AI confidently argued with the record holder on victories in debates and rejected his point of view. The audience was more convinced by “their” participant.IBM computer and one of the best debaters in the world Harish Natarajan A shot from the broadcast
At an IBM conference in San Francisco, California, an open debate was held with the participation of artificial intelligence from IBM. After they ended, the audience voted with the help of smartphones and awarded the victory to the person. The full recording of the broadcast, which lasted almost an hour, was published on YouTube.
Artificial Intelligence Project Debater with a female voice (which can be translated as “Project Contributor”) was opposed by the finalist of the 2016 World Debate Championship Harish Natarajan. The discussion was moderated by a journalist and four-time winner of the Emmy Award, John Donvan.
IBM announced the topic of debate, subsidizing preschool education, right before the start. The Project Debater was in favor, and Natarajan was against. The participants were given 15 minutes to prepare. Just before the start, the Project Debater made the audience laugh, greeting the opponent.
[perfectpullquote align=”full” bordertop=”false” cite=”” link=”” color=”” class=”” size=””]I heard that you hold a world record in the debate against the people, but I believe that you have never argued with the machine. Welcome to the future.[/perfectpullquote]
The parties made a four-minute speech, refutations of the opponent’s arguments, as well as a two-minute closing remarks. At the same time, the AI was disconnected from the Internet, so I based the arguments on the already “gained” knowledge. The difference also lies in the fact that the Project Debater analyzed more than 10 billion in-memory suggestions from news stories and scientific journals, and Natarajan only sketched notes on a piece of paper.
Project Debater provided meaningful arguments, dividing them into sub-topics, and also backed them up with data from research and policy statements. The AI explained that government support for preschool institutions would help “protect children from the most vulnerable segments of society.” As CNET noted , it was the computer that reminded people how to behave: “Providing the less fortunate should be a moral obligation for any person.”
Natarajan also noticed that the allocation of subsidies to kindergartens is far from necessarily improving the quality of education for the poor. He also called such measures “no more than a politically motivated distribution of money to middle class […] people, and not to people in a disadvantaged position.”
In the debates, the Project Debater showed the ability to draw parallels, emphasizing that subsidizing health insurance can be useful. The AI also demonstrated refutation skills: for Natarajan’s assertion that “there are more important branches for subsidizing,” the computer remarked that it is not talking about distributing money without a limit, but about “targeted and strictly limited” help. However, sometimes a person responded more specifically and to the case than the AI.
— Thiago Teixeira (@thiago_teixeira) February 12, 2019
The audience of 416 people voted twice: before the start of the debate and after the end. According to the results, Natarajan won against the Project Debater, since he reassured 17% of the audience. However, the audience answered the additional question differently: in the opinion of about 50%, it was AI that brought them more knowledge, while Natarajan and the “equally” variant got about equal votes.
After the victory, Natarajan admitted that he was struck by “the potential value of the Project Debater thanks to the amount of knowledge she can comprehend.” A professional debater and a Cambridge alumnus called the AI’s arguments well-worded and relevant to the context of the discussion.
IBM began to develop the Project Debater after the victory of another Watson computer over the participants of the quiz show Jeopardy, which is similar to Own Game.
Project Debater has debated publicly with people for the third time. In June 2018, debate debtor Noa Obadiah defeated AI in a dispute over state support for space research, but on the same day, another expert Dan Zafirir gave way to a computer when discussing telemedicine. Both decisions, as in 2019, were made by the audience. IBM developers noted that in a year the Project Debater took a big step forward.