Have you ever been a part of a debate, where a speaker ended up contradicting himself or made a superficial argument such that his entire stance got diluted? Or have you sat amidst the audience of a debate and wanted so badly to point out some obvious facts that the speaker missed which could otherwise strengthen the argument?
Humans err and apparently the fact that we err is what makes us human. How contradicting is that considering that we are the most evolved species on the planet? If computers had a life, I guess they would soon overtake humans in this race and be the most evolved ones out there.
Earlier computers that were used just for crunching some numbers have crossed another milestone. The world is at a stage where there are zillions of facts, opinions, and arguments about the things that matter the most. In many areas, there is a dire need to make policy changes or come to a common consensus to implement any solution. But the immensity of information available and the constraints of a human mind make it impossible to compile all of it to come to a conclusion.
This is where the machine, computers, backed by AI may come handy. Computers use AI to perform a technique called argument mining which allows it to process large amounts of information from various different sources and come to a conclusion. Most of it is used to process legal cases where based on several sections and data available, a claim may be refuted or accepted. However, IBM has been attempting to use this technology to solve such problems through its project called Project Debater.
Project Debater is the first AI-powered, computational argumentation tool that can absorb massive amounts of information and perspectives to arrive at persuasive and meaningful conclusions. When the project was first introduced in 2016, it was incapable of even making an elementary level argument. Over the years, the program has evolved under the leadership of Noam Slonim, the engineer who leads Project Debater at IBM, and in 2019 it reached a stage where the system squared off against a world champion debater Harish Natarajan. However, Project Debater lost to Natarajan as he came out to be more persuasive just like humans naturally are.
Human beings can be very persuasive and use a lot of background knowledge to make arguments. This can be very difficult to implement using AI. However, in the past 18 months, new machine learning models from the likes of Google and OpenAI have been made available which have not only improved the comprehension abilities of Project Debater but the software has become so sophisticated that it can effortlessly weed out obscene or racist language that has been deliberately fed to the system to tamper with the output. Using this software, the system can predict correct meanings of phrases and compose human-like paragraphs of text that could help in making meaningful arguments.
Until now, the Project Debater has been introduced and tested through debates. The way the Project Debater works is that it listens to a long human speech using its listening comprehension. It then prepares for a rebuttal by modeling human dilemmas and forms principled arguments made by humans in various debates and all other information that it has been fed. It then uses data-driven speech writing and delivery, or the ability to automatically generate a whole speech, reminiscent of an opinion article to deliver its rebuttal persuasively.
To train the system, millions of articles and thousands of statements have been fed into it. The statements have been rated for their persuasiveness to help the system understand what makes an argument persuasive to human listeners.
A major challenge that Project Debater faced was to identify which information backed a claim and which did not. Recently, the neural networks of the system were tweaked and the resulting network helped it classify which statement backed the claim and which neither supported nor undermined it. What about these sentences helps Project Debater classify the sentences are still unclear but it serves the purpose.
IBM hopes to be able to help the government better understand the views of its citizens like it did in Switzerland, by collecting opinions from people about whether the government should invest in autonomous vehicles.
If this system can be rolled out successfully, a lot of our time that gets wasted reading fake news, superficial opinions and pointless debates can be used to implement the much-required policy changes all around the world.