meta_pixel
Tapesearch Logo
Log in
Science Talk

AI Can Now Debate with Humans and Sometimes Convince Them, Too

Science Talk

Scientific American

Science

4.2 • 644 Ratings

🗓️ 17 March 2021

⏱️ 15 minutes

🧾️ Download transcript

Summary

Today on the Science Talk podcast, Noam Slonim of IBM Research speaks to Scientific American about an impressive feat of computer engineering: an AI-powered autonomous system that can engage in complex debate with humans over issues ranging from subsidizing preschool and the merit of space exploration to the pros and cons of genetic engineering.  In a new Nature paper, Slonim and his colleagues show that across 80 debate topics, Project Debater’s computational argument technology has performed very decently—with a human audience being the judge of that. “However, it is still somewhat inferior on average to the results obtained by expert human debaters,” Slonim says.  In a 2019 San Francisco showcase, the system went head-to-head with expert debater Harish Natarajan.  Beyond gaming, it’s rare to see humans and machines go against each other, let alone in an oratory competition. Not unlike its human counterpart, the AI was given only 15 minutes to research the topic and prepare for the debate—rifling through thousands of gigabytes of information at record speed to form an opening statement and layer counterarguments that were later delivered through a robotic female voice, in fragments and with near perfect diction.  It couldn’t best Natarajan in San Francisco, but in a different debate, the system—co-led by Slonim and fellow IBM researcher Ranit Aharonov—has managed to change the stance of nine people in a debate on the use of telemedicine, essentially swaying the debate to its side and rebutting the argument of its opponent. In other words, in this realm, humans still prevail. But how do you build the architecture for a complex system like this? Is the AI capable of recognizing meaning or larger contexts in a debate? Can a system descended from Project Debater one day intervene in real-life social media arguments to quell misinformation or stir a debate in one direction or another, for better or worse? We answer these questions and more in the podcast. Learn more about your ad choices. Visit megaphone.fm/adchoices

Transcript

Click on a timestamp to play from that location

0:00.0

Here's the truth about AI. AI is only as powerful as the platform it's built into.

0:05.7

ServiceNow puts AI to work for people across your business, removing friction and frustration

0:11.2

for your employees, supercharging productivity for your developers, providing intelligent tools

0:16.9

for your service agents to make customers happier, all built into a single platform you can

0:21.9

use right now. That's why the world works with ServiceNow. Visit ServiceNow.com

0:27.8

slash UK slash AI for people. This is a science talk podcast from Scientific American.

0:45.0

I'm Pakina Maimir.

0:47.2

It's as old as ancient Greece and a practice driven by passion as it is by facts and evidence.

0:53.0

It's also exclusively human. Animals cannot

0:56.1

debate. Machines cannot debate, right? Well, here's the thing. This truly is a first for us,

1:03.0

the first time that an artificial intelligence, namely Project debater, will be on our stage,

1:08.3

arguing with a human being and may the best debater win. And as we like to

1:13.3

say at every debate, may civil discourse win as well. That debate between man and machine happened in

1:19.4

2019 in San Francisco in front of a live audience. My guest today is Noam Sloanim, a distinguished engineer

1:26.6

at IBM Research Haifa, who along

1:29.0

with colleague Granite Ahrenov and others created the machine that was on that stage, one that

1:34.3

can debate a human without a script and occasionally win, one with a supercomputer for a brain.

1:42.3

Project Debater is a cloud-based AI system created by IBM, built with an

1:46.9

NLP or a natural language processing model, and trained using deep learning and machine

1:51.9

learning techniques. It took about seven years to develop, and since it has proven itself

1:57.1

a formidable opponent to champion debaters worldwide. Its model can scan over 400 million

2:03.4

newspaper articles and Wikipedia pages in the time it takes you to finish a cup of coffee. It has a

...

Please login to see the full transcript.

Disclaimer: The podcast and artwork embedded on this page are from Scientific American, and are the property of its owner and not affiliated with or endorsed by Tapesearch.

Generated transcripts are the property of Scientific American and are distributed freely under the Fair Use doctrine. Transcripts generated by Tapesearch are not guaranteed to be accurate.

Copyright © Tapesearch 2026.