AlphaGo Turns 10: How DeepMind’s Breakthrough Set the Stage for AGI

While generative artificial intelligence (AI) gained mainstream attention after the launch of GPT-3 in 2020, the underlying technology, machine learning, is decades old. From typing prediction in keyboard apps to recommendation algorithms in social media platforms, all rely on the same technology and have contributed to what AI has become today. One definitive precursor of modern-day AI is said to be AlphaGo, a system created by Google DeepMind in 2016. In fact, it showcased a single moment which convinced researchers and domain experts that artificial general intelligence (AGI) is possible.

AlphaGo and the AGI Moment

Marking the 10th anniversary of AlphaGo, DeepMind CEO Demis Hassabis penned a post about the impact of the technology on today's AI and the path towards AGI. AlphaGo is an AI system designed to play the Chinese board game Go autonomously. The two-player game is played on a grid of 19 x 19 lines (361 intersections) with identical black and white circular stones which are placed on the intersections. The objective of Go is to surround more territory than your opponent.

In March 2016, AlphaGo faced the 18-time world champion Lee Sedol in Seoul, South Korea. In Game 2 of the four-match tournament, AlphaGo did something unprecedented. In the 37th move of the game (now famously known as Move 37), it made an unconventional play that commentators initially thought was a mistake. However, by the end of the game, it was understood to be a decisive move that allowed the AI opponent to win the game.

“It was a display of incredible foresight and the AI system's ability to go beyond mimicking human experts and find entirely new strategies,” Hassabis said. This exact move is also considered a preview for AGI. The biggest determinant of the yet-to-be-achieved technology, which is said to match human-level intelligence, is innovation. In other words, the ability to be truly inventive and creative and not just regurgitate the knowledge it has acquired via its dataset.

What is interesting about AlphaGo is that it was built using techniques that are still being used to develop AI models — deep neural networks combined with advanced search and reinforcement learning. Hassabis revealed that today's Gemini AI models rely on these foundational techniques.

But the path towards AGI is much more challenging. The DeepMind CEO acknowledges that one of the key barriers researchers need to solve is making an AI model understand the physical world, and have it use specialised tools to take actions in the real world. This combination is what Hassabis thinks will crack AGI.

“Move 37 was a glimpse of AI's potential to think outside the box, but true original invention will require something more. It would need to not only come up with a novel Go strategy, as AlphaGo impressively did, but actually invent a game as deep and elegant, and as worthy of study as Go,” the DeepMind CEO said.