The future we will be like training a puppy tuning computer

Before the invention of the computer, experimental psychologists believe that the brain is an unknowable black box.play Frozen Games You can analyze the behavior of an object - such as the bell rang, the dog will salivate - but how do you go analytical thinking, memory and mood it? These things mysterious, beyond the scope of scientific research. Thus, behavioral scientists to study a range framed in the stimulus and response, feedback and reinforcement, and did not try to understand the internal mechanisms of the mind, this period lasted for forty years. Then, in the mid-1950s, a group of psychologists, linguists, information theorist and early AI researchers have proposed a different concept. They believe that man is not just a collection of conditioned reflex, but also to receive information, process it, and then act accordingly. People have a system that can write information to memory, stored in memory, and call information from memory. This is achieved by a logical, formal syntax performed. The brain is not a black box, it is more like a computer. This "cognitive revolution" is the beginning of budding little bit, but as computers became standard in laboratories across the United States psychology, "cognitive revolution" gained wide recognition. By the late 1970s, cognitive psychology subversion of behaviorism, it uses a new language to describe the mental activity. Psychologists began to thinking likened program, ordinary people have begun to use "memory bank" like metaphor. Control of the code, control the world? The digital revolution penetrated into our lives, but also penetrated into our language, and our basic views on the matter being. Science and technology is one such. In the Age of Enlightenment, Newton and Descartes to inspire people to think of the universe is an elaborate clock. Industrial era, with the piston machine gave rise to enlightenment, psychodynamic Freud would draw the steam thermodynamic mechanism. In this day and age, computer modeling from a fundamental mindset of the people, because if the world is a computer, then the world can be programmed. The code has a logic, it can be modified. This is the core principle of the digital age. Software is everywhere, as a venture capitalist Mark Anderson (Marc Andreessen) said, our exposure to the jungle machinery, machine our behavior, thoughts and emotions converted into data - and data is a raw material, can be used for coding engineers carried out. We look upon it as a regular thing, it is a series of rules can be found, utilization, optimization, and even rewrite instruction. Technology companies use the code to understand our most closely linked, even some inspirational article also said that you can modify your own source code, and for your love pattern, sleep habits, consumption habits and reprogramming. In this world, the ability to have more than just an ideal programming skills, but also become a language, if you speak the language, you are a "insiders", with the sexually explicit avenue. "If you control the code, you control the world," futurist Marc Goodman (Marc Goodman) said. Paul Ford (Paul Ford) wording is more cautious: "control code even if the person does not control the world, and also control the things you can control the world). Now, whether you like it or hate it situation, whether you're familiar with the programming, do not obsessed with it. Since the beginning of our machines speak a completely different language, and even the best programmers can not fully understand the language. "Machine learning" and Deep Neural Networks Over the past few years, Silicon Valley's top tech companies began to develop a computational method called "machine learning."play Barbie Cooking Games The traditional way is to write a program step by step instructions, let the computer comply. But in machine learning, programmers do not write instructions for the computer, but the computer training. If you want to church neural network kitten, you do not tell it to find a beard, ears, fur and eyes. But to show it a lot of cat photos, eventually it will be able to learn. If it is the fox wrongly classified as a cat, you do not need to rewrite the code, as long as it can continue training. This practice is not new, there will be a few decades ago, but recently there have been leaps and bounds, thanks in part to the rise of deep neural networks. Deep neural network is a large-scale distributed computing systems that mimic brain neurons multilayer connection. We have a lot of online activities are based on machine learning backing, such as Facebook use it to determine what content appears in your news stream, Google (microblogging) Photo Services use to identify faces, translation Microsoft Skype also uses machine learning, real-time converting content into different languages. In addition, unmanned vehicles also use machine learning to avoid accidents. Even the Google search engine also has begun using deep neural networks: the company was appointed in February this year, machine learning expert John(John Giannandrea) as the search for department heads. Google has launched a major program for qualified engineers to master these new technologies. "By learning to build the system,"he said, "we do not have to write the rules." End of an era The issue here is: the use of machine learning, computer engineers never know how to complete the task. Mechanisms of neural network is largely opaque, mysterious. In other words, it is a black box. As these black boxes begin to take on more tasks daily figures, they will not only change the relationship between us and technology, but also change the way we look at ourselves, look at the world, and its own way in the world position. If the former, the programmer is like God, developed a computer system to run the rule, so now, they are more like parents or dog trainer, this is a more elusive relationship. Andy Rubin (Andy Rubin) is an experienced programmer, participated in the Android operating system structures. "When I entered this line of computer science is still very young ...... it is a blank canvas that I can create something from scratch," he said. "For many years, this has brought me a great sense of control." But now, he says, this era coming to an end. Rubin interested in machine learning, his new company Playground Global is the field of machine learning start-up companies, the main popularity of smart devices - but these things make him a little sad, because the machine learning has changed "when an engineer" connotation. "People do not write programs with a linear fashion," Rubin said. "Neural networks learn later how to make voice recognition, the programmer can not see what's deep inside. It's like your brain as you can not cut off a man's head to look at what he was thinking." If the engineer determined to see what deep neural networks is that they will see is a sea of ​​mathematics: a huge, multi-layered structure calculus problem through continuous billions of data points derived relation between the results of the world guess. Just a few years ago, the mainstream AI researchers also believe that, in order to create intelligent, we must instill the correct logic to the machine. Prepared to wait until a sufficient number of rules, we will ultimately create a kind of sophisticated enough to understand the system of this world. They largely ignored machine learning some early supporters, these supporters believe that the data should be poured to the machine, let them draw their own conclusions. For many years, the computer has not been developed to the