AI and multilevel neural network are nothing new. In 88 when I was student , Yann Lecun was a reference in the domain... Older than Cold Fusion But the size of the network and the data were too small. Internet also overtake the priority on AI, Expert System, neural network, Natural language Processing, and it was followed by Grid computing, Mobile internet... Statistic translation also But in fact neural network was discretely integrated in many niche programs, and deep learning get back in fashion because it could manager, and worked better, with huge data. I've met a guy using deep learning, to detect pictures of Luxury good from French Brand, bought for cheat in France and sold on Alibaba in China... Facebook is known to detect nudity or forbidden image that way... with some well known mistake, that i find quite logical. Google plan to start that to organize anti #fakenews censorship (down indexation)...
All that could be fantastic if it was not centralized, if you could choose your censorship ruling, your nudity preferences. I see two moves for AI evolution. One, leading to anthropomorphic/animality, is integration of motivation and feeling/mood into AI in physical devices... the famous Chicken bot of Jed. Another move is the distribution. AI have the power to be more perfect than human, so it must stay associated with humans, empowering people, not controlling them. the nightmare could be what I see as Google/facebook AI filtering, which could be more efficient, blind, and totalitarian than the Catholic Church and it's index. an AI can implement the Law, and all the Law, and this cannot work. A, AI designed by Caltech could have made Cold Fusion disapear on earth like some demand Climate skepicism be banned, and we try to ban nazism or hate speech. "Hell is paved with good intents". I prefer an imperfect earth, and a perfect hell. In fact like there is many people, there should be any AI, competing, associated, specialized, with different approach, different tastes, different strategies, and if possible educated differently by different people, different culture. An AI should be always in competition with dissenting AI, a rule of free market. No monopoly in ideas, in market, or in AI. 2017-03-16 15:31 GMT+01:00 Jed Rothwell <jedrothw...@gmail.com>: > Bob Higgins <rj.bob.higg...@gmail.com> wrote: > > That is the problem with the work of Futurists - many of the massive >> changes in our lives comes from seminal inventions whose timing cannot be >> predicted. Once that seminal invention is proved, progress from >> engineering can be rapid, or can be slow, but it usually moves forward. I >> think LENR is still in need of at least some seminal understanding that is >> presently missing. >> > > Definitely it needs this! > > > I believe AI is in a similar state of waiting for that seminal invention >> that makes AI practical. >> > > I think the breakthrough has already come. In the last few years, > tremendous progress has been made with multilevel neural networks. This > technique that made it possible for a computer to beat the world's best go > player, and it has recently had a tremendous impact on Google translate, > making it far more like a human translator. This all happened in the last > several months. > > Neural networks were proposed in the 1950s. A lot of work was done on them > in the 1960s, but not much progress was made. Computers back then were not > powerful enough to implement an effective version of these networks. I > think they had roughly as much computing power as an insect brain. The > biggest computers today have roughly as much power as a bird or mouse brain. > > This article shows an interesting animation: > > http://www.motherjones.com/media/2013/05/robots- > artificial-intelligence-jobs-automation > > Scroll down to where it says, "how long until computers have the same > power as the human brain?" They are predicting this will happen around the > year 2025. > > As I said, modern neural networks are multilevel, meaning one network > interfaces to another, which goes to another, and so on. This is called a > deep neural network. The original networks were single level. This does not > work anywhere near as well. The program that won at go has billions of > individual decision points (artificial neurons), as I recall, in two main > deep networks, policy and value. Twenty years ago, it would have taken > weeks or months to run such a gigantic program. > > https://gogameguru.com/i/2016/03/deepmind-mastering-go.pdf > > https://www.tastehit.com/blog/google-deepmind-alphago-how-it-works/ > > - Jed > >
