Teaching computers how to argue

I'll believe this when I see it, but it's an interesting idea.

Over the past couple of years, however, researchers have started to think that it might be possible to model some aspects of human arguments. Work is now under way to capture how such exchanges work and turn them into AI algorithms. This is a field known as argument technology. The advances have been made possible by a rapid increase in the amount of data available to train computers in the art of debate.

Some of the data is coming from domains like intelligence analysis; some from specialised online sources and some from broadcasts such as the BBC's Moral Maze. New methods to teach computers how arguments work have also been developed. Researchers in the area draw on philosophy, linguistics, computer science and even law and politics in order to get a handle on how debates fit together. At the University of Dundee we have recently even been using 2,000-year-old theories of rhetoric as a way of spotting the structures of real-life arguments.

Until very recently even the most sophisticated AI techniques would have been completely flummoxed by pronouns. So if you say to your smartphone's personal assistant: "I like Amy Winehouse. Play something by her," the software would be unable to work out that by "her" you mean "Amy Winehouse". Hardly the stuff of robot-apocalypse nightmares.

If such simple things can be too difficult for AI, what chance is there that computers could argue? Narrowing our focus down, there are at least two ways in which computers could argue that are tantalisingly close. The first is in justifying and explaining.

It's one thing to look up online how video game violence affects children, but it's quite another to have a system automatically harvest reasons for and against censorship of such violence - an area being explored by IBM, with whom we collaborate. The system that results is like an assistant, making sense of the conflicting views around and allowing us to dig into the justifications for different standpoints.

The second is to develop artificial intelligence that can play dialogue games - following the rules of interaction that can be found everywhere from courtrooms to auction houses.
http://www.bbc.com/news/technology-41010848