IBM once again pitted a computer against two human debaters — similar to a demonstration it held in San Francisco last month — but this time the show was held it IBM’s Givataim offices, in Israel, where the Israeli researchers who pulled off the so-called Project Debater feat proudly took ownership of the product in front of the local press.
At the event on Tuesday, as in San Francisco last month, the computer, called Project Debater — a tall black pillar with three blue, flashing, round, eye-like lights at its front — and the humans, in this case two professional Israeli debaters, Yaar Bach and Hayah Goldlist-Eichler — sparred over two subjects. The first, held against Bach, debated the pros and cons of mass surveillance methods; the second, which the computer held against Goldlist-Eichler, was about the pros and cons of genetic engineering.
The IBM team in Haifa has developed a “very special technology” that is “a significant milestone in the development of Artificial Intelligence technology,” said IBM’s Israel CEO and country manager Daniel Melka to an audience of journalists and IBM employees who worked on the project.
The project is being hailed as a new era for the interaction between humans and machines, in which the computer is challenged to beat humans, transitioning from games and trivia to the ability to talk and undertake significant dialogue.
Over the past six years, a global IBM Research team led by the Haifa, Israel, lab worked on Project Debater, giving it three crucial capabilities, each one breaking new ground in the field of AI.
The first feat was creating a computer that is able to generate an opinion — similar to an op-ed in a newspaper — driven by data and putting it into comprehensive sentences.
The second capability enables the robot to listen and understand its opponents, allowing it to identify key claims, hidden within long, continuous spoken language. The third feat was giving the computer the ability to express human dilemmas succinctly and with principled arguments.
“Project Debater moves us a big step closer to one of the great boundaries in AI: mastering language,” Arvind Krishna, the director of IBM Research said in a post on June 18. “It is the latest in a long line of major AI innovations at IBM, which also include ‘Deep Blue,’ the IBM system that took on chess world champion Garry Kasparov in 1997, and IBM Watson, which beat the top human champions on Jeopardy! in 2011.”
However, unlike a chess game or even Jeopardy, where the rules are clear, the problem is very specific, and results, win or lose, can be ranked, in debates the boundaries are not as defined, and thus there is not always a clear winner. This makes it harder for the machine to learn from its mistakes, Ronit Aharonov and Noam Slonim, the Haifa-based project manager and principal investigator, said at the event.
Instead of playing by the rules of the game, the computer needs “to follow the logic of thought,” Aharonov said. “There is no real scoring function to see who is better. So it is more challenging but also more relevant to the real world,” where things are not always black and white. The computer must adapt to human rationale, set out lines of argument that can be followed, and navigate an unstructured world, all while deviating from a pre-defined set of rules as per a board game, the researchers explained.
During the debate, each side gave a four-minute introductory speech, a four-minute rebuttal to the other’s arguments, and a two-minute closing statement.
The audience of journalists was asked beforehand what their stance was on the subject and how knowledgeable they were about it, in order to gauge after the event whether the debaters had been able to change anyone’s mind.
The subjects were chosen in advance, but nothing else was pre-prepared, the organizers of the event said. While the robot spoke the three round, blue flashing lights transformed into one blue line, similar to a talking mouth. The voice, female, was metallic and robotic, but clear. It made its case by dipping into the vast amount of information it has been fed by the researchers — some 300 million articles and 10 billion sentences. It cited Ayn Rand when it debated against mass surveillance, and FDA studies and other medical research when it spoke in favor of genetic engineering. It also made jokes.
Among the things the robot said: Mass surveillance damages free expression and hinders the free flow of information, thus it has no place in democracy. The UN has said that mass surveillance is an infringement on human rights and the NSA surveillance program is unconstitutional and wrong. Mass surveillance is becoming a dangerous habit, as opposed to being used just as a necessity, the metallic female voice said. Regarding genetic engineering — which the robot advocated in favor of — the robot cited FDA reports and studies from Australia and Pennsylvania, which showed that the techniques are beneficial and safe to use.
“I don’t have any nerves”
Some of its choicest quotes were:
“Let us not deny ourselves advancements, life is too short.” And, “holding on to status quo is like holding on to a dead bouquet of flowers: It makes the room stink.”
And even: “I can’t say this is getting on my nerves, because I don’t have any,” when addressing its opponent, while accusing Goldlist-Eichler of making “populist” statements.
While listening to the human make its points, the robot’s lights flashed away — showing that the computer was registering what was being said. And between its opening statement, its rebuttal, and closing statements, the machine, just like the human, took a few minutes to collect its thoughts — with the flashing lights showing the churning of the algorithms inside. Sometimes it rambled, mentioning points that were not that relevant and pertinent, but all in all it was an impressive show of capability.
The slight monotone of the metallic voice, however, often made it difficult to hold the attention of this listener, who also found the arguments rather data heavy. The human debaters showed more passion, through body language and increased emphasis on words.
Indeed, admitted Slonim, the researchers are currently focused on providing the robot with logic to hold a debate and make a rational case, with less emphasis on pathos. The main idea, he said, was to provide a tool that can help people make decisions. And even if emphasis was placed on making the voice as expressive as possible to keep the public engaged, more research needs to be done in this whole field, he said.
At the San Francisco demonstration, Project Debater beat the human debaters on the subject of telemedicine, changing the mind of nine audience members, but in Givataim the results were not as clear cut. The robot and the human debater tied in the first debate. In the second, the robot scored higher on knowledge and speech clarity, but the human debater managed to change more minds.
While in certain subjects — playing chess, especially — the machines can easily beat humans, Slonim said, there are other areas, such as language, reasoning, and logic, where the rules are not clear cut. It is in these areas that humans maintain their superioriority to machines.
“This is the start of our journey into this territory,” Slonim said. “We need to address these areas in which humans are better and see what can be done in this field.”
Israel team was sure to succeed
The idea for the project started in 2011, after IBM’s success with Jeopardy, with management looking for the firm’s next big challenge. Slonim and his team pitched their idea on one slide to develop an AI system that could debate with humans. In 2012, IBM chose Project Debater as its next big challenge, and things progressed from there.
Oded Cohen, the head of the Israeli research team, said he was extremely hesitant about the proposal and doubtful that the researchers would be able to pull it off. It seemed like too ambitious a project, he told the audience.
But the researchers insisted they could do it and worked painstakingly for six years until they reached the current stage.
“We are starting to translate science fiction into reality,” Aya Soffer, VP at IBM Research in charge of Artificial Intelligence, told the audience.
“AI is a partnership between machines and humans,” she told The Times of Israel in an interview on the sidelines of the event. “Each is better at some things.”
She said more work needs to be done before commercialization of Project Debater, but that meantime, parts of the technologies developed will most likely make their way to existing IBM products like Watson.
“Anyone who needs to make decisions” will be the likely target for the technology, she said, including lawyers, students, financial analysts, or company executives.
The challenges ahead for AI will be in strengthening its ability to reason and learn so it can use logic and common sense, while also democratizing the technology to make it accessible to all.
IBM set up its first branch in Israel in 1949 and was the first multinational to set up an R&D center in Israel in 1972. Its Haifa research center is the company’s largest outside the US. The Israeli team works on key developments for the computing giant, including in the field of Artificial Intelligence, IoT, health, and cybersecurity.