extent strong enough to really prove the merits of these two methods, so this argument has become a philosophical proposition. "Most of the debate have been based on some of the firm belief: how the world should be organized, how the brain works," Google's driverless car research and development, the former Stanford University professor of artificial intelligence Sebastian Thrun (Sebastian Thrun) said. "Neural networks do not sign, no rules, only numbers. This makes a lot of people at arm's length." Programmers do not unemployment A non-analytical machine language, this is not just a philosophical idea only. In the past two decades, learning program has been one of the most reliable ways of employment - so some parents that their children go to after school programming remedial classes. However, deep machine learning neural network is connected to another world,play Angela Games another employee is needed. Analysts have begun to worry about the job market will have artificial intelligence and what impact the. Like some of the equipment after the invention, a number of jobs before the meaningless, the programmer may soon enjoy the taste of this. When asked about this change, Tim O'Reilly (Tim O'Reilly) said that the traditional coding will not completely disappear, but in a very long period of time, we still need programmers - but needed The number may be reduced programming skills will be one yuan (meta skill). According to Allen Institute for Artificial Intelligence (Allen Institute for Artificial Intelligence) O'Neill, CEO Oren (Oren Etzioni) saying, machine learning needs "scaffolding" to run, and set up "the scaffolding" needs to be programmed. Not because of quantum mechanics, Newtonian mechanics will be abandoned. Programming will continue to be a powerful tool to explore the world, but other people may also need more tools. However, the specific functions of the drive, the main work is done by machine learning to perform. Of course, people still want to train these systems. This work requires both mathematics have a very deep understanding, but also need to have a teaching on intuition. "It's almost like an art, these systems lead to the best side out," Google DeepMind AI team leader Dai Mies Hasa Bies (Demis Hassabis) said. "There are only a few hundred people can do it well. World." However, even such a small number of people, but also enough to give just a few years the technology industry has brought change. Influence of Culture Whether this change will impact how employment, impact on culture will certainly be even greater. If human beings write software engineers led people to worship, so that we feel that the human experience can eventually reduced to a series of instructions can be understood, then began learning machine in the opposite direction to promote. The operation of the universe laws, may not be of human analysis. European antitrust investigation, said Google's search results exerted undue influence. However, even if the company's own engineers could not say exactly how search algorithms play a role, then such allegations will become a legal case without a head. Uncertainty is not news, even simple arithmetic, may produce sudden and unpredictable behavior - this argument can be traced back to chaos theory and the random number generator. Over the past few years, as networks become increasingly intertwined, functions become more complex, the code seems increasingly like a god alien behavior of the machine becomes more elusive, difficult to control: the stock market appeared a series can not prevent sudden collapse; inexplicable blackouts occur. Due to the rise of these forces, technical expert Danny Hillis (Danny Hillis) announced the "Age of Enlightenment" has ended. For centuries, our logic, certainty and control over natural full of faith, this era is now over. Hillis said, "entangled age" (age of Entanglement) began. "We created something in the techniques and mechanisms become more complex, our relationship between them has changed," he wrote in "Science and Design" (Design and Science) magazine. "We are no longer the masters of what we create, we learn to negotiate with them, cajole and guide them forward toward our direction. We have created our own jungle, they have their own life." In this article on the road, machine learning is the rise of a new development, just might be the last one. The prospects worrisome? This may make people feel terrible. After all, after ordinary people participated in short courses, how many will have some programming skills. Programmer or at least human. Now not only scientific and technological elite circle narrowed, and create something for themselves, their control also reduced. Companies create these things that they find difficult to control. Last summer, Google's photo-identification engine picture the black flag gorilla. In apology while, immediately took a correct approach: let the system do not put anything marked as gorillas. Some people think that this means that the machine will take away the authority of the era of human arrival. "It is conceivable that the technology to overcome the financial markets, the researchers better at than human inventions, more adept than humans leader in surgical manipulation, but also developed a number of weapons we can not even understand," Stephen Hawking wrote, " Although the short-term impact of AI depends on the control of its people, but its long-term impact will depend on whether it can be controlled in the end. "elon Musk and Bill Gates, who agree with him. But it should not be too afraid. And we are just beginning to learn a new skill, "entangled" rule. Currently, engineers are studying how the process of deep learning system's visualization. But even if we can never fully understand the idea of ​​these new machines, this does not mean we will not do anything in front of them. In the future, we will not be too concerned about the root causes of their behavior, but to learn to focus on their behavior itself. The importance of the code will be reduced, we used to train its data will become more important. Regain behaviorism You may think this seems a bit familiar, it is because it is the 20th century and is very similar to behaviorism. In fact, the process of training a machine learning algorithm is often likened to the early 20th century, a great behaviorist experiments: Pavlov with bells let the puppy drool, it is not derived from a deep understanding of hunger, but over and a play Dora Games routine over again. He provided data over and over again, until the code rewrite itself. Whatever your views on the behaviorist, who is to have the ability to control subjects. Thrun said that in the long run, machine learning will bring a democratizing influence. Like now you do not need to learn HTML website will be able to build, finally, you do not need a PhD, you can take advantage of the enormous power of deep learning. Programming is no longer the exclusive territory mastered a series of mysterious language programmers. As long as you have had puppies church roll, you can do the job. "For me, this programming is the coolest thing," Thrun said, "because anyone can be programmed." In the history of computing, how machines work, many times we have adopted a perspective from the inside out. First, we write the code, and then express it by machine. This outlook suggests a plasticity, but also reflects a decision based on the rules, in a sense, the underlying instruction is everything. Machine learning on the contrary, it represents a Perspective from the outside: Not only code determines behavior, behavior also determines the code.