IBM’s debating A.I. is added to Watson
IBM is making artificial intelligence technology it created to debate humans available to businesses, the company said Wednesday.
The connection between debating and the corporate world may not be obvious. With the notable exception of the late Chrysler CEO Lee Iacocca, the roster of high school debate champions who gained prominence is heavy on politicians and lawyers and light on industry titans.
But while few businesses engage in the rhetorical competition for which IBM’s “Project Debater” A.I. was trained, IBM is banking that some of the underlying skills the system perfected will have clear uses in the corporate world.
IBM said it is adding several of the technologies it honed with Project Debater to its Watson-branded digital assistant, which businesses use much like consumer digital assistants such as Alexa or Siri. The new skills will automate either customer service tasks or internal business processes, and improve its Watson-branded document discovery tool, which extracts and sorts information from documents.
These new capabilities include a more advanced system for categorizing the sentiments expressed in text and finding particular clauses in documents, a tool that can summarize information, and another that can group information into more precise clusters.
Since 2012, IBM researchers have worked on perfecting an A.I. system that could challenge humans at debating. In order to succeed, the software would have to master a number of challenging tasks for a computer: speech recognition, searching for relevant information, and then grouping it in by topic and argument, summarizing those groupings, ranking them by effectiveness and then generating persuasive text.
In the past two years, IBM has showcased this Project Debater A.I. in a series of demonstrations. Last year, Debater narrowly lost to world champion debater Harish Natarajan before a live audience in San Francisco. In that debate, the A.I. had to source its arguments from news stories, editorials, op-eds, and journalistic essays.
In November, it assisted two teams of human debaters in a demonstration at the famous Cambridge Union debating club in Cambridge, England. In that case, Project Debater took its arguments from crowd-sourced opinions submitted through a website in the weeks leading up to the debate.
The fact that IBM is now moving to commercialize aspects of Project Debater shows how the company has tried to speed the movement of ideas from its well-respected research division into its commercial Watson A.I. products, Rob Thomas, general manager of IBM Data and AI, says. The company has been leaning heavily on its Watson product line to boost its cloud computing division’s sales and help arrest a long-term decline the company’s technology services revenues.
While Thomas acknowledged that IBM had, in the past, sometimes been slower than rivals in bringing technology from its research into its products, it was consciously trying to speed up that process. “This is a very different innovation pipeline coming from IBM than you’ve seen in the past,” he says.
The new sentiment analysis technology that IBM is taking from Project Debater will be integrated into its Watson Natural Language Understanding tool later this month, the company said. This won’t turn the Watson digital assistant into a debate champ. Instead, Thomas says the new sentiment analysis software will better understand words in context as well as understanding idioms that natural language processing systems have often found tricky, such as “a can of worms,” or “hardly helpful.”
In another advance taken from Project Debater, the Watson platform will also get better at finding and classifying clauses in business documents, such as legal contracts, Thomas says.
The new summarization technology will help businesses summarize even millions of documents into “into bite-sized insights,” the company says. It says an early version of the technology was tested at this year’s Grammy music awards to help provide capsule biographies and summaries of fan views of certain artists.
Thomas says IBM’s client accounting firm KPMG has also used the summarization technology to help its clients get R&D tax relief. The system scans documents and summarizes information that indicates the company qualifies for the tax benefit.
Finally, the topic clustering technology, which also comes from Project Debater, will allow companies to create more fine-grained groupings of information, Thomas says. He points to an example of French bank Crédit Mutuel, another Watson A.I. customer, and how many of its contact center calls were being routed into a “general inquiry” queue instead of more specific ones because the existing technology could not correctly categorize the caller’s questions in order to assign it to a narrow, more-focused category.
Thomas says that improved clustering would now allow that “general inquiry” bucket to be cut down by 80%, increasing the speed with which customer inquiries can be resolved.
More must-read stories from Fortune:
—Google Doodle celebrates International Women’s Day
—Growing coronavirus threat weighs on Apple
—When will PlayStation 5 and Xbox Series X debut?
—NASA hiring new astronauts for the first time in four years
—WATCH: Best earbuds in 2020: Apple AirPods Pro Vs. Sony WF-1000XM3
Catch up with Data Sheet, Fortune’s daily digest on the business of tech.