AI

Behind the Scenes of Project Debater – Speech by Crowd at CES: Part One

Share this post:

This week at CES we unveiled Project Debater – Speech by Crowd, a new and experimental cloud-based AI platform for crowdsourcing decision support. Speech by Crowd uses the core AI behind Project Debater to collect free-text arguments from large audiences on debatable topics and automatically construct persuasive viewpoints to support or contest the topic. Every day at CES brings a new topic and a new challenge for Project Debater—and we’re proud to see it rising to the challenges.

The challenge of crowdsourced content

First, we welcomed CES attendees and people around the world to submit arguments for and against the topic “Gambling should be banned.” The majority of arguments were against the proposal. Using that input, Project Debater constructed two original speeches, which you can check out here. To construct the speeches, Project Debater analyzed the arguments, compiled them into related clusters, and selected the most representative and highest-quality ones to include in its speeches. You can explore how the speeches were constructed through the Speech by Crowd experience online.

When Project Debater debated two humans back in June, it relied on 10 billion sentences from newspapers, magazines and journals. Speech by Crowd puts a new spin on this concept by analyzing information submitted by participants. Using crowdsourced content as input poses some new challenges, such as typos, grammar mistakes, and variable argument quality. We are proud that Project Debater handled these challenges so well today.

We also enjoy Project Debater’s ability to deploy wit – evidenced today in its speech supporting the idea that gambling should be banned. A key theme in this speech was that gambling facilitates criminal activity. Project Debater introduced this point by asserting, “It is no crime to share the following comments about the subject,” prompting chuckles from the crowd that had gathered to hear the speeches.

The challenge of ambiguity

Project Debater – Speech by Crowd faced its second challenge today when we invited people to contribute arguments both for and against the idea that we should stop the development of autonomous cars. Hundreds of arguments were submitted—it is probably no surprise that autonomous vehicles are a hot topic at CES—and almost two-thirds of them were against the idea of stopping the development of autonomous cars. But Project Debater did not choose sides, constructing compelling speeches representing both the pro and con positions. You can find the resulting speeches here and drill down into how Project Debater came up with them.

Arguments were written by hundreds of people in their own words, and from these very disparate submissions, Project Debater extracted key themes, summarized them concisely, and chose appropriate arguments to support each theme. This resulted in two coherent and persuasive narratives, representing both sides of the debate around autonomous cars.

It was interesting to see how Project Debater handled statements whose polarity (whether they were pro or con) was ambiguous. Project Debater isn’t told a priori whether a statement supports or contests the topic; it determines this for itself. For example, in its speech against stopping development of autonomous cars, Project Debater included the argument, “Autonomous cars need new insurance systems and new legal rules, because it must be clear who bears costs if an accident happens.” We felt this argument seemed to suggest that development of these vehicles should stop until such measures are in place. But a member of our audience pointed out the argument could instead be interpreted as encouraging the involvement of insurers and regulatory authorities, which would be a positive outcome of continued development. This ability to present new points of view is a unique and exciting aspect of Project Debater; the technology was designed to help expand minds and help people see more than one side of an issue.

Add your voice

Submissions are now open for our next topic of the day, “We should ban the sale of violent video games to minors.” Visit the Project Debater – Speech by Crowd experience now to share your arguments, then come back tomorrow for the resulting speeches!

IBM Research

Noam Slonim

Distinguished Engineer, Project Debater, IBM Research-Haifa

More AI stories

The path to the “perfect” analog material and system: IBM at IEDM and NeurIPS

Researchers from the IBM AI Hardware Center will showcase at IEDM and NeurIPS new analog devices, algorithmic and architectural solutions, a novel model training technique, and a full custom design.

Continue reading

Quantum-Inspired Logical Embedding for Knowledge Representation

In our new paper, to be presented at NeurIPS 2019, we develop a new knowledge representation, which we call “quantum embedding”, that represents conceptual knowledge using a vector space representation that preserves its logical structure and allows reasoning tasks to be solved accurately and efficiently.

Continue reading

Learning and Characterizing Causal Graphs with Latent Variables

Researchers from the MIT-IBM Watson AI Lab developed a new approach to characterize the set of plausible causal graphs from observational and interventional data that has latent variables.

Continue reading