We’re in the Midst of an AI Revolution, says Shashua

“We are in the midst of a revolution” in the field of artificial intelligence. “Things are moving very, very fast.” If there’s one thing to take away from the recent webcast conversation between our CEO and one of Samsung’s top executives, let that be it.

Professor Amnon Shashua recently joined Young Sohn, the President and Chief Strategy Officer at Samsung Electronics, for the latest episode of the latter’s web series “The Next Wave.” Over the course of 21 minutes, the two discussed a broad range of subjects, from the challenges inherent in developing autonomous vehicles to the latest breakthroughs in AI and even a bit of advice for budding entrepreneurs eager to follow in Shashua’s footsteps.

As Prof. Shashua relates, our CTO Prof. Shai Shalev-Shwartz once observed that “There are two types of products. One family of products is very, very sophisticated from a technological point of view, but low accuracy” – like smartphones and personal computers, which (for all their capabilities) can afford to crash from time to time with little risk beyond annoyance and inconvenience. “The second family of products are not so much sophisticated but very accurate” – like airplanes, which are based on time-tested technology, but cannot afford to fail. “Autonomous driving is both,” noted Shashua: “very, very sophisticated but on the other hand very, very accurate. The tolerance for failure is almost zero. And this is where the conundrum, this is where the big challenge is.”

Successfully and reliably replacing the human driver with sensors and computers requires harnessing the latest advancements in artificial intelligence. Not the least of those developments are being undertaken at companies that Shashua has cofounded – including Mobileye, of course, but also OrCam (which develops computer-vision systems for the visually impaired), AI21Labs (which specializes in natural language processing), and Israel’s first-ever digital bank – on top of his academic career and awards received in the field.

“The kind of AI that we are used to is Narrow AI. Solve particular problems: play chess, play Go, have a perception system for autonomous driving, do decision-making for autonomous driving… these are all narrowly defined problems. The new frontier right now is language: natural language understanding.”

Whereas current methods of training AI require manually labeling elements for the system to recognize, said Shashua, “the way you train these language models is completely self-supervised. It means that you can build these monster networks that are trained on unlimited amount of data.” That, in turn, unlocks virtually unlimited potential – and it’s all advancing at a rapid pace: “If we simply continue this path, within this decade we will have Broad AI – which is something that two-three years ago I would not say.”

Watch the full interview in the video below.