AI Learns To Debate On Controversial Topics | Project Debater

  • IBM’s Project Debater performed live in public debates with humans and it was really impressive. 
  • In a debate, it actually convinced 20% people to change their minds in favor of it.
  • The AI works on unscripted reasoning and can argue both sides of an issue. 

Can artificial intelligence (AI) expand a human mind? Soon, it will. IBM is working on a technology, named Project Debater, that can engage in complex debates and provide impartial, evidence-based decisions.

This technology goes beyond than just a keyword search approaches. It can work with human to discover new points of view (both positive and negative) and even ask questions.

On 18th June, 2018, Project Debater, for the first time, performed live in public debates with humans. A champion debater and AI system argued about topics that have no right or wrong answers. Both sides prepared arguments for and against the topic, and then delivered a 4 minute opening statement, 4-minute rebuttal, and a 2-minute summary.

What Happened In Debate?

They were given a statement: “We should subsidize space exploration”. The AI presented an opening argument supporting the statement with facts and statistics. It stuck to a point that space research could advance scientific discoveries and inspire people to think beyond themselves.

The human debater, on the other hand, opposed this statement, arguing that the government subsidies could be used for good causes and researches here on Earth. Project Debater then countered the human’s argument with the point that economic benefits and technical advances from space research can outweigh other spending made by the government.

After listening to closing summaries of both sides, majority of the audience agreed that the AI enriched their understanding (about the topic) better than its human counterpart.

They also conducted a 2nd debate on the statement: “We should increase the use of telemedicine”. This time, Project Debater actually convinced 20% people to change their minds in favor of it.

Source: IBM

How It Works?

The Project Debater works on unscripted reasoning to persuade the audience on a complex topic. The system scans hundreds and millions of articles to develop its position on a topic.

It relies on 3 pioneering capabilities-

  1. Data-driven speech writing and delivery: to generate and deliver speech persuasively.
  2. Listening comprehension: to understand long spontaneous speech made by human and generate an effective rebuttal.
  3. Modeling human dilemmas in a unique knowledge graph: to form principal arguments.

To train the system, IBM has developed numerous datasets featuring argument detection, argument stance classification, debate speech analysis and text to speech systems.

The AI can argue both sides of an issue. It finds positive and negative points of the topic, and discovers which claims are for and which are against the given statement.

According to the IBM, Project Debater isn’t aimed to prove something right or wrong. It’s just an algorithm developed to help people see more than one side of an issue.

As of now, Project Debater uses a pre-defined set of rules (like in a board game) and therefore it often makes mistakes. Although the system is far from complete, it is capable of assisting with tons of complicated decisions. For instance, it can present pros and cons of public policies, or help employees to decide whether to support or oppose a financial thesis.

Read: Google AI Can Now Pick Particular Voice In A Crowd

The aim of the AI is to allow decision-makers to make more informed decisions. This is new territory which isn’t explored much; the more explainable and transparent we can develop this technology, the more we can trust it.

Written by
Varun Kumar

Varun Kumar is a professional science and technology journalist and a big fan of AI, machines, and space exploration. He received a Master's degree in computer science from Indraprastha University. To find out about his latest projects, feel free to directly email him at [email protected] 

View all articles
Leave a reply