Artificial intelligence into the public free, musk devil"
leifengwang· 2016-04-28 16:51:27
pictures: Michal Czerwonka / Redux
Friday afternoon usually is the prime time of bad news release, this is traditional politicians and capitalists love of public relations. Therefore, when in December last year, Tesla founder Elon Musk and Y combinator's founder, Sam Altman after a week of AI conference, at the end of Friday suddenly announced to set up a new AI company, we all feel strange.
but they were late to announce the news for a reason. Not because no one will notice, but because the attention is too hot. Some of the most powerful companies in Silicon Valley heard the sound of the wind, began the backbone, all INSERM, composed of an to fresh OpenAI throws a lot of checks to their possession. These last-minute check - there are some is opened in the general assembly, is too big that musk and Altman has to be put off to announce the new company. "The money is going to go crazy. "Zaremba Wojciech said that he had worked as an intern at Google and Facebook, and then joined OpenAI, and he received a high price job offer at the last second.
how much money is too much to go crazy? Two years ago, with the market of machine learning really began to hot, vice president of Microsoft Research, Peter Lee said, the cost of a top AI researcher let top quarterback in the national football league are eclipsed, said he is usually case, if Silicon Valley's two most famous venture to hunt your talent. That, of course, is another kettle of fish the. Zaremba said that with the establishment of OpenAI, other companies to give him a salary of two to three times the value of his market.
OpenAI did not give the same high salary. But company gives the other a temptation: a purely for the sake of future research opportunities, for any product or quarterly earnings and, eventually, to with any wish to obtain research results share most, or all of the research results. You heard us: musk, Altman and company want free may be in the 21st century to change the world of science and technology - free.
Zaremba said that those who are too crazy to pay the salary actually make him uncomfortable, even though he has a great respect for Google and Facebook. He felt that this money is to buy his human, but also in order to prevent the establishment of OpenAI, which makes him more like the OpenAI atmosphere. "I realized that OpenAI was the best place I could go. "Zaremba says.
this is the story the greatest irony: < / strong > even if the world's largest tech companies want to retain the researcher, is like the football team to retain their star player, the researchers themselves or to share < strong >. in the AI study of pure air, the most intelligent mind is not the next product cycle and profit space - at least, not just these. They want AI to get better, and if you turn the latest discovery into your own little secret, AI can't get better. < p > today, OpenAI will release its first batch of AI software, a build artificial intelligence system toolkit, using one of the key technologies behind a technique called "reinforcement learning" -- this is the beat chess champion of Google AlphaGo. with this tool kit, you can build a system to simulate the new robot generation, play Atari games, and - on, you can also - to become a master of chess.
but playing the game is just the beginning. OpenAI is a multi billion dollar project, in order to promote the future of AI. You can see a new wave of innovation from the way you created the company and the future. We still don't know if OpenAI will become the main force of this change. But with the thrust of this particular company, we can see that a new generation of AI will not only bring new technologies, but also a new way to build technology. Class= img_box "
pictures: OpenAI research director Sutskever Ilya. Photograph: Christie Hemm Klok / Wired
everywhere in the Silicon Valley is not what a rare thing. It is best to remain skeptical that anyone who is not in a good tone of voice. But in the field of AI, change is coming true. In the Google and Facebook, called deep learning technology has helped them achieve a variety of services, such as face recognition in the photo. This technique can not only allow the machine to complete the task, but also let them learn by doing. Some people believe that this will allow the machine to eventually acquire common sense, a ability to think like a human.
followed by a variety of concerns. Musk and Altman worry that if people can create a good AI, then people can also create a bad AI. It is not just the two of us who are worried about the robots that dominate humans, but perhaps this is contrary to intuition. They believe that the best way to fight against the "bad" AI is not to limit people's access to AI, but to expand it. this has attracted a group of super smart young people to their new project. The origin of
OpenAI in the last year of a summer night, in Silicon Valley rosewood hotel a private rooms, a farm style luxury hotel in the city, in Menlo Park, sand hill road, in the venture capital world center. Musk Elon and Sutskever Ilya eat together, the latter was working on Google Brain. Sutskever is one of the top scholars in the project. At that time, a bigger idea was in the bud. Y combinator in the
Sam Altman help such as airbnb, Dropbox and coinbase and so on firm growth, he arranged the meeting, several AI researcher and a young but actual combat experience of enterprises to build home Greg Brockman gathered together. Brockman is the Silicon Valley digital payment startups Stripe (also a Combinator Y company) CTO. This is a select group. But they all have a common goal: creates a new AI lab, which is not only controlled by Google, but is not controlled by any one. "I think the best thing I can do," Brockman says, "is to make people build real AI in a safe way. "
Musk comes because he is an old friend of Altman - and because AI is very critical for the future of his business, or that the key to the future is very important. Tesla needs AI to build it sooner or later. SpaceX, Musk's other company, needs AI to send people to space, and to make people live in space. However, Musk is also one of the biggest AI critics, warning people if the system is strong enough to be able to learn independently, and perhaps one day we will lose the ability to control the system. The problem with
is that the people who are most able to solve these problems are already working for Google (and for Microsoft and Facebook and Baidu and Twitter). That night, no one was sure that the scholars could be attracted to a new company, even if it was Musk and Altman behind the company. But a key figure at least remains open to job hopping. "I feel like there's a risk here," Sutskever said. "But I think it's a very interesting thing, it's worth trying. "
break the deadlock Brockman in the dialogue with Musk, Altman and other Rosewood people later,
strong guts, soon determined to build the laboratory they want. He did his best to conquer the project and went to find Bengio Yoshua. Bengio is a computer scientist at the University of Montreal, is also one of the founders of the movement of deep learning. Another two in the field - Hinton Geoff and LeCun Yann - are currently working in Google and Facebook, but Bengio is committed to the academic community, generally do not participate in the activities of the industry. He listed some of the best researchers in the industry, and in the next few weeks, Brockman contacted all the people on the list, and some of them were out of the list.
