IBM built language-processing AI technology that was a hit after debating against humans. Now it's ready for businesses needing to plow through data.
- IBM's Project Debater became a hit last year when the technology took part in its first verbal showdowns with humans. It lost, but it highlighted advances in language-based AI.
- On Tuesday, IBM announced that it has incorporated features of Debater into its Watson AI product, giving businesses more language-based capabilities in mining and analyzing their data.
- "It's about understanding language and how you can turn that into reasoning," Rob Thomas, head of IBM Watson told Business Insider.
- Click here for more BI Prime stories.
IBM's Project Debater became a hit last year after the groundbreaking AI technology took part in its first public debates with humans.
It lost - but the IBM project turned the spotlight on language-based AI, which Big Blue is now making available to its customers.
IBM said Wednesday that it has incorporated features of Project Debater into its Watson product to give businesses more language-based capabilities in mining and analyzing their data.
"The whole way a company operates is based on language, Whether it's voice-to-voice communication, documents, email," said Rob Thomas, general manager of IBM's data and Watson AI. "This is a treasure trove of information that can be turned into insight and become actionable if you have the capabilities to do that."
And these are precisely the capabilities that IBM harnessed with Project Debater, a major AI project at IBM Research which created a bit of a stir during its public debut last year in debates in San Francisco and Cambridge. The technology underscored IBM's advances in natural language processing research. Now, businesses using IBM's Watson AI platform will have access to national language processing tools based on the Debater project.
"If you look at the guts of what's inside Debater, it's all around language," Thomas said. "It's about understanding language and how you can turn that into reasoning. And so we've taken a number of components out of Debater and put those into the core Watson products."
The tool is able to go beyond basic meanings in words and sentences by also conducting "sentiment analysis," Thomas said.
"Think of idioms and the ways we speak casually," he said. "How do you actually convert that into an understanding? … I could say something that my tone is a negative tone, but in just reading the text that seems positive."
The tool can also help businesses summarize huge amounts of material quickly.
"How do I take 100,000 different documents, and summarize that into two or three key words?" he said. That was a key part of what the IBM Debater was doing, he said.
Thomas cited the example of a law firm that needs to quickly review huge amounts of documents for a case. With IBM's AI technology, that process can be dramatically faster.
"What today takes a lawyer 30 days to do, or it could be two or three lawyers, they can now do in an hour," he said. "So It's incredibly powerful in terms of case discovery, understanding, sentiment, creating summaries of what's going on in all the documents."
IBM's move to commercialize Debater underscored a shift in the company's approach to research. IBM had been criticized for investing heavily in groundbreaking R&D but not moving fast enough to turn that research into viable products.
"I think the biggest thing we changed in the last two years is used to be there was a product organization and then there was IBM Research," Thomas said. "And they were appropriately separate, meaning researchers really focused on five to 10 years plus. We've tightened that a lot in the last couple years because there was a lot of stuff we were doing that wasn't getting into products that I thought was a big opportunity."
Got a tip about IBM or another tech company? Contact this reporter via email at bpimentel@businessinsider.com, message him on Twitter @benpimentel or send him a secure message through Signal at (510) 731-8429. You can also contact Business Insider securely via SecureDrop.