How Good of a Debater is AI? Pretty Good
But it may take all the fun away
August 7, 2018, Bruno Jacobsen
We like to say that artificial intelligence, for all its advantages, cannot do all that we can. And that's true. But if debating with other human was one of those things, IBM's Project Debater shows that it keeps sneaking into human territory.
Recently, IBM's AI debater went head-to-head with experienced human debaters. The two topics for the day were: should we subsidize space exploration and should we increase the use of telemedicine. This was the first public debate of the Project Debater, which had been training with debaters for almost six years behind closed doors.
The result? With the help of an enormous data bank, machine learning techniques, and a few attempts at humor, it managed to win one debate.
If this doesn't seem impressive, keep in mind that it needs to understand what humans say, both their main points and their rebuttals, search information on the topic, create a coherent narrative, and argue back in a way we understand. Some may say this is already too tough for some people, let alone a machine.
But this is what exactly happened. The AI receives no prior preparation, according to IBM, and is ready to debate topics it's unfamiliar with. To do this, it scans hundreds of millions of articles from several newspapers and magazines.
What does it do after that? For a certain topic, it searches the most "compelling, diverse, and well-supported" arguments. Then it's a matter of arranging them coherently and presenting them. This suggests it is almost incapable of missing good arguments for a particular point of view.
According to IBM, three things make Project Debater unique. First, it relies on data-driven speech writing and delivery, which means it can "read" opinion articles and form an argument based on them. Second, it's its listening comprehension and ability to understand spontaneous human speech, to rebut it. Finally, its ability to model human dilemmas.
OK. So the AI isn't taking over yet, nor is it winning debates throughout the planet. However, this is a significant step.
If extrapolated, its abilities could be useful in a range of disciplines. Without the bias of humans and armed with much more data, an AI's ability to form coherent, plausible, and well-supported arguments could revolutionize politics, finance, sales, law, among many others.
While, we may need real people to judge a decision and give it a final go, having AI assisting in forming arguments may turn it into an invaluable tool. But, of course, that's still a bit far away.
Are you interested in future trends, developments, and other phenomena? Futures Platform's foresight radar includes all trends in this blog and hundreds more. Check it out here: Futures Platform