Dr. Ranit Aharonov (L) and Dr. Noam Slonim (R) with IBM's Project Debater – Speech by Crowd machine.
(photo credit: OR KAPLAN)
Could the most contentious and complex issues in society be more persuasively argued by artificial intelligence?
That hypothesis was tested on Thursday as American technology giant IBM unveiled its Project Debater–Speech By Crowd technology at the company’s Think Summit 2019 in Tel Aviv.
The artificial intelligence-based Project Debater technology, conceived at IBM’s Haifa research laboratory, has previously debated with individual humans, taking on and narrowly losing to world champion debater Harish Natarajan in February in an argument over subsidizing preschools.
Yet aiming to foster quality decision making when faced with masses of opinions on contentious issues, Project Debater’s Speech by Crowd technology is able to collect large quantities of text arguments on issues from a wide audience, and contribute persuasive arguments that can assist decision makers.
Developed by a team headed by Dr. Noam Slonim and Dr. Ranit Aharonov, Speech by Crowd uses artificial intelligence to analyze arguments, generate coherent narratives out of them and present them as two speeches – one for the pro arguments, the other for the con arguments.
Among the most contentious debates raging in many societies today concerns cannabis legalization, with countries worldwide – including Israel – increasingly evaluating new approaches toward medical and recreational use of cannabis.
Accordingly, that was the tricky topic selected for Speech by Crowd’s first live debating experiment.
Some 1,037 short arguments – 637 in favor and 388 against – were put forward by conference-goers to the question: “Should cannabis be legalized?” before Speech by Crowd turned them into concise speeches on the conference floor.
“Shalom everybody,” said the machine debater, which proceeded to formulate four key issues submitted by respondents supporting legalization: reducing pain for sick patients, cannabis being no more harmful than alcohol, increased tax revenues and improving general well-being.
“It is illegitimate for the government to limit people’s choices of consumption based on the harm the substance might cause them, especially because alcohol and nicotine are currently legal,” Speech by Crowd told the audience, speaking in a near-perfect, human-like manner.
Arguing against legalization, the computer highlighted cannabis serving as a gateway drug; the substance’s potential to harm memory; the increase in black market demand; and the distraction from work caused by marijuana.
“Legalizing cannabis will lead to high public health costs in dealing with addiction, psychosis and other side effects of this dangerous drug,” said the computer. “It would begin a process of legitimizing and legalizing more and more dangerous drugs.”
Dr. Aya Soffer, VP of AI Tech for IBM Research AI, told The Jerusalem Post that the Project Debater initiative is about understanding language.
“There has been incredible progress over the last five or so years in artificial intelligence in general. But one of the areas which is still nowhere near artificial intelligence is the ability of the computer to understand human language,” she said.
“It’s a difficult problem because language is what differentiates human beings from other beings. For computers to eventually be intelligent, language is one thing we need to advance significantly.”
Soffer said that key challenges in training an artificial intelligence system to build arguments include needing to automatically identify whether an argument is pro or con, especially given the subtleties of human language, removing parallel arguments made using different words, detecting underlying themes, and capturing the amorphous notion of argument quality.
“When I think about the future and artificial intelligence as something we believe will help but not replace humans, communication is key to us being able to work together,” said Soffer.
“We need to understand what are the real pain points that we want to solve with this technology,” she said. “It’s applicable to all fields where you’re summarizing masses of information in a coherent way. It’s applicable, for example, to journalism, to health professionals dealing with many clinical studies and also the financial domain.”
IBM Israel general manager Daniel Melka hailed the technology as one of the key breakthrough moments he has witnessed during his time at the company.
“It doesn’t surprise me that an artificial intelligence system that is able to argue was developed in Israel,” Melka said jokingly.