DOWNLOADDownload Free Guide Book [7 QUICK TIPS FOR SURVIVING A DISASTER]

Alphago – In case you live in real world, mankind just had a new virtual threat!

Google’s AlphaGo beats Lee Se-dol to win the AI vs Human Go Board Game Series 4-1

The game of Go came from China over 2,500 years ago, it was considered at the time one of the four essential arts that was required to be a true Chinese scholar.  Go game is Played by over 40 million people worldwide nowadays.  The rules are very simple:  each game has two players,  each player takes turns to place his/her “stones” on the game board, trying to capture the opponent’s stones, or surround empty space to make territory.  The game is played primarily through intuition and feel, and because of its beauty, subtlety and intellectual depth it has captured the human imagination for centuries.

 

shutterstock-go-board

 

But as simple as the rules are, there is significant strategy involved in the Go game, and the number of possible games is vast (10761 compared, for example, to the estimated 10120 possible in chess),  making it an extremely complex game.  To put it visual, 10761 looks something like 1,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000, 000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000, 000,000,000,000,000,000,000,000,000,000, 000,000,000,000,000 – that’s more than the number of atoms in the universe, and more than a googol times larger than chess.

 

 

Humans have been developing Artificial Intelligence (AI) for a long time.  The first game mastered by a computer was noughts and crosses (also known as tic-tac-toe) in 1952.  Then fell checkers in 1994. In 1997 Deep Blue famously beat Garry Kasparov at chess.  It’s not limited to board games either—IBM’s Watson [PDF] bested two champions at Jeopardy in 2011, and in 2014 Google’s algorithms learned to play dozens of Atari games just from the raw pixel inputs.  But the complexity of Go game is what makes it hard for computers to play, and therefore an irresistible challenge to AI researchers, who use games as a testing ground to invent smart, flexible algorithms that can tackle problems, sometimes in ways similar to humans.

Google subsidiary DeepMind’s AI program AlphaGo has been in the works for the past two years. It relied on machine learning and neural networks to study a database of 100,000 human Go matches, then played against itself millions of times to reprogram itself and improve.  But only a few months back, Go had thwarted AI researchers; computers could still only play Go as well as amateurs.

However, in March 2016, this has changed, AlphaGo program triumphed in its final game against South Korean 9 Dan Go grandmaster Lee Sedol to win the 5-game series 4-1, representing a landmark achievement for an artificial intelligence program.

“This is definitely the moment when we realized computers could beat us at this game,” says Andrew Okun, the president of the American Go Association.

 

AlphaGo logo

 

The defeat of Lee Sedol, one of the world’s best Go players, by AlphaGo, looks like another milestone towards a world where computers can do almost anything a human can.  There are uncountable things that only a human can do, and that no computer seems close to. The problem is that the purely human things are not economically useful to anyone. The things that computers can be taught to do are by contrast economically fantastic. The most powerful programs in the world are no more human than a shovel is, or a nuclear submarine. They are not moral actors and they have no feelings. What they have is power, but this power is growing at a rate that should frighten us all.

What can computers do next?  Can we survival if they turn to humans? That is scary!

alphago

 

Source: Goolgel Official BlogThe Verge

 

Share this page with someone who would love it!
Share on Facebook0Share on Google+0Tweet about this on TwitterPin on Pinterest0Share on Reddit0

Comments

comments

 

Add a Comment

Your email address will not be published.