How Pro Debater Harish Natarajan Beat A Machine, And It's Still A Big Win For Future Of AI

If there's one thing people are great at, it's arguing. We do it on a regular basis for the smallest of grievances.

So after millenia of arguing amongst ourselves, we've probably gotten really good. IBM wanted to test just how good, and whether computers could be better.

Images courtesy: IBM

The IBM Debater is the brainchild of the technology company. It's basically an AI-powered machine that's designed to emulate professional debaters. Except instead of reading articles, essays, and books at human speed, the IBM Debater can pull from a trove of about 10 billion published sentences.

The AI was pitted against Harish Natarajan last night, the grand finalist at the 2016 World Debating Championships. To be clear, the IBM Debater has faced two human opponents before, soundly defeating one and losing a close battle to the other. But Natarajan was a completely different beast, and ended up handily stealing the competition.

Also Read: Google AI Taught Itself Chess In 4 Hours, Came Up With Moves Never-Before-Seen In Chess History

The two were debating the subject of whether the US government should subsidize preschools. The actual debate itself didn't matter, so much as the results of it. These contests are scored based on how many audience members change their minds. Before it began, 79 percent of attendees were in favor of preschool subsidies. But after the debate concluded, with Natarajan arguing against subsidies, that figure had dropped 17 percentage points to 62 percent, giving him the win.

So yes, it wasn't as awe-inspiring as IBM's Deep Blue beating the world chess champion in 1997, or Google's AlphaGo defeating the top Go players 20 years later. However, that by no means makes this any less monumental a feat.

Humans have a natural advantage in a debate, being able to express emotion and therefore appeal to people on an emotional level instead of just logical. Not to mention, Natarajan has a BA in Politics, Philosophy and Economics from Oxford University, and is currently a Masters Student at the University of Cambridge.

Also Read: AI Will Soon Be So Good At Hacking, We¡¯ll Only Be Able To Stop Them With Other AI

And yet, even though the AI lost, it proved a point. The IBM Debater was able to make people listen to it not as a computer, but to actually evaluate its "opinion". It was able to create rational arguments, break them down into points, and obviously pull data from studies to back it up.

At one point, when talking about preschool subsidies, the computer even chastised humans for not being human enough. "Giving opportunities to the less fortunate should be a moral obligation for any human being," the IBM Debater said.

You may not be too surprised by an AI being able to formulate sentences, after all you probably talk to Siri, Alexa, and the Google Assistant all day. But consider that IBM Debater wasn't just responding to human queries, it was monologuing for minutes on end.

And to be honest, she sounded like a natural.