Can You Beat AI in an Argument?

It wasn’t long ago that Google’s Duplex made headlines for an unsettling demonstration in which its AI tricked an innocent human into thinking that it was human as well.

Not to be one-upped, IBM has revealed an AI capable of arguing — and maybe even sometimes winning against humans.

Holding Its Own

Headquartered an hour north of New York City, IBM is tucked far away from the usual AI developer circles in Silicon Valley. But that doesn’t mean it isn’t still pushing the boundaries of cutting-edge technology. Its latest innovation, known as Project Debater, recently had a debate with professional debater Dan Zafrir.

In a demo of what’s known as “computational argumentation,” the AI found supporting points for its opinion by searching “several hundred million articles.” Similar to Google Duplex, what was even more impressive was also somewhat unsettling: the AI managed to sprinkle in a few passive-aggressive digs and jokes in its argument.

We know how disturbing this sounds, but keep in mind that this was a demo at the end of the day, just like Google’s Duplex demo. It’s worth noting that, after the polarizing reception, Google made it known that the successful demo only came after too many failed attempts to count.

Human Error

Even though IBM’s Project Debater was doing a demo, it still made some mistakes that even the best human debaters fall for. For instance, sometimes, it gave generalized comments that weren’t actually direct responses to any of the human debater’s argument points.

Jeff Welser, the VP for IBM research, explains: “If it’s less confident, it’ll do its best to make an argument that’ll be convincing as an argument even if it doesn’t exactly answer that point — which is exactly what a human does, too, sometimes.”

It’s important to understand that the AI wasn’t exactly composing arguments on the fly. Rather, Project Debater takes supporting sentences from documents, puts them together, then adds a bit of human touch (the jokes and sarcastic tones). This method still isn’t as refined as, say, what a professional human debater would compose. But it’s still impressive.

Besides, IBM didn’t set out to trick us into thinking its AI was human, anyway. The company’s main objective was to show that AI could perhaps help refine our own reasoning by inserting more objectivity through efficient evidence-based research and removing the biases and unnecessary emotions that humans inadvertently put in their perspective.

More Progress to Be Made

IBM’s demo is undoubtedly impressive in showing how far AI comes in combining logical thought and human speech to actually sound like a human. But achieving self-awareness and the ability to generally handle many tasks is a completely different game (and frankly, one that’s much more difficult to make a reality).

Yes, AI can beat us in ultra-specific tasks now, like our favorite games. Yes, it can even create a fake video that fools us into thinking it’s real. But the ability to learn from new situations and switch between a multitude of varying tasks is still far beyond its grasp.

What would you like to see AI tackle next? Let us know in the comments!