IBM Computer Beats Debating Champion

John Lister's picture

An IBM computer has out-argued a debating champion according to an audience vote. It did so despite not having an Internet connection to look up facts during the debate.

The demonstration wasn't designed to test how much the computer "knew," but rather it's ability to form arguments, listen to an opponent's case, and then respond to it.

The computer system, dubbed "Project Debater", took on Noa Ovadia, who was crowned national debating champion in Israel in 2016. The pair debated two topics: whether to have more publicly funded exploration of space and whether to put more money into telemedicine technology. The latter uses telecommunication and information technology to provide clinical health care from a distance.

100 Possible Topics On The Table

Rather than an extended back and forth argument, the debate had a simple format: for each topic, the two debaters took turns to make a four minute statement, then spend four minutes replying to their opponent's argument, then give a two minute concluding argument.

The computer couldn't go online but did store a bank of "hundreds of millions" of documents such as academic journals and newspaper reports. IBM staff drew up a list of 100 topics that might be covered by the databank and randomly selected two for debate. Neither the computer nor Ovadia knew in advance what they would be asked to talk about or which side of the debate they would present.

According to IBM, the computer had to pull off three tasks that aren't naturally suited to automated systems.

Firstly, it had to write the initial speech in natural-sounding language. (Source: ibm.com) Delivering the speech through a synthesized voice was relatively simple by comparison, as was recognizing the human voice and turning the audio into text for analysis.

The second challenge was the ability to process the text to find the key arguments made by the speaker.

'Rules Of The Game' Unclear

Finally, the computer had to figure out how to respond and address the points in a persuasive manner. That was tricky, as debating isn't so much about objectively correct or incorrect points, but instead often involves appealing to a mix of human logic and emotion.

There were some glitches with the computer repeating some arguments and occasionally using clunky language, but an audience poll eventually crowned the computer the winner. (Source: bbc.co.uk)

In the long term, IBM says the technology could be developed so that the computer system can look at a topic and try to analyze the arguments on both sides of the debate. That might throw up conclusions for any for humans who approached it with their own preconceptions and biases.

What's Your Opinion?

Is this a worthwhile use of technology? Are you surprised by the result? Could a computer adapt to election-style debates where candidates aren't always necessarily taking completely opposing positions on every topic?

Rate this article: 
Average: 4.7 (6 votes)

Comments

Dennis Faas's picture

Where I can see this technology really shining is in the diagnosis of patients in a hospital, or perhaps in a telemedicine application whereby smartphone users could ask a query (by speaking to the phone), and then have the IBM computer respond back in their language with a highly probable answer. Now that would be impressive.

Google Android smartphones and Apple Phones can process speech for simple tasks. This can be very helpful if you need to be hands free or don't want to be bothered typing in a query.

For example, when I've opened the lock screen on my Samsung Galaxy S6 Edge, I can press and hold the main menu button and the Google Assistant appears. I can say "set an event for 3 days from now to check car oil levels" and it will put it in my Google calendar or my email calendar. You can also send text messages in the same manner. "Send a text message to Stephen", and then it will say "OK", then ask you what you want the text message to say - all without having to type anything.