On 11th February 2019, the world saw a first of its kind face-off between machine and man when IBM’s six-year-old Project Debater battled against the world champion Harish Natarajan in a live debate on providing subsidies for pre-schools. This was the first time that Project Debater was put into a full-fledged live debate and it did not disappoint. Project Debater, which has been assigned female gender by its creators, debated in favour of pre-school subsidies while Mr. Natarajan debated against them. The audience voted in favour of Mr. Natarajan and he prevailed as the ultimate winner.
The fascinating thing about Project Debater’s debate was the way in which she created a coherent speech, relevant to the topic within a stipulated time limit, complete with snarky remarks. The format of the debate constituted of 4-minute opening and closing statements on the opposition and the proposition sides and each side was given two minutes to conclude their speeches. While the format of the debate was framed to help Project Debater, there were definitely some bigger feats that she achieved which went way beyond winning or losing the debate.
Not only was Project Debater able to give a brilliant opening speech replete with facts and figures to support her argument but was also able to listen to Mr. Natarajan’s speech and provide targeted rebuttals to his nuanced arguments. This is a machine which is capable of intellectually engaging with a human being, a truly remarkable feat for an AI, which had so far been limited to playing games by the rules.
But what is it that sets Project Debater apart from other game playing AI technologies? The answer lies in the fact that most of the AI’s that have defeated their human counterparts have the ability to play thousands of games against themselves in order to explore every possibility in which the game can go. However, human interaction, especially debating happens in an extremely dynamic environment. Not only is every debate different than the last, but it is also impossible to pre-empt all possibilities in this scenario. Such linguistic battle requires the AI to be highly-functional in real time and adapt itself quickly to meet the intellect of its human opponent.
According to IBM, Project Debater has three core elements. The first is data-driven speech and delivery, which is the ability to generate a speech, contextualize it and deliver it in a persuasive manner. The second is listening comprehension, which is the ability to listen to the opponent and provide meaningful rebuttals. The third is its profound ability to model human dilemmas and form principled arguments that are characteristic of human debates. By combining these abilities, it has become a revolution in the way we look at a machine’s ability to engage with a human constructively.
The most astounding part of the debate was the way in which Project Debater engaged with Mr. Natarajan. It seemed as if it was an intelligent machine that was capable of thought. The secret behind its language ability is the huge corpora of 10 billion sentences that have been pre-fed to Project Debater. While in a live debate, Project Debater picks up the correct sentences from these pre-fed sentences, contextualizes them, makes sure that they follow a logical and persuasive speech within a stipulated time limit. While she has been trained on other topics and against other debaters but this was the first time that Project Debater was debating on pre-school subsidies and she managed to do more than a decent job. Another edge that Project Debater had over Mr. Natarajan was the fact that it could make use of hundreds of factual sources to support its argumentation, something that is impossible for a human debater. However, it lacked in making nuanced principled arguments and contesting claims, something that Mr. Natarajan fared much better at. All in all, it can be seen how an AI debater and a human debater have complementary capabilities that can be brought together to put the ultimate debates of human history to rest.
Project Debater, as noted by Mr. Natarajan and its creators after its first live demo, is a technology that can aid humans in the way they communicate and make decisions. The ultimate aim of Project Debater is to become a crucial aid in human decision making since it can understand principled nuances of arguments and can also combine this knowledge with huge amounts of factual information and contextualize it in a manner that can be understood by humans. It can harness the power of the knowledge available to humankind and be a driver of effective decision making by collating the facts and giving multiple perspectives, allowing humans to make more informed decisions. But this is still a goal that is far off for Project Debater. However, the revolution has started and it is only a matter of time before machines become our new subject matter experts.
Picture Courtesy- Bloomberg Quint