IBM HAS showcased its AI holding its own in a debate with a human.
Project Debater can call upon its library of research papers to create arguments and counter-arguments against human debate adversaries.
Demonstrating the process at its San Francisco office, a six-foot tall box with blue lights, debated two humans – Noa Ovadia and Dan Zafrir. It looks an awful lot, but not entirely like Watson.
These were no beginners debate topics either – “we should subsidise space exploration” and “we should increase the use of telemedicine” are certainly going to call on the research library.
The audience agreed that IBM Project Debater was nowhere near the standard of a human debater in terms of delivery, but when it comes to the sheer amount of data it could convey, it was better.
The AI was also more convincing and therefore caused more audience members to change their mind, in the second debate.
The Guardian notes, however, that in an audience made up of the press and some IBM employees, it could well be that it wasn’t a completely fair fight.
IBMs aims aren’t really based around Oxford-style debates (hey, we know all about them at INQ) but rather to build better virtual assistants that can actually collate and form a cogent argument, not just “play me a fart”.
One example could be boardroom arbitration, where a bot could listen dispassionately and give a decision based on the evidence, without being cluttered by biases or emotions.
The IBM bot did make a few mistakes, such as rambling on with irrelevant anecdotes, and in one case saying the word “Voiceover” when it read the transcript of a video.
Wait a minute – that’s not a robot – he’s Ron Burgundy?
Also noted was the fact that Project Debater has neither facial nor hand expressions and that is a guaranteed way to lose a lot of the sentiment behind statements. It’s worth noting that over 90 per cent of human communication is non-verbal, so there may be a bit more work to do. Maybe they could give Boston Dynamics a call? µ
Source : Inquirer