IBM has once again showed the world that machines can do everything that humans can, sometimes even better.
Computer systems have already triumphed over their fleshy counterparts at chess, Jeopardy, as well as the complex game of Go. Now, there's a computer that can defeat humans in a far more nuanced competition: debate.
IBM Debuts Project Debater
On Monday, June 18, IBM Research introduced the world to Project Debater, an AI supercomputer that can hold a substantive debate with a human with a sense of humor, little tact, and occasionally powerful arguments.
Project Debater, which IBM has been training since 2012, took on two Israeli debating champions: Dan Zafrir and Noa Ovadia. The debaters, both human and computer, were not aware of the topics in advance. Each side started off with a four-minute introductory speech, a four-minute rebuttal, and finally each competitor was given two minutes two make a closing argument.
The AI system participated in debates about whether governments should subsidize space exploration and whether telemedicine should play a bigger role in health care.
Debater Can Argue, Rebut, And Debate With Humans
Debater demonstrated its ability to listen to the opponent's argument and then try to undermine it, just like humans would. There were times when the computer even attempted to guess what the opponents would say and deliver a preemptive argument against it.
Here's how it works. Debater can pull up information to support its position from its data bank, which has "several hundred million articles." It also has a framework of pre-built arguments and even jokes that it seeks opportunities to deploy.
The AI also dropped the odd joke when arguing in favor of telemedicine.
"I can't say it makes my blood boil because I have no blood," the computer quipped.
How Did The AI Perform?
The Debater's performance was evaluated by audience members that included journalists and IBM staffers. The audience members were asked their opinion on the subject before and after the debate and which participant made the most compelling arguments, so IBM could measure how Project Debater fared against its human opponents.
In the first debate, the audience voted Ovadia as the winner in "delivering" the argument, but Debater scored more points when it came to "knowledge enrichment."
In the second debate against Zafrir, IBM's AI computer was the clear winner. Not only did it convince nine audience members' to change their stance but also outscored Zafrir in both argument delivery and knowledge enrichment.
