The staging is not futuristic. In the center of the stage there is a black column, like a narrow television set up. That's Project Debater. It has a prototype name and no pretension to humanity: it barely shows a discreet oval light when it speaks. It has a mechanical female voice, like Alexa or Siri, but it's much more than a home speaker. Project Debater reasons, understands arguments and even jokes. His intellectual agility shows how far he has advanced artificial intelligence, but also the slowness of the progress: for now its themes are limited, its exposure may waver and the format of its interventions is fixed.
"I can not experience poverty directly and have no complaints about my own standards of living," says Project Debater
"I can not experience poverty directly and I have no complaints about my own standards of living," says Project Debater shortly after beginning his first intervention. The audience laughs without much conviction, as if wondering if that machine had made an acceptable joke.
After its public presentation in June 2018, Project Debater faced this Monday in the framework of IBM Think 2019 in San Francisco its biggest challenge. Before her, a formidable rival, Harish Natarajan, graduated from Oxford and Cambridge and with more victories in the world in debate championships. The format was common for a competition of the genre: two rounds of 4 minutes and another final of 2 minutes to conclude.
The occasion recalled other definitive battles between machines and humans, such as chess, go or Starcraft. "Today we can make history," said the moderator, if a robot passed another milestone. Although a debate is different. The victory is awarded by a jury with a subjective criterion, not a result on a board. The subject to be discussed was: "The State must subsidize pre-school education". Project Debater defended the proposal, Natarajan rebutted it.
In the end, Natarajan won. The victory was for those who changed the opinion of more people in the audience. Natarajan had in favor that few people believed that the State should not subsidize the education of the smallest. So convincing a few attendees was enough.
It's like playing the Trivial against the Encyclopedia Britannica, Wikipedia and the file of the 'New York Times' together
The organization had added a second question: who has enriched your knowledge? It was a question for the brilliance of Project Debater, which devastated 60%. With reason: their knowledge is infinite. Its database includes 10,000 million phrases, 400 million articles. To compare: from 1851 to 2016 the New York Times has published some 15 million articles. Project Debater manages in minutes more than 25 files of that magnitude, most of them taken from newspapers and scientific journals. Not even the Encyclopedia Britannica (5.8 million articles) or the Wikipedia in all languages (40 million) are comparable. The challenge of confronting Debater is how to play the Trivial against the Encyclopedia Britannica, Wikipedia and the archive of the New York Times together.
The distinction in knowledge was overwhelming. Project Debater was taken out of the system studies of the University of Melbourne or Duke on children without studies and delinquency, cited Australian or British prime ministers word for word. Given this fact, Natarajan balanced with arguments expressed in a more floral way. But no evidence, no proper quotation, just more rhetoric: Debater never exceeded the allotted time, Natarajan always.
This does not mean that the arguments of Proyecto Debater did not have finesse. In his huge memory there are also refined articles: "My intention is not to leave a suitcase full of money for anyone who wants it," he said, to defend that he did not want to give away public money.
He is not aware
Project Debater is not aware. He is programmed to look for the best arguments for his thesis. The jokes, for example, are in a "bag" of ironies that give robot personality: "Their challenge is to take them out at the right time," says Noam Slonim, director of the Debater Project in Haifa (Israel), where it develops. It is not easy to define the right moment or an argument in favor. Hence part of the fascination. Its merit is not only to reason but to listen to its rival, to understand the core of its arguments and to refute it. "For starters, sometimes I listen to my opponents and I wonder if they want to, if they prefer people at their doors asking for money, that there are people without food and drinking water." Giving opportunities to the less fortunate should be a moral obligation of any human "says Project Debater.
There was the machine using "moral" and humanity arguments against a human, which in contrast seemed cruel. Again, the robot does not understand why it is moral, it only knows that it is effective. Its creators, who have been with Project Debater for more than 6 years, they get surprised when they hear some subtle argument, that it takes out of something that could be called "knowledge". His artificial intelligence is "wider," says Slonim.
The technological marvel must not forget a key detail: it improvises only in a prepared and favorable terrain. The researchers behind the project offer a list of topics to discuss. The communication department of IBM chooses one that is current and little thorny: abortion, for example, no. Project Debater defends him or attacks him with a rival, but could not face an interview where the questions are addressed. You need a thesis to support or refute. At least, at the moment.
Languages are another problem. Debater understands and speaks in English. Preparing a Debater in another language is not automatic. The corpus of knowledge -the academic journals- is not the same as in English.
The trail of evidence that Debater would take out in Congress would leave his lordships stifled
That said, it's wonderful to think about how Project Debater would work in a human gathering and screaming on TV or even in the Congress of Deputies, where public policy is so badly debated. The trail of evidence that Debater would take would leave his lordships stifled. His list of controversial topics is not only general, but also topical. Brexit was one of the options for Monday's debate in San Francisco. Also newspaper columnists must fear. If it's for arguments, Debater has them all and well ordered.
"The truth is that your first intervention could become a newspaper column," says Slonim.
Debater, like any specialist in debates, is as good at defending a position as his opposite. From IBM they do not see another application that together with humans: Debater would give all the arguments in favor and against an issue and the human would decide which one he prefers. Thus, its possible biases due to the data flow would be visible. Education can be one of its commercial destinations: "It would foster critical thinking," says Slonim. It is not competition, but collaboration. At the moment