Theoretical Speed of Computers

In this episode, Byron talks about how machine learning systems compete with and teach themselves and are thus evolving ever more rapidly. For more on Artificial Intelligence: https://voicesinai.com https://gigaom.com https://byronreese.com https://ift.tt/31WftGA... Transcript: In 1983, the movie War Games came out. I was 15 years old at the time, and I saw the movie then, and have not seen it since, and that was, oh I don’t know, over 35 years ago. The premise, as you might remember, is that a computer is contemplating launching nuclear weapons to win a war game. Matthew Broderick, our star and hero, comes to the conclusion that he should play the computer in the game of tic-tac-toe a number of times to model for the computer something that it had not considered, a game that cannot really be won. At first, he does this by playing the computer himself, but he can’t play fast enough, and the launch sequence is counting down. So he had the computer play itself over and over at the speed of light. My memory of the quote of the computer when it comes to its pivotal conclusion, was that it was quote ‘an interesting game, the only way to win is not to play’ and in writing this video, I looked up the actual quote to find that the computer says it is a strange game and the only winning move is not to play. Interesting that the idea, that one idea stuck in my head for over 35 years. I bring all of this up right now because that’s the way we’re training AI right now, whether through adversarial networks or other systems, where we’re getting the AI's to either play themselves or play each other - to evolve better systems. Nowhere is this more evident than with AlphaGo Zero, the system that replaced AlphaGo after its spectacular win over Lee Sedol in the game of Go. AlphaGo Zero was not trained on Go with human games, but just played against itself. The implications of this are profound. When we see that learning systems are learning ever faster and that is, in part, because we’re getting better at training them, and that’s becoming more refined. And it’s mainly because our computers are growing ever faster. In spite of what’s recently been written about the limits of computer chips in their current forms, the theoretical speed of computers is vastly beyond what we are currently experiencing, and this, to my mind, is our best shot at making dramatic, new advances in AI. https://ift.tt/340vNXe gigaomhttp://ifttt.com/images/no_image_card.pngNovember 14, 2019 at 06:34PM https://ift.tt/eA8V8J