Watson, an artificial intelligence system built by International Business Machines (IBM -0.89%), made headlines in 2011 by beating two champions on the game show Jeopardy! This initial version of Watson was built for answering Jeopardy!-style questions, backed by a vast database of information. Since then, Watson has been adapted to a wide variety of business applications, in industries such as healthcare and financial services.
Soon after Watson's debut, IBM began work on another AI project. Project Debater, an AI system that's been in the works for six years, was shown off for the first time earlier this week at an event in San Francisco. Project Debater participated in two debates with two expert debaters, and it was able to successfully hold its own in each. The system wasn't perfect, but it still represents a major step forward for AI.
IBM's Project Debater and debate expert Dan Zafrir. Image source: IBM.
An AI that can argue
Project Debater goes well beyond interpreting and answering questions. Given a topic, the system was able to produce an opening statement, a rebuttal based on the argument of the human debater, and a summary to close out the debate. Project Debater was trained on various data sets, including information from Wikipedia. But it wasn't trained on the specific topics chosen for the debate.
Project Debater was able to come up with coherent arguments on topics it didn't know ahead of time, and then effectively respond to the human debaters. And it did so convincingly. A snap poll following the closing arguments for Project Debater's first debate showed that the audience thought the AI system enriched their knowledge more than the human debater, according to IBM director of research Arvind Krishna in a blog post.
Project Debater is powered in part by Watson, including Watson's Speech to Text API. But the system also implements new technologies. It can comprehend and identify key claims from multiple minutes of spoken language, and it can make principled arguments by modeling human dilemmas. The goal is not just to have the system generate an argument, but to generate an argument that people can follow. "AI must learn to navigate our messy, unstructured human world as it is -- not by using a pre-defined set of rules, as in a board game," said Krishna.
Building on Watson
IBM plans to commercialize the technologies that underlie Project Debater as part of IBM Cloud and IBM Watson. Some of Watson's capabilities will be expanded by the breakthroughs made developing the debating AI system.
Krishna gave a couple of examples of how this technology could be used to assist in complex decisions: "For example, by helping to identify financial facts that either support or oppose a financial thesis, or by presenting pro and con arguments related to public policies."
It's not hard to come up with cases where Project Debater could be helpful. In its 2016 annual report, IBM talked about how Watson's natural language capabilities were being used by law firms to sift through millions of legal documents and return relevant passages and citations. The technology behind Project Debater could go even further, coming up with its own arguments or strengthening existing ones.
IBM's AI strategy ultimately boils down to providing tools that help people make better decisions. Project Debater expands IBM's AI capabilities, and it helps further differentiate IBM from the slew of other tech companies betting on AI.
