IBM AI wins debate against human champion
Checkmate – it's a simple concept: the final decisive move that ends a game of chess with one winner and one loser – just like IBM's Deep Blue which beat world champion Gary Kasparov in 1997. Not so debating: an altogether softer battle of wits, where the question of who won is far from straightforward – and arguably subjective. Which is much like the art of debating itself – lacking the easily-defined rules of the game of kings. All the more impressive, then, that IBM's AI technology won, according to an audience vote, against a seasoned human debater at a San Francisco event on June 17.
We first reported on IBM's debating AI back in 2014 when it fell under the Watson umbrella, which describes IBM's embeddable question and answer AI platform. Now dubbed Project Debater, the project is designed to sift trusted information sources to provide valuable insight in an age flooded with information, misinformation, and information accused of being misinformation.
"Eventually, Project Debater will help people reason by providing compelling, evidence-based arguments and limiting the influence of emotion, bias, or ambiguity," states the project website, somewhat boldly. Yet the events of June 17 would seem to lend credence to the claim.
According to a BBC report, Project Debater performed without an internet connection. Nevertheless, the AI has access to hundreds of millions of locally-stored documents – largely newspapers and academic journals. More impressively, it had no warning as to the subject of the debate, though there was a list of 100 topics that IBM thought the AI would be up to tackling.
Its opposition in one debate was Noa Ovadia, national debating champion of Israel in 2016, who recently started working with IBM on the project. Their topic was public funding for space exploration: Project Debater for and Ovadia against. They each made a 4-minute statement followed by a 4-minute rebuttal and a 2-minute conclusion.
"It is very easy to say that there are more important things to spend money on, and I do not dispute this," argued Project Debater in response to Ovadia's argument that the money could be better spent. "No one is claiming that this is the only item on our expense list. But that is beside the point. As subsidising space exploration would clearly benefit society, I maintain that this is something the government should pursue."
The audience vote went in favor of the AI as being the more informative.
It's still a work in progress. Project Debater is prone to repeating itself, albeit in different words, and it sometimes uses peculiar phrasing. But these are relatively minor quibbles. Much less important than the AI having the rehearsed delivery of a human speaker (though it does crack the odd joke) is its ability to mine a hoard of information and present a cogent, informed argument based upon it. Arguably the quality of that argument will always be limited by the quality of the data its drawn from.
The development it raises all kinds of questions about the future of information and opinion. For instance, if we're to be told what to think about an issue, is it better to be told by a person or organisation with, at best, unconscious biases or at worst a hidden agenda, or a dispassionate machine? But then is a machine dispassionate if it ultimately serves human masters – or develops a bias or two of its own?
You can see a short video of Project Debater below