53 Language Processing Lab of the Department of Computer Science is combining theoretical and computational linguistics, data science, and software engineering to encode visual and textual input in Hebrew and then ascertain its meaning. The lab delivers technology, models, and algorithms for enabling AI or data-science based practical applications, whether for shopping, studying, or turning on smart lights. The Hebrew NLP project will also deliver technology that is critical to the state. For example, as Israel continues to purchase advanced weaponry, Hebrew NLP will ensure that the IDF can use it seamlessly. Hebrew NLP can help Israel’s security services analyze the large volume of language data it collects, leading to better and faster identification of potential threats from within. And Hebrew NLP can be used to scan patients’ medical files, helping doctors identify areas of concern for predictive medicine and effective treatment. TAKINGAI TOTHE STREETS If we want safer, more functional, and more efficient societies, humans and robots will need to learn to get along. Fortunately, says artificial intelligence expert Prof. Sarit Kraus, robots are pretty quick studies. A researcher inBar-Ilan’sDepartment of Computer Science, considered the leadingdepartment in Israel for robotics and AI, Krausworks tooptimizemulti-agent systems, or systems whosemultipledecision-makingagents all shareacommon goal. Given that increasingnumbers of such systems include both computers andhumanbeings, ensuring that each can support the other is key to a successful outcome. “We can’t relinquish all control to software agents—at least not yet,” Kraus explains, in what is at once an affirmation of and sobering comment on humanity. “The ideal is for as few humans as possible to be involved in areas that can be managed by AI, but the reality is that there are situations in whichwestill needorwanthumans todecide. Socomputers need to know when to ask for help, and how best to help their human operators help them. As with most things in life,” she concludes, “the key is collaboration.” “Computers candoamuchbetter job thanwecan. Thequestion iswhether we’ll be smart enough to let them.” She gives the example of her work in the field of driverless cars, for which she received a major grant this year from the Israel Innovation Authority. By integrating machinelearning modelling into tele-operated vehicles, Kraus helps computers recognize when they’re in over their heads. “When an autonomous vehicle (AV) approaches a construction site, for example, there may be a new lane created by traffic cones that requires weaving in and out. It would be too dangerous for the AV to maneuver autonomously, and therefore in such complex scenarios, we still prefer for humans to have the final say. However, manually tele-driving the car can be exhausting and too time consuming to work.” Kraus’s algorithms prompt the AV to trigger a request for tele-assistance, or the involvement of a human operator in choosing between alternatives, drawing up a newcourse of action, or overridingastandardpolicy. Theoperator canthen guide thevehicle,whilenever actually takingover thewheel. “The truth is, we would all be better off if we let AVs rule the roads,” says Kraus, whopoints out that humans, despite theirmany attributes, are in fact terrible drivers. If, however, we can get to the point at which just one person is needed for every 1000cars—she says she’ll call that awin. “Incertain situations, fromdriving to repetitive tasks toevenhigh-level negotiations, computers can do amuch better job thanwe can. The question,” she finishes, “is whether we’ll be smart enough to let them.” Prof. Reut Tsarfaty
RkJQdWJsaXNoZXIy NDU2MA==