IBM computers famously won at “Jeopardy!” and beat world-class chess masters. Now, they’re taking on human debaters.
At a media gathering here Monday afternoon, a black, artificial intelligence-infused IBM computer with a screen for a face more than held its own debating seasoned human debaters.
In one debate faceoff, IBM’s “Project Debater” AI computer made the case in favor of the government subsidizing space exploration against Israeli debate champion Noa Ovadia, who took the opposite position.
Ovadia was judged the winner by the crowd of journalists in “delivering” the argument – the computer’s attempts at humor didn’t measure up to the personality of a human – but IBM outscored Ovadia handily on the question of “knowledge enrichment.”
IBM’s computer fared better in a second debate persuading the crowd that telemedicine is worth pursuing against another human debater, Dan Zafrir. Again, the human prevailed on delivery, but this time only by a slim margin, and the computer was a big winner in knowledge enrichment. And at least nine audience members’ minds on the topic changed to the point of view of the computer.
The debaters, both human and computer, were not made aware of the topics in advance. Each side had four minutes to make an opening statement, followed by a four-minute rebuttal and a two-minute closing summary. The computer went first each time.
The San Francisco event was the first time anyone outside of the company was able to witness a live IBM debate between a human and its AI system. But IBM researchers have been conducting debates in the lab for awhile, on such topics as “Should income taxes exist?”, “Will autonomous cars help safety?” and “Should antibiotics be used in our food supply?”
Through the IBM Cloud, the computer scanned billions of sentences to generate a coherent and persuasive position on the various topics. The machine then listens to its opponent’s speech and generates what IBM claims is a spontaneous compelling rebuttal, exhibiting a type of argumentation that until recently was simply out of reach for the machines.
“We believe that mastering language is a fundamental frontier that AI has to cross,” IBM Research Director Arvind Krishna says. “There’s aspects like speech recognition, speech to text that AI already does and does quite well. But that is not the same as listening comprehension or constructing a speech that can either be spoken or written or understanding the nuances of claims, meaning what supports a proposition or what may be against a proposition.”
Tech’s biggest companies – IBM, Google, Apple, Microsoft and Facebook – are among those engaged in a high stakes race for AI supremacy.
But the ability for a computer to not only persuasively compete in a debate against a live person, but to actually win the argument, is only likely to feed into fears expressed by Tesla and SpaceX CEO Elon Musk and the late cosmologist Stephen Hawking that artificial intelligence could spell doom for human civilization.