IBM's artificial intelligence loses face to a human debate, but remains a persuasive technology

The debate was about whether the government should subsidize kindergartens. But the real question was whether a machine called IBM Debater could underestimate a prominent human debater.

The answer, Monday night, was no.

Harish Natarajan, the grand finalist of the 2016 World Debating Championships, has more influenced an audience of hundreds of people than his interlocutor IBM Debater. Humans, at least those with degrees from Oxford and Cambridge Universities, can still win against the subtleties of knowledge, persuasion and argumentation.

It was not a major title win as we saw when IBM's Deep Blue computers beat the best human chess player in 1997 or defeated the best human players in the old game of Go in 2017. But IBM has always shown that an artificial intelligence could be useful in situations where there is ambiguity and debate, not just an easy score to judge who won a game.

"What really has struck me is the potential value of IBM Debater when [combined] with a human being, "said Natarajan at the end of the debate.The IBM AI was able to tap into a mountain of information and offer a useful context to that knowledge, he said.

This was the second time that IBM Debater attacked humans in public, even though he had held dozens of debates behind the walls of Big Blue. In the first IBM Debater contest, artificial intelligence defeated a human debater while losing closer competition with another. This time, however, the human opponent was harder. Indeed, IBM researchers involved in this multi-year project were waiting for the loss of their AI.

Computer Persuasion

IBM Debater lost, but there is no doubt that he won in one way or another: by listening to him, you evaluate what he says, and not just saying it, it's a computer that says it. She marshaled her argument, broke it down into a few points and substantiated it with data from various studies. It was not perfect, but it was about.

And oddly for an AI, it told us how to behave homo sapiens.

"Giving opportunities to the less fortunate should be a moral obligation for every human being," said Debater.

During the debate, each party had 15 minutes to prepare – although only IBM Debater has the advantage of being able to rely on 10 billion phrases of press and research articles. academics. Each party presented their argument in turn, refuting the other, and then presenting a final argument.

At a time when Apple's Siri, Amazon's Alexa, and Google Assistant are listening to our questions and answering them in human voices, it's easy to forget how remarkable it is to be able to converse with computers. IBM Debater goes beyond, speaking for a few minutes.

"She was surprisingly charming and human," said John Donvan, host of the debate on Intelligence Squared Debates, which animates the debates and broadcasts them through a radio show.

Reading in progress:
Look at this:

The new IBM AI can debate you

2:00

Do not expect to run something like Project Debater on your laptop in the near future. It was running primarily on a powerful server with 28 processing cores and a gigantic memory of 768 GB, about 50 times that of a high-end laptop. It was supported by a quartet of servers, each with 64 GB of memory and two terabyte hard drives containing text.

Preschool Grants

IBM Debater argued for the idea of ​​subsidizing nursery schools and Natarajan objected.

According to Debater, kindergartens "bring benefits to society as a whole and it is our duty to support them". Good preschools mean that children – especially poor children – do better in life.

Natarajan countered that preschool subsidies were "nothing more than a political gift for the middle clbad … and not for the most disadvantaged people". He also highlighted Debater's badumptions, for example, that a grant would significantly improve the education of the poor.

Debater has shown improvements over his 2018 debate. A new tip lies in his ability to offer a parallel argument – in this case, subsidizing health care can be beneficial. Another was the improvement of the refutation skills. After Natarajan had explained that some children might not benefit from immersion in the potentially competitive world of preschool education at the age of 3 or 4 years old, IBM understood this point of view and challenged this view: "My opponent has claimed that preschools are harmful".

"We have been working very hard since June to improve the system," said Noam Slonim, senior Project Debater investigator at IBM Research. Debater's source material – academic publications and press articles – has also been enriched with data from another year until the end of 2018.

The most difficult contest so far

The competition was the most difficult to date for IBM's AI.

Natarajan "is at a different level from the debaters we've met up to now," said Ranit Aharonov, Debater Project Manager at IBM. "He is the most decorated debater in the history of university competitions debating with the world record of the number of victories."

The event, at the IBM Think conference in San Francisco, is the latest big debate of IBM Debater. "Debating is nice, and it's good to show, but we should focus on how to take this technology and make something that is commercially viable," said Aharonov.

"We are at the stage where we are going to finalize the first use case we are going to work on," said Aharonov. It could be something like helping a company understand the point of view of its employees or customers, or helping the media or governments engage in discussions with people on controversial issues, she said.

Indeed, the technology behind Project Debater is about the mess and shades of the real world we live in, not the black and white world of games.

"We are coming out of the AI ​​comfort zone to get into a grayer territory," said Slonim.

Facebook acquires an AI startup: GrokStyle is here to help you with your purchases.

CNET Magazine: Check out a sample of articles published in the CNET online edition.