We’ve already seen artificial intelligence reason its way through data to make savvy medical diagnoses. Now IBM’s Project Debater software has shown its ability to sort through information, to structure persuasive arguments and counter opposing ones.
In June, the software squared off with Dan Zafrir, Israel’s national debate champion, to test two topics. The first: should governments subsidize space exploration; the second taopic: should medical care providers make more use of telemedicine. The human and AI debaters delivered four-minute opening arguments, followed by a four-minute rebuttal and two-minute closing statements.
The audience, largely IBM staffers who might have had a bias for the company’s creation, rated Project Debater worse in the first round than its human opponent, in its delivery style (it made some stumbles and seemed to free-associate at times) but better in the amount of information it could marshal to support its position.
In the second debate, a majority of the audience reported that Project Debater was more effective than Mr. Zafrir in persuading them to agree with its position.
IBM has spent six years developing Project Debater in its Israeli R&D center, training it not only to sort through millions of bits of information on a topic, but also to organize its “thoughts” in a logical way that respects the fact that persuasion is often a subjective art.
IBM sees its new software as both an able assistant and a devil’s advocate. For example, it could listen to a lawyer’s structure of a legal case and then poke holes in it. It could also listen to a boardroom discussion about corporate strategy and support or question facts or logic used to argue a position. The company hopes that the software can lead to better decisions, based on factual evidence and logic.