What’s All the Fuss About? IBM Debater

Robots can now make more persuasive arguments than humans

What is it?

A new supercomputer, six years in development, that uses artificial intelligence to debate topics with humans. In a recent public debate in San Francisco, it ‘beat’ its human opponents with stronger arguments.

Who decided that it won?

The audience. They judged that while the humans were better at delivering points, the computer’s reasoning was more persuasive.

What were they debating?

Two propositions: that government should pay for space exploration, and that telemedicine – diagnosing patients remotely by video – should be used more. IBM Debater supported both statements; its human opponent argued against.

How does it work?

By drawing on evidence from its huge database, which comprises hundreds of millions of newspaper and magazine articles on “about 100 areas of knowledge”. This means it should be able to debate a topic without knowing beforehand what it will be.

It can work out whether an article is for or against the topic being discussed, which its creators say makes it less biased than humans. Dr Noam Slonim, IBM’S ‘principal investigator’ for Debater, claims the computer doesn’t regurgitate its entire argument from a single source. He told BBC Radio 4’s Today programme that Debater “is not copying a whole article, it is not even copying a whole paragraph either, it is picking single sentences – and in some cases just clauses – taken from many different articles, then it glues that together into a coherent, persuasive narrative”.

It’s an extremely clever method, but it didn’t stop Debater being caught out plagiarising some points.

Such as?

It used the phrase “having a space exploration program is a critical part of being a great power”, mirroring the exact words Mark Whittington wrote in the Wall Street Journal in an article about Japan’s plan to send people to the moon. Its debating style is rather sneaky, too.

In what way?

It’s been programmed to suggest its opponent is lying about certain points, emulating a distracting technique often used by humans when they have a weak case. It also tries to engage with the audience by telling jokes.

Are they funny?

Judge for yourself. Arguing for telemedicine to be used more, it said: “I am a true believer in the power of technology… as I should be”. Not exactly Oscar Wilde.

What could Debater be used for?

On its blog ( www.snipca.com/28237), IBM says Debater could deliver fact-based arguments that help humans make better-informed decisions in a variety of sectors, including healthcare, criminal justice, counter-terrorism and hiring people in the workplace. Companies might end up using it to resolve conflicts in the boardroom. Debater could listen to both sides, and challenge arguments.

What does it look like?

A bit like the enigmatic black monoliths in Stanley Kubrick’s 2001: A Space Odyssey (pictured left). It’s six-foot tall, with a blue animated ‘mouth’ (see main image, and watch it debating at www. snipca.com/28262).

So will it be appearing on Question Time soon?

Not yet – it’s still far from perfect. Rather than respond instantly to its opponents, Debater spends a few minutes processing what they said (we’re not sure a TV audience would wait that long). Also, some of its sentences don’t make sense, while its anecdotes feel unnatural. Mind you, some might say those flaws would make it feel right at home on the Question Time panel.