IBM’s CES booth featured an experiment in crowdsourced debate

In June, I wrote about the fist public demo of IBM’s Project Debater, a piece of software–created in one of the company’s AI grand challenges–that can engage in competitive debates. This week at CES in Las Vegas, Project Debater was back in a new version with its own twists.

As originally demonstrated in San Francisco, Project Debater generated its arguments by trawling a vast corpus of existing text relating to the topics at hand–a technique that often worked surprisingly well, though it also led to some oddball moments which, IBM says, it embraces as part of the experience. At CES, the company let attendees (and online participants) contribute their own brief arguments–pro or con–on issues such as “social media brings more harm than good” and “gambling should be banned.” The software then rated these submissions for quality, attempted to eliminate redundancy, and stitched them together into arguments.

The June demo had Project Debater facing off against real live champion human debaters, and let audience members vote on who won; the CES version skips this competitive aspect. Instead, the software has been generating two brief arguments for each statement–one supporting it, and one contesting it.

You can see the debate topics, the human-generated arguments, and Project Debater’s speeches here (and can hear the speeches if you’re using Firefox). The results are less flashy than the debates I witnessed last year at IBM’s Watson West office, without the earlier attempts to make the digital debater ingratiating and even amusing. But maybe this more basic approach brings the technology closer to IBM’s long-term goal for the whole idea–which is to use AI to help businesses synthesize complex information and understand it better.