Watson AI is Debatable
An AI system under development by IBM researchers seeks to move its Watson platform beyond chess matches and game shows, combining natural language processing (NLP), listening comprehension and the ability to model human dilemmas to create an agile debating machine.
Project Debater also uses sentiment analysis, deep neural networks and machine learning techniques to “mine” the claims and evidence behind an argument. The debater can tackle subjects it has not been trained on, instead scanning text and key sentences in minutes, selecting the strongest evidence for its position, then delivering an open statement in debates with humans.
It then listens to an opponent’s response before formulating a rebuttal.
Rather than an academic exercise, the AI research is intended to augment human decision makers with tools that will inform their decisions.
The six-year-old AI project has so far generated 45 technical papers and benchmarked data sets on subjects ranging from NLP and “argument mining” to “weak supervision” deep neural networks.
In a reference to the rise of “Fake News,” IBM researchers note the need for better decision-making tools in a “world awash in information, misinformation and superficial thinking.”
The researchers added that Project Debater represents the latest advance in machine learning that started with tools like email spam filters and virtual assistants. They now seek an AI system that moves beyond the IBM Watson system that could answer open-ended questions on “Jeopardy!” to a free-form AI platform capable of debating humans on complex subjects.
Among the AI advances yielded by Project Debater is a tool used to automatically and concisely summarize arguments. The framework represents the summaries as a set of “key points.”
“By analyzing a large dataset of crowd-contributed arguments, [it was shown] that a small number of key points per topic is typically sufficient for covering the vast majority of the arguments,” IBM researchers reported in a recent paper.
“Furthermore, we found that a domain expert can often predict these key points in advance,” they added.
The IBM researchers proposed a large data set that could be used for “argument-to-key point mapping.”
Project Debater data sets are here.
The other advance addresses the ability of machines to ponder human dilemmas, considering both sides of an argument. That vexing problem is reminiscent of the existential debates between HAL the Computer and Commander Dave Bowman in the film, 2001: A Space Odyssey.
Astronaut and machine first argue over whether HAL should allow the stranded Bowman back onboard the mothership. Sneaking in through the emergency airlock, Bowman and HAL then debate the wisdom of human intervention in a momentous mission to Jupiter, as well as Dave’s heated and arguably ill-informed decision to disable HAL’s higher cognitive functions.