AlphaGo Zero required no human input to learn how to play Go, and beat every single AI on its path to become the ultimate player. Source: DeepMind

The DeepMind team published a paper in the journal Nature this Wednesday detailing the journey of AlphaGo Zero, the latest AI to beat all expectations about machine learning. The Google AI reached the status of the strongest Go player in history in just 40 days with no human input.

The new machine represents a significant achievement for the field due to the fact that it was essentially self-taught. In this case, this also translated to far more efficient algorithms and less resource consumption, which also meant more all-around efficiency.

AlphaGo Zero is but a demonstration of what artificial intelligence can achieve, now mostly on its own. Researchers at DeepMind believe that this new breakthrough means their machines are ready to tackle more real-life challenges in the field of medicine, for example.

How did AlphaGo Zero beat everyone on its own?

With AlphaGo Zero, Google’s AI division decided to take a different approach to see how it would do under a new model of reinforcement learning for machines of its kind. This involved feeding it just the rules of the ancient Chinese game and let it figure out how to play.

As it became its own teacher, AlphaGo Zero showed a striking resemblance to the learning process of novel human players. Just a few hours in, it acted like most people who start playing Go and focused on capturing as many stones as it could. A couple of hours later, it was already figuring out its own strategies.

By the time the AI reached the 72-hour mark, it had already surpassed the abilities of AlphaGo Lee, one of its most notorious predecessors for being the first machine to beat a human Go champion, Lee Se-Dol in 4 out of 5 games.

21 days later it had bested AlphaGo Master, the version who recently defeated the current champion Kie Je and 60 other Go professionals. In 40 days it surpassed all other AIs and became, possibly, the best player in the world.

Science is the next frontier for DeepMind

The scientific field remains one of the most challenging areas there is, and in which some of the most pressing matters of today and tomorrow are found. Even with the application of today’s most groundbreaking technological advancements, we as a species always have something to look forward to.

DeepMind believes that new AlphaGo cousins are ready to see some real-life action to help scientists make new discoveries. Researchers mention how the machine has shown it could be capable of discovering new materials and designing new drugs in the field of medicine.

Of course, some people are always wary of the idea of a machine that can think for itself. AlphaGo Zero can learn from its own experiences, but the team at Google’s DeepMind says they are still a long way from actual artificial intelligence as most people conceive it.

Source: DeepMind

LEAVE A REPLY