IBM AI Debater Vs Human Champ

When Ancient Greece philosophers encountered a divisive question they turned to debate for resolution. For centuries, this formalized back-and-forth of words and arguments has helped us explore the pros and cons of ideas to make more informed decisions. Debating is a hallmark of human civilization, and few do it better than World Debating Championship Finalist Harish Natarajan. Last night at Yerba Buena Center for the Arts in San Francisco, Natarajan stepped up against an AI-empowered debating machine from IBM.

“We should subsidise preschool” was the motion, with Project Debater arguing for, and Natarajan against. Audience members declared their position before the debate started, and the winner would be the debater that shifted the most people over to their side. The proceedings were overseen by professional moderator John Donvan.

Rules:

• Each side had only 15 minutes advance notice of the resolution to prepare arguments
• Both sides present a four-minute opening statement, a four-minute rebuttal, and a two-minute summary

Project Debater argued we should subsidise preschools because they benefit impoverished populations, create better students, and contribute to lowering crime rates. Speaking in a female voice, the bot impressed many observers with her suggestion that social welfare is a moral duty, an expression of “basic human decency.”

Natarajan’s counter — that history has shown money spent on subsidies could be better applied elsewhere — boosted his side’s support from 13 to 30 percent and he won the debate. There was a silver lining for the bot: in a post-debate poll, almost 60 percent of the audience said Project Debater had enriched their knowledge on the subject, with Natarajan’s corresponding rate only 20 percent.

The debate with Harish Natarajan was not Project Debater’s first test against humans. In June 2018 it took on Israeli National Debate Champion Noa Ovadia. That post-debate poll also showed the audience believed the bot had done a better job of enriching their topic information than the human.

The IBM Debater system was trained on 10 billion sentences in hundreds of millions of articles from various prominent newspapers and magazines. It doesn’t learn a specific topic in a human way, instead it builds narratives from massive data. This enables it to debate on many different issues, provided they are well-covered and explained in the corpus where the system was trained.

The debate challenge reflects IBM’s ambition to enable AI to interact with humans in increasingly complex discussions. In our world of information and misinformation overloaded, developments in AI reasoning and language organization can also bring new perspectives and give a voice to underrepresented opinions to initiate more meaningful discussions.

Journalist: Fangyu Cai | Editor: Michael Sarazen

2018 Fortune Global 500 Public Company AI Adaptivity Report is out!
Purchase a Kindle-formatted report on Amazon.
Apply for Insight Partner Program to get a complimentary full PDF report.

Follow us on Twitter @Synced_Global for daily AI news!

We know you don’t want to miss any stories. Subscribe to our popular Synced Global AI Weekly to get weekly AI updates.