Google unveils a superior AI capable of self-learning 'true meaning'

Called AlphaGo Zero, Google's newly revealed AI has the ability to independently learn to play Go, gain experience by playing with yourself , rather than using data from professional players. like previous generations.

Better than the best

Zero taught himself how to play Go by playing with himself.

We already know that Google is one of the most prominent technology companies in the race to develop artificial intelligence. As the months went by, their AI division, DeepMind, constantly showed great progress on AI capabilities, especially when referring to AlphaGo.

This AI of Google became famous all over the world after becoming the master of the Chinese go game, then defeated the 18-time world champion Lee Se-dol, which only happened. out last year.

Since then, DeepMind has begun to plan to add imagination to its AI, and they also use other games to teach AI how to manage tasks better. AlphaGo continues to beat the current world number one player at the moment Ke Jie, thereby showing his almost unlimited learning potential.

Recently, a new development plan has been "brought to light" by Google .

According to Futurism, DeepMind introduced this new generation AlphaGo this week, named AlphaGo Zero , saying it "is even more powerful than before and can be considered the greatest chess player in history". According to the company, Zero taught himself how to play Go by playing with himself - while previous versions of AlphaGo learned by studying the data collected from human players.

Zero began the process of learning as a beginner, but after only three days it defeated the AlphaGo version that won Lee Se-dol. After 40 days, Zero won the AlphaGo version that defeated Ke Jie.

In his press conference, DeepMind shared: "AlphaGo Zero has also discovered new knowledge, developed unique and creative tactics, acquired and improved the steps that its previous versions. each show during matches against Lee Se-dol and Ke Jie ".

Step outside the Go board

While AlphaGo Zero's ability to play Go is highly commendable, it should be noted that Go is very different from completing other events with more events. According to Eleni Vasilaki, a professor of neurology at Sheffield University, "AI fails in jobs that are easy for humans. We just need to look at the ability of robots in everyday tasks like walking. , run and kick a ball ".

In order to be on par with people at complicated tasks, AI still has a long way to go - even the AI ​​like Siri or Google Assistant cannot pass the knowledge of a 5th grade child.

But that does not mean that this is not revolutionary progress.

DeepMind's CEO, Mr. Demis Hassabis is fully aware of the difference between people and AI, that AlphaGo's growth and development are more important than becoming masters of an ancient game. According to him, this is "a big step for DeepMind in building common algorithms".

Over the next decade, Hassabis believes that Who will be able to work with people to promote development in areas such as science and medicine - even as much medical progress has emerged in current time.

For example, AlphaGo Zero is currently trying to figure out how proteins fold into three-dimensional structures, which, if successful, could accelerate drug discovery (the process by which we find those new medicine). Since then, we will be able to save countless people and bring humanity into a new medical era.

Mr. Hassabis shared: " I hope that algorithms and things inspired by AlphaGo in the future will go hand in hand with us as experts in science and medicine to bring us to new horizons. ".

Artificial intelligence may not be as smart as we are, but AlphaGo Zero is a sign, that AI learns very fast. Perhaps the biggest problem now is what we will have to do when the AI ​​day takes over all the work - an inevitable future - comes.