IBM Project Debater FAQ

What it is, how it works, and why it matters.

1. What is the purpose of Project Debater?


Project Debater is an AI technology developed by IBM Research that aims to expand human minds through impartial debate. IBM researchers are exploring the boundaries of AI by teaching computers to create engaging and better-informed points of view. The goal is to build a system that helps people make evidence-based decisions when the answers aren’t black-and-white. In development since 2012, Project Debater is IBM’s next big milestone for AI, following previous breakthroughs like Deep Blue (1996/1997) and Watson on Jeopardy! (2011).

2. How is this different from an AI personal assistant?


The AI assistants we’re used to have become very good at certain tasks like searching for photos of loved ones, shuffling a playlist and dimming the lights. AI is also helping enterprises deliver better customer service and conduct deeper analysis. Project Debater is exploring different territory by using AI to engage in long-form discussion and provide impartial arguments on various topics that have no right or wrong answers.

3. Why teach a machine how to debate?


Culturally, the origins of debate lie not in conflict and competition, but in democracy and discussion. Debate enriches decision making, helping people weigh the pros and cons of new ideas and philosophies. It lies at the core of civilized society. We debate not only to convince others of our own opinions, but also to understand and learn from each other’s views. In the future, we believe machines will be able to help humans with many important decisions we make daily.

4. My attention span is short. Can I have the abbreviated version?


Project Debater goes beyond search-based discovery technologies to AI technology that can work with humans to discover, reason and present new points of view.

5. What is the format of the debate?


It follows the format of a traditional debate. Project Debater and its human counterpart each present a four-minute opening, followed by a four-minute rebuttal and a two-minute closing statement. It’s important to note that the debate topic is not known in advance, and Project Debater is not pre-trained on any specific topic.

6. How does Project Debater learn a topic?


Actually, it doesn’t learn a topic, but it is very good at quickly creating a persuasive narrative based on available data. The Debater system was taught to debate unfamiliar topics. Hence, it can debate many different topics, as long as these are well covered in the massive corpus that the system mines, which includes hundreds of millions of articles from numerous well-known newspapers and magazines.

7. Wait, what’s a corpus?


The dictionary defines it as “a collection of written texts, especially the entire works of a particular author, or a body of writing on a particular subject.” In our case, it’s a very big collection of documents.

8. How does Project Debater form an argument?


Given a topic, the Debater system scours its massive body of knowledge looking for the most relevant points and evidence to support or contest the topic. It then picks the most compelling, diverse and well-supported arguments and arranges them to construct a complete persuasive narrative.

9. Can Project Debater argue both sides?


Like human debate experts, Project Debater can argue either side of an issue. It finds pros and cons based on the topic stance it’s given. So, when presented with a topic like “governments should regulate artificial intelligence,” it figures out which claims are for and which are against the statement. For example, if Project Debater is arguing for the topic motion, the claim “AI needs regulation to both protect individuals and accelerate adoption” would be classified as pro. And a claim like “heavy-handed regulation of AI by any country will stunt its AI progress” would be classified as con. Because it is a machine, Project Debater is impartial—it isn’t out to prove a position or to be “right.” It is designed to help expand minds and help people see more than one side of an issue.

10. How does the Debater system know if a claim is for or against the topic it’s given?


This is one of the many aspects that make Project Debater unique. It has been taught to understand the nuances of language and decide the stance of an argument given the topic. Imagine debating the pros and cons of the use of traffic enforcement cameras. When given the claim “the photo radar program fails to provide any clear safety benefit,” a human debater instinctively understands it contests the use of traffic cameras. But this type of understanding is very hard for AI.

The Debater system approaches this by breaking it down into smaller tasks. The system understands that “the photo radar program” is associated with “traffic enforcement cameras” and further understands that the rest of the sentence—even though it includes positive words like “safety” and “benefit”—is, in fact, contesting the use of traffic enforcement cameras.

11. How is this different than, say, a keyword search?


A keyword search will bring back a collection of relevant documents. Or it may trigger a simple response that has already been prepared. Project Debater, on the other hand, develops a much deeper understanding of the topic at hand, and constructs a point of view based on its findings. That’s no easy task for a machine. To construct the speech, it has to first remove all the redundant claims so it doesn’t repeat the same point again and again. Then it looks at what’s left and extracts the most important points to make an argument.

12. What else makes Project Debater unique?


Project Debater relies on three pioneering capabilities. The first is data-driven speech writing and delivery, or the ability to automatically generate a whole speech, reminiscent of an opinion article, and deliver it persuasively. The second is listening comprehension, which is the ability to understand a long spontaneous speech made by the human opponent in order to construct a meaningful rebuttal. The third is the system’s ability to model human dilemmas and form principled arguments made by humans in different debates based on a unique knowledge graph. By combining these core capabilities, it can conduct a meaningful debate with human debaters.

13. But is the Debater system actually listening to the human debater?


Yes, it is listening using Watson Speech to Text. The system identifies key concepts and claims as its opponent is talking to prepare for rebuttal. Project Debater can listen and digest long, continuous spoken speech—up to four minutes of it.

14. When can I see Project Debater in action?


Check out the video∶


And stay tuned for more info!

is just

Have a question for the Project Debater team? Click here.

Follow us @IBMResearch