SAN FRANCISCO: IBM'S AI DEBATING SYSTEM has been put through its paces live on stage for the first time at the firm's THINKQ conference in California.
Touted as the first computer system able to debate complex topics with humans, IBM's Project Debater is based on a raft of deep artificial intelligence such as data-driven speechwriting and delivery, listening comprehension and modelling human dilemmas.
It was first unveiled in a small closed-door event in summer last year, and this week the cocky computer flexed its muscles in front of a proper audience. It was pitted head to head with the grand finalist at the 2016 World Debating Championships, Harish Natarajan, also head of economic risk at AKE International.
Hosted by debate promotion firm Intelligence Squared US moderator John Donvan, the live debate focused on for and against the resolution: "We should subsidise preschools".
Each debater had 15 minutes to prepare three four-minute speeches in three rounds: an opener, a rebuttal and a summary speech. IBM said nothing was recorded or scripted in the lead up the debate.
The Project Debater AI machine kicked off the event, referencing a host of studies to back up the benefits of subsidising preschools for four minutes, while the human debater took notes. While IBM said the AI that makes up the debater is genderless, it chose the voice of a woman to make its arguments.
The AI debater argued in favour of subsidising preschools because it can benefit the impoverished, make for better students and reduce crime rates. It made these points while citing a broad range of sources including the National Institute for Early Education Research and the Centers for Disease Control and Prevention.
Doing what's considered morally right by humans were also core parts of Debater's opening argument and rebuttal.
Natarajan was very positive of Project Debater, but did say that it doesn't deal with underlining problems in society.
"I think project debater explains something intuitive, but I don't think its ever enough justification for subsidies," said Natarajan.
Unfortunately for IBM, the AI bot lost the debate, which was scored based on how many people that made up the audience changed their minds. Before the debate, 79 per cent agreed with the position in favour of preschool subsidies, but afterwards, the figure dropped 17 percentage points to 62 per cent.
Under the hood
The AI system works by pinpointing relevant evidence from a massive memory of 10 billion sentences and can pinpoint text that is relevant to topic argumentative in nature and support the debate and glue them together in a relevant narrative.
Using Watson speech recognition capabilities, it can understand the words being said by its fellow debater and digest these into meaningful sentences.
"The trick is to capture the gist of what [the fellow debater] has said using AI engines in order to anticipate what he will say next to try and determine whether he was saying his argument explicitly or implicitly," explained one of the system's creators during the debate's break.
"The system uses a unique collection of arguments relevant to the topic, touching on core issues of what to say, when to say them, and what is justified, using listening and comprehension tools."
Intelligence Squared US hosts lots of debates on a massive range of topics to "help raise the level of public discourse" insisting on civility but make a contest out of the challenge to debate intelligently and persuasively, a bit like TED talks on steroids.
The IBM Project Debater machine is a testament to this. Just before the event kicked off, host Donvan said: "Envision a future of the tech beyond the podium, to help people form better arguments and make a better decision."
But the real point of the debate was not who won or lost. It was to demonstrate how the AI technology that has been developed can bring skills as well as people and AI together.
It's not a question of AI being better, the goal of the tech is to show how clever AI can really be - to find information in a mass of text and make it contextually useful, bringing it into a position understandable for humans so that they can use it to help improve how they think and interact when it comes to debating in the future. µ
It's the week in Google news
Erik Estrada wouldn't have stood for this
Hacks in support of WikiLeaks founder target gov websites