The history of AI research is littered with human vs. machine challenges. IBM’s Deep Blue was the first serious challenger, back in 1996. In 2019, IBM’s Project Debater challenge has raised the stakes quite a bit higher — taking the competition away from boards and strict rules to the more general and subjective field of human debate competition.
The first groundbreaking competition between man and machine, Kasparov vs. Deep Blue pitted IBM against the best living chess champion, Garry Kasparov. In 1996, Kasparov proved victorious, but Deep Blue nicked the first-ever game off a grandmaster in history. In the 1997 rematch, Deep Blue went a step further and won the whole thing. Next up was Jeopardy, in which IBM’s Watson pushed the competition between man and machine even further. It was a towering achievement for AI competitiveness.
More recently, Google/DeepMind’s Alpha Go & Alpha Go Zero took on an even more complex game than chess, Go, and dooooominated one of the top players in the world… and then crushed the best player on Earth. But, albeit complex games, they are indeed games — the rules are very specific and immutable, there is no subjectivity to the game, and the more mental horsepower/computational power you bring to the table, the better chance you will have to win it. When it comes to more ‘general’ artificial intelligence, like attempting to pass the Turing Test, AI to date hasn’t fared quite as well.
It was IBM in 1996 who had the gall to challenge the best player in the world, and it was IBM in 1997 who beat him. It was IBM who took down the best living Jeopardy champions in 2011. But besting a human at a board game or trivia gameshow isn’t nearly as complex as besting a world champion debater in a real-time debate.
IBM didn’t succeed in that endeavor — yet — but the result was inspiring nonetheless.
From Forbes, recounting how the competition played out:
But IBM’s “Project Debater” still made history as the first AI machine to make a persuasive argument about a given a topic it had not been programmed to learn.
Harish Natarajan holds the world record for most international debate victories. Monday’s debate was the most unusual—and perhaps the most challenging—he’s ever had. After all, Project Debater had access to 10 billion sentences in hundreds of millions of documents. Natarajan had no access to the Internet. He just had a pen, his notepad and his brain.
Each side—human and machine—had 15 minutes to prep for the debate. They were each presented with a topic: “Should we subsidize preschools.” Project Debater had to argue in support of it. Natarajan had to argue against it.
The machine did pretty damn well!
The humans in the audience ultimately gave the win to Natarajan (to win a debate like this, it rests on whichever debater changes the most audience members’ minds from their pre-debate opinion), but it marked the first time a machine was able to make a public and persuasive argument in a freeform format like this. No game show rules, no board game rules, no head start in learning a specific subject matter — just 15 minutes to prepare an argument whole cloth.
Natural language processing really is one of the holy grails of AI integration into modern society, and displays like this prove we’re getting closer and closer to that reality.