Humans just won IBM’s six-year-old AI debate

Natarajan and IBM’s AI system, Project Debater, began by preparing the debates and arguments against the resolution, “We should subsidize preschool.” Both sides had only 15 minutes to prepare their speeches, after which they proposed an opening statement. which lasted four minutes, a four-minute rebuttal, and a two-minute summary.

The winner of the event is determined by Project Debater’s ability to convince the audience with convincing debates. But even though Natarajan was declared the winner, 58% of the audience said Project Debater “enriched their knowledge significantly on the topic at hand, compared to Harish’s 20%”.

The results are displayed in real-time online polls.

Debate Project IBM engaged in the first ever live, public debate with humans in June 2018 when it debated the topic of whether or not to subsidize space exploration.

Project Debate is credited to IBM as the next big step in AI, having been in the works for almost seven years. It is taught to discuss unfamiliar topics, as long as these are well covered in the huge corpus of the mine system – which includes hundreds of millions of articles from many popular newspapers and magazines. The system uses the Watson Speech-to-Text API (application programming interface).

The international IBM Research Group led by IBM’s Haifa, Israel laboratory gave the project discussion three powers. First, data collection in text and speech delivery. Second, listening comprehension that can identify important claims hidden in long spoken language. And third, modeling human confusion in a unique knowledge graph to enable principled discussion.

But what is interesting about machines beating humans in debates and games, apart from showcasing the capabilities of technology companies?

Consider this development:

Ten years ago, IBM’s supercomputer Deep Blue defeated then-world chess champion Gary Kasparov.

■ In March 2016, AI company Alphabet’s DeepMind computer program, AlphaGo, beat Go champion Lee Sedol.

■ On December 7, 2017, AlphaZero—modeled on AlphaGo—took just four hours to learn all the chess rules and learn enough of the game to defeat the world’s strongest open-source chess engine, Stockfish.

The AlphaZero algorithm is a general version of the AlphaGo Zero algorithm. It uses reinforcement learning, which is an unsupervised training method that uses rewards and punishments. AlphaGo Zero does not need to train human hobbyists and professional players to learn how to play the Chinese game of Go.

Further, the new version not only learned from AlphaGo—the world’s strongest player in the Chinese game of Go—but also defeated him in October 2017.

Moreover, in July 2018, AI bots beat humans in the video game Dota 2. Published by Valve Corp., Dota 2 is a free-to-play multiplayer battle arena video game and is one of One of the most popular and difficult e-sports games Professionals train all year to earn a share of Dota’s $40 million prize pool, the largest of any e-sports game. Therefore, this type of player beating machine underlines the power of AI.

AI bots, though, have lost professional players in Dota 2, which has been actively developed for more than a decade, with the logic of the game implemented in hundreds of thousands of lines of code. This logic takes milliseconds to execute, compared to nanoseconds in the Chess or Go engines. The game is updated once every two weeks.

IBM’s Project Debate and AI bots lost out to humans, but given the lessons learned from DeepMind’s AlphaGo, this isn’t the last we’ll hear of AI-powered machines. The game has just begun.