Back in the late 18th century, a mechanical marvel in the form a robot that could outsmart a human opponent in a game of chess gained international fame.
The Automated Chess Player, or The Mechanical Turk, wowed audiences for over 80 years, defeating many challengers including Napoleon Bonaparte and Benjamin Franklin in games of chess.
This was later proved to be an elaborate hoax when it was discovered that a chess master was controlling The Turk from underneath a cloth, but ever since, there has long been a fascination with competitions between robots and humans.
With AI’s capabilities advancing every day, this fascination has not died down. Computer hardware company IBM has led the way in the form of its AI Grand Challenges.
Beginning back in the 90s, Deep Blue, a chess-playing computer developed by IBM, became the first computer system to win both a chess game and a chess match against a world champion after beating grandmaster Garry Kasparov.
More recently, humans and AI went head-to-head in 2011 when IBM’s question-answering computer system Watson competed on quiz show Jeopardy!, beating two former champions to prize money of $1m.
Debating, however, is a different ball game.
Requiring mastery of human language as well as skills in persuasion, winning an argument presents a unique challenge for AI.
Approved back in 2012 after an IBM scientist in Israel suggested that the company should build a computer that could have an argument with a human being, Project Debater aims to see whether AI can be used to build “well-informed arguments and make better decisions”.
IBM has developed the first ever AI system able have a full live debate with expert human debaters, and the latest version will be put to the test at IBM’s Think conference, kicking off in San Franciso this week.
The human taking on IBM’s Project Debater
With over ten years of competitive debating experience including the 2016 World Debating Championships Grand finals and winning the 2012 European Debating Champion, Harish Natarajan should make a worthy opponent.
This evening, Harish is due to go head-to-head with AI in front of an audience of hundreds for a debate that is being livestreamed around the world.
The State of Technology This Week
Speaking while in-transit to San Francisco, Harish explains how he first got involved with the project:
“I found out about the existence of the project in June or July as there was a demonstration of it and I think this made a lot of the world’s press so I had a little bit of knowledge about it. A friend of mine who I’d met through debating is now working with IBM on the project. A few months ago he messaged me and said that there was going to be this challenge and asked if I’d be interested in taking part.”
Project Debater was first unveiled in 2018 when it debated against two human debaters, Noa Ovadia and Dan Zafrir. On two topics: “We should subsidise space exploration” and “Should we increase the use of telemedicine.” The system was able to debate with its opponents and even crack jokes, winning the competition.
Neither Harish not Project Debater will know the topic for their debate until ten minutes before the competition, and as a result the competition is difficult to prepare for. Describing his knowledge of AI as “close to a novice”, this is the first time Harish will see Project Debater in action:
“It’s an odd experience. I’ve never seen the machine work so I only have the analytics preconceptions of what it’s like. So I have guesswork of how it’ll work, what it will be good at, and what it wont be good at. One of the consequences is it’s reasonably difficult to prepare for it.”
However, with ten years’ experience in competitive debating, Harish says that his preparation for the competition is largely the same as going head-to-head with a fellow human:
“There’s not a huge amount of different preparation I would be doing…in terms of actual material, the preparation is largely the same as I would for any major tournament, so I’m doing normal reading, I’m making sure I’ve been giving a couple of speeches so I’m not totally out of practice, so it isn’t particularly difficult, in part because I don’t know what the machine is going to throw out at me!”
With a database of hundreds of millions of articles at its disposal, the system is able to search through for relevant information from which it can then debate unfamiliar topics.
Human vs AI: the strengths and weaknesses in debating
Harish believes that both AI and humans have strengths and weaknesses when it comes to debating, as Project Debater is not yet able to directly engage with its opponent in a way a human could:
“My preconception is that it will be very good at coming up with relevant information and probably a lot more than I will be able to do. Both the machine and me will get the topics we’re debating ten minutes before the debate starts so I wont have had a great amount of time to come up with specific evidence, so it will certainly be in an advanced position at that level.
“What I assume humans will still be better at than machines is coming up with well-reasoned, logical arguments, and the human side of persuasiveness as ultimately debating is about persuading a human audience.”
He believes that where Project Debater may struggle is in constructing an argument:
“Convincing is only partly about the provision of facts. It’s also going to be able logic, it’s going to be about language and my assumption is that is where humans are probably in a better position than machines.”
Regardless of the outcome of this debate, the ramifications for the project are far greater, shedding light on whether AI can be a useful tool to aid human decision-making.
Harish believes the technology may be particularly useful in the policing of fake news:
“What strikes me about Project Debater is the immediate human ramifications of it in terms of what it can do for everyone is much greater. This is the ability to process a billion pieces of information, find what is credible and be able to present that. So while in some ways this event seems to be a contest, what it really is is showcasing the ability of AI in an area where its value to human decision-making could potentially be the greatest in that there is so much information out there so the ability to find it, find what is relevant and particularly nowadays what ends up being correct and what ends up being fake news, for the lack of a better term.”
It will also be extremely useful in informing research into “natural language understanding,” that is how well AI can understand and respond to human speech and may eventually be able to hold a natural conversation.
Future AI will be better at debating than humans
Harish is confident that in the future, AI’s capabilities will be such that it can win an argument against humans:
“There will come a point, I don’t know when that point is, maybe it’s this week, where AI is at least as able as the best human debaters. If not, I’m sure it will come at some point in the near future…certainly with the progress of AI and technology it is highly likely in the not too distant future.
“I think the application for what is underlying Project Debater in two or three years could be huge for human’s ability to make good decisions.”
However, when it comes to his own imminent debate, Harish is diplomatic in whether he thinks he will win:
“I can’t really answer that. Partly because I’ve never seen the machine and I guess like always as a human being you want to back yourself but the record of humans versus machines in this sort of thing hasn’t been great. I still think that debating at the highest level may still be an area where humans are able to beat machines but there is certainly more than a good chance that I am going to be surprised and I will be very impressed!”