• Log In
  • Sign Up
    • Since 2012, IBM has been conducting a research project to develop an artificial intelligence that can engage in a real-time formal debate with humans. The first public demonstration featured a debate between a machine and Noa Ovadia, the 2016 Israeli national debate champion on the assertion "We should subsidize space exploration." The topic was chosen from among a set of possibilities designed to ensure meaningful debate, but the AI was not specifically trained on the subject. Each side gave a four minute opening statement, a four minute rebuttal and a two minute conclusion. A poll showed that a majority of the audience thought the AI enriched their knowledge more than the human debater. AFAIK, no "winner" was chosen.

      As a former debater, I am aware that winning a debate is not the same as arriving at truth--sometimes it's quite the opposite. Still, to debate effectively requires more than simply having a large set of facts to choose from; it requires an understanding of the logic of argument and the ability to persuade. This goes way beyond the abilities of Siri or Alexa. For now.

      I haven't been able to locate a complete video of the debate, though I imagine one will be available soon. In the meantime, IBM has posted a teaser here, and you can find details in their project blog. Like most of the developments in AI one can see both good and bad possibiliites: imagine conversing with an AI as a means to clarify your thinking about a topic. Now imagine an AI robocaller with superior persuasive abilities. Fascinating stuff, I think

    • Okay, that's just crazy. Here's the clash I imagine will be epic: listening to the Cambridge Analytica guys, their claim is logic and facts do not win large public debates, only emotion does. There is an old saying in the law: if the facts are on your side, argue the facts. If you have the law on your side, argue the law. If you have neither, use emotion.

      I am the type of person who keeps coming back to facts and logic, feeling that surely it has to be the way to solve big public issues like health care and taxes, but if the answer really is no, it's predominantly about emotion, then doesn't AI have a hell of a challenge?

    • It's probably not as hard as you might think. Many of today's customer service automated voice response systems are already capable of detecting a caller's anger or annoyance and will transfer the call to a human agent quickly. So it's easy to imagine an AI which is attuned to the emotional response of humans even though the AI has no emotions itself. It probably wouldn't take long for the AI to learn how to evoke a sympathetic response. I'm not saying this is desirable; it's certainly not a good way to develop public policy. But focus groups have been around for decades to help politicians and marketers hone their messages. In principle, an AI debater might not be very different, only more effective. The question really is whether the goal of the AI is to promote the common good or to serve narrower interests.

    • Yeah, me too, but the last time I looked, no full video was available. The blog did have some additional details on the training sets used, if you're technically inclined.