Human Opponent Beat IBM's AI in Debate, AI Shows Promise
IBM’s latest AI competitor, the IBM Debater, lost its match against one of the highest ranking debaters in the world Monday night but showed off just how far AIs had developed in the process.
Humanity Triumphs For Now
The debate, which took place at the IBM Think conference in San Francisco, put IBM Debater up against Harish Natarajan, who holds the record for most University debate victories and was a 2016 World Debating Championships grand finalist. Project Debator began the evening by greeting Natarajan, "I have heard you hold the world record in debate competition wins against humans, but I suspect you have never debated a machine. Welcome to the future.”
This was IBM Debater’s second public debate. The first IBM Debater competition saw IBM Debater defeat one opponent fairly easily but lose a close contest with a second opponent. Going into Monday night’s competition, the IBM team behind IBM Debater believed defeat was certain.
“[Natarajan] is at a different level compared to the debaters we faced so far," said IBM's manager of Project Debater, Ranit Aharonov. "He's the most decorated debater in the history of university debate competitions with the world record in the number of victories."
The way a debate is scored relies on how many audience members changed their minds about the issue being debated. In this case, the topic was whether preschools should be subsidized by the government, which 79% of audience members agreed with at the start and which IBM Debater argued in favor of.
By the end of the debate, that number had fallen 17 points to 62%.
Sticking up for Preschool Subsidies
In many ways, the competition is a remarkable achievement for Natarajan, who had the unenviable task of arguing against a highly popular proposition against an artificial intelligence with over 10 billion lines of text from academic journals and news articles to rely on when making its case—as well as 28 processing cores, 768GB of RAM, and 4 assisting servers with 64GB of RAM and 2TB of text on their hard-drives.
Still, we cannot overlook just what IBM Debater was able to accomplish in defeat. It was able to take the data it had and construct moral arguments in favor of subsidizing preschools, arguing at one point that preschools "carry benefits for society as a whole. It is our duty to support them."
In an improvement over its earlier performance, IBM Debater’s rebuttal skills saw a marked improvement, grasping at one point that Natarajan was arguing that immersing some 3- and 4-year-olds into a potentially competitive environment of a preschool would be of no benefit to them.
IBM Debater was able to contextualize this argument and opened its rebuttal by saying that “my opponent argued that preschools are harmful.” Piecing together this statement may seem simple to us, but we have a lifetime of natural language training and brains specifically designed to parse the meaning of nuance like this. IBM Debater had to do it using mathematical operations using 1s and 0s, which is an astonishing feat.
IBM says, however, that this is Debater’s last big showcase debate.
“Debater is nice, and it's good to showcase, but we should be focusing on how to take that technology and make something that's commercially viable," Aharonov said. "We are at the stage where we'll finalize the first use case we'll work on.”