratory at Stanford University (United States); Rudolf Kruse, Professor Emeritus at the Faculty of Computer Science at the Otto von Guericke University of Magdeburg (Germany); and Mario Piattini, Professor of Computer Languages and Systems at the University of Castilla-La Mancha (Spain).
You can follow THE COUNTRY TECHNOLOGY in Facebook Y Twitter or sign up here to receive our weekly newsletter.
#Frontiers #Knowledge #award #recognizes #development #transparent #artificial #intelligence #systems