Thanks to popularity both in academic circles and commercial markets, AI research and technology have grown exponentially in the past few years. Now AI and neural networks can not only beat humans in games, they can also craft sometimes disturbing works of fiction. But one thing that AI have not yet been able to do sufficiently is to actually explain their “thinking” in plain language. IBM’s Project Debater can and it can be impressive or scary, depending on which side of the AI debate you’re on.
Most people might be familiar with AI as it is found in the likes of Siri, Alexa, and Google Assistant. While non-trivial, these AIs are often trained on simpler data, mostly what you have stored on your phone or in the cloud or what’s available via a Web search. Their most sophisticated capabilities fall in the realm of natural language processing (NLP) and figuring out the best answer to give from a dozen possible search results.
But imagine an AI that is not only able to understand your statements but can also comb through massive amounts of data and then respond to you, in plain English, arguing the opposite position. And all in real-time. That’s the remarkable feat that IBM’s Project Debater is accomplishing. It’s not yet a perfect debater, but it’s significantly more advanced than an AI Go player.
Project Debater combines at least three capabilities, all of which are heavyweights in the AI field. There’s the ability to actually comprehend an opposing human debater’s speech and identify the key points in sometimes long-winded sentences. Then there’s the modeling of data into knowledge graphs to enable it to argue its position. And finally, the AI has to formulate not just comprehensible but also logical spoken arguments.
While some might worry this kind of AI will eventually argue its sentience and superiority over humans, IBM’s goals for Project Debater are less lofty and more practical. Eventually, IBM hopes that these AI technologies will help in broadening human knowledge on a matter, the way debates do, by arguing for or against a particular topic, even one that hasn’t been digested by the AI before hand.