Read on Twitter
Morning guys! So continuing on the Artificial Intelligence discussion.
How far have we gone, on having intelligent systems? I'll explain in this thread.
Have ur Cup of coffee and be my guest....
How far have we gone, on having intelligent systems? I'll explain in this thread.
Have ur Cup of coffee and be my guest....
We can determine how sophisticated the AI systems are by analysing how well they play with humans in complex board game Go. It's a good choice because this is a non-chance combinatorial game that doesn't have dice 
and ur moves are seem by d opponent. Pure tactics & smartness
and ur moves are seem by d opponent. Pure tactics & smartness
It's a Chinese abstract-strategy game that dates back 2500 years. Enough to say- it's a very complex game.
Let me put it in context: The numerical estimate of moves in game far exceeds the number of atoms in observable universe!!.. Think about this for a while.
Let me put it in context: The numerical estimate of moves in game far exceeds the number of atoms in observable universe!!.. Think about this for a while.
It led to field in mathematics called Surreal numbers. Anyway, about 45m ppl worldwide can play Go and about a half play it.
Now imagine making a computer than can play this complex game. How to you tell the computer the rules and tactics to win.
Now imagine making a computer than can play this complex game. How to you tell the computer the rules and tactics to win.
Well, in 2015, AlphaGo (Google's Artificial Intelligence) became the first AI to defeat a Human Go champion. Not only that, it defeated the 5-times GO Champion Fan Hui. It did that by learning as they play and developing itself untill it became better than the world champion.
AlphaGo was able to learn from humans by playing, untill it's able to beat them. Now hear this, in 2017, Google made a master version called AlphaGoMaster. This version of the AI was not only capable of learning and defeating world champions, but it actually made non-human moves.
It fascinated everyone by making moves that were never thought of, and it won professional Go players 60 times straight!
Ke Jie said "Humans perfected our moves and tactics for over 2500 years, only for the AI to beat us within a month" with it's 'innovative moves'.
Ke Jie said "Humans perfected our moves and tactics for over 2500 years, only for the AI to beat us within a month" with it's 'innovative moves'.
We're not done here, it gets crazier.
By end of 2017, Google's team publish a paper in Nature, on an advanced version of AlphaGoMaster called AlphaGo Zero. It created this version without using any data from humans.! Basically a stand alone intelligence.
By end of 2017, Google's team publish a paper in Nature, on an advanced version of AlphaGoMaster called AlphaGo Zero. It created this version without using any data from humans.! Basically a stand alone intelligence.
It played by itself and taught itself, without ever playing or knowing a human being. After the third day of playing itself, it competed with AlphaLee and won 100-0. (just so you know, AlphaLee was able to beat the 18-time world champion, Lee, by 4-1). Mind blowing.
So this self-taught AI (AlphaGo Zero) was able to beat AlphaLee 100 times straight. And it reached the level of AlphaGo Master by day 21 of playing itself, and by 40, it exceeded every other known variant. It outsmarted every human and it's AI alike. The craziness didn't stop.
In early 2018, an improved version of AlphaGo Zero was made (called AlphaZero), it could master Go, shogi and Chess all together. Without any data or practice with human, but by practicing and playing itself. After 8 hours of practice, it surpassed its predecessor AlphaGo Zero.
The results of AlphaZero was published in journal Science last year December. Where it says AlphaZero had the ability of updating its Neural networks.
