IBM’s new AI supercomputer can argue, rebut and debate humans (IBM, GOOG, GOOGL)

Dr. Ranit Aharonov, Dr. Noam Slonim, Project Debater, IBM Research, artificial intelligence

  • IBM gave reporters in San Francisco a demonstration of Project Debater. Think Deep Blue, but instead of playing chess, this system debates.
  • Computer scientists have labored for six years to enable an unprecedented system that culls millions of documents of information, listens and comprehends and then forms arguments based on that data. Debater even made jokes.
  • Debater fared well, winning one of the two debates against humans.Warts were visible, however. At times, Debater cited barely relevant data.
  • This is all part of IBM’s pursuit to create computers that master human speech.

IBM once again gave the world an impressive update on the competition between humans and machines.

The company known for building supercomputers that can defeat grand master chess players and champion Jeopardy contestants, hosted another Man vs. Machine contest in San Francisco on Monday. A system that IBM calls Project Debater faced off against two humans in two separate debates.

The verdict: Humans are still ahead, but the gap is closing.

Debater won one of the two debates as voted by the audience, but who won was almost beside the point. What mattered most is that this is the first artificial intelligence system to demonstrate the ability to argue. According to IBM, the technology represents a breakthrough in equipping computers with the ability to " truly understand language" and then be "expressive."

In the war of words, two experienced debaters, Noa Ovadia and Dan Zafrir, took on IBM’s AI. Each competitor opened with a four-minute speech. Then received another four minutes to rebut, and finally each was given two minutes to make a closing argument.

Some of the highlights included Debater’s ability to listen to an opponent’s argument and then attempt to undermine it, much as a human would. Sometimes, Debater guessed at what the opponents would say and launched a preemptive attack against it. Debater even made a couple of attempts at humor, one time acknowledging that its "blood would boil" if it had blood.

Noa Ovadia, Dan Zafrir, Project Debater, IBM, artificial intelligence

Some of the lowlights came when Debater drew upon barely relevant data to make a point. In the debate over the use of telemedicine, Debater cited a Supreme Court ruling in Iowa that was near nonsensical.

We’ve seen a lot of whiz-bang AI applications of late. This is one of the tech sector’s hottest areas and one that companies around the world are vigorously pursuing. The promise is that AI can help us access and analyze big amounts of data faster and more efficiently than humans. For the companies who stake out AI turf, there’s a chance to make big bucks.

Skeptics fear that AI could snatch jobs away from humans, or at the extreme, some day lead humans to a robotic apocalypse.

No doubt, IBM isn’t afraid to continue the march toward more advanced AI. For half a century, the company has  showcased new AI tech by pitting it in head-to-head competitions against humans. The debate Monday hearkened back to February 10, 1996, when IBM’s Deep Blue supercomputer made worldwide headlines by defeating Garry Kasparov, the Russian grandmaster, in a single game of chess.

Kasparov won the match that day, but the next year, Deep Blue would best him in a full match. At this stage of Debater’s evolution, IBM isn’t as concerned about winning debates.

“We’re not super obsessed about who wins and loses,” said Dario Gil, vice president of Artificial Intelligence and Quantum Computing at IBM Research. “We’re more interested in seeing AI deal with the human ambiguity, reality, and context. We want to see if it can write well, be persuasive, and give good evidence.”

The audience was asked to identify what their opinions on the two debate subjects (We should subsidize space exploration and we should increase the use of telemedicine)  prior to the start of the debate. Later, they were asked which one of the debaters made the most compelling arguments, the human or the computer. 

To train a computer to debate, IBM said three major hurdles had to be overcome. The system had to gather huge amounts of text, analyze the information, and from that, retrieve relevant and convincing information.

Debater must listen to, and comprehend the opponent’s argument to identify concepts and claims and identify weaknesses.

Finally, the system must recognize human controversy and dilemma so it could “suggest principled arguments.”

Dr. Noam Slonim, Dr. Ranit Aharonov, IBM, artificial intelligence, Project debater

To do all this, IBM’s team of scientists, led by Dr. Noam Slonim and Dr. Ranit Aharonov, developed “a methodology and tooling framework” that they called Cicero, which improves the algorithms that control natural-language processing.

This is all supposed to one day enhance language-based applications. That’s good because for all the promise of AI, all the tech available to consumers so far– especially Siri, Google Assistant and Alexa–none of it seems anywhere near ready to have but the most rudimentary conversations.

Sometimes these digital valets don’t understand basic commands. We seem a long way from the day we can rely on our computers to help us determine if we should vote for one candidate or another, the right and wrong of having an abortion or the pros and cons of joining the military.

But that’s the kind of assistance that IBM appears well on its way to some day giving us.

Join the conversation about this story »

NOW WATCH: How Apple can fix HomePod and Siri

from SAI