a lot of researchers like the idea, but they also worry about changing. In order to break the deadlock. Brockman chose 10 what he wants most researcher, invited their Saturday to a cellar Napa Valley wine and dine. For Brockman, even driving into the Valley Napa this section of the road is also a catalyst for the project. "People don't realize that this kind of situation can make people go to a great, because you can't make the car go faster. "He said," you have to go, so you have to chat. "When they reached the cellar, atmosphere is maintained. "You can know the atmosphere is great," Brockman said. Or according to Sutskever's words: "compared with the chat, the wine is second. "At the end of the Brockman,
invited all ten researchers to join the lab and gave them three weeks to think about it.". By the end of the day, nine of the researchers had been in, and they had been on the program, despite the high price paid by the Silicon Valley giants. "They're really going to leave me, so it's not a simple decision to quit. "Sutskever said of his former employer, Google, ," but in the end, I decided to go to OpenAI, partly because of the strong people, and to a large extent, because of the OpenAI's vision. "
deep learning movement originated from academic circles. Until recently, companies like Google, Facebook, and Microsoft did not use it, as a result of the advances in computing power that makes the depth of the neural network a reality rather than a theoretical assumption. Hinton and LuCun, who left academia to enter the business, because the enterprise has a large number of resources. But they also retain the willingness to cooperate with other scholars. Indeed, as LeCun has said, deep learning studies require the flow of this kind of freedom. "When you do research, you will fall behind. "LeCun says.
therefore, big companies now share a lot of their research. This is a big change, thanks to Google. Google has always been regarded as the core of the technology secret. Recently, Google will promote the neural network software to do the open source. But Google still retains the core of the future competition. Brockman, Altman and Musk hope to further promote openness, said they do not want to see one or two big companies control the future of artificial intelligence. He is one of the OpenAI CTO and founder of a
Greg Brockman. Image source: Hemm Klok Christie / Wired.
open to the limit
above all sounds great. But for OpenAI's idealism, a lot of researchers may still face the same compromise. There is a limit to the opening. AI's long-term consideration is not just the only consideration. OpenAI is not a charitable organization. Musk's companies can benefit from the OpenAI research, as well as the Y Combinator Altman support companies. "Of course there are some competing goals. "LeCun said," this is not a public service organization, but also has a close relationship with Combinator Y. And, as in other companies, people are paid to do the work. "
according to, the salary of the lab is not as high as Google and Facebook," said Brockman. But he said. < strong > laboratory do hope to researcher "good pay, but also provide equity < / strong >, first in the Y combinator, then perhaps the space X (unlike Tesla, SpaceX or an unlisted company).
despite this, Brockman insisted that OpenAI will not provide special treatment to the sisters. OpenAI is a research organization, he said, not a consulting firm. But under questioning, admits that OpenAI's idealism has its limits. may not have all the results open source, although the final hope will share most of the research, whether through research papers or Internet services. "Open to all research is not necessarily the best way. You need to develop an idea to see what it will develop into, and then publish it. "Brockman says," we're going to bring a lot of open source code. But we also have a lot of things that are not ready to be released. "
Sutskever and Brockman have added that OpenAI may also be a part of the success of the patent. "We won't make any patents in the short term," Brockman said. "But the change in our captain's strategy remains open, and if we think it's the best way to do that. "For example, he said that OpanAI might be the first to do a patent application, a strategy to prevent others from applying for a patent.
but for some people, the patent means that the profit motive - at least, compared to the founder of the open source vision, this motivation is even weaker. "This is the real meaning of the patent system," said Etzioni Oren, head of the Alan Institute for artificial intelligence. "It makes me wonder what they really want to do. "
when Musk and Altman for the OpenAI unveiling, they describe the project as a threat to neutralize the" malicious "super artificial intelligence. Of course, what they say is also likely to be born from the OpenAI technology, but they insist that any threat will be reduced, because the technology can be used by anyone. "We feel that it is more likely that many AI have come together to organize the occasional bad AI. "Altman says.
not everyone eats this set. Oxford philosopher Bostrom Nick warned of the dangers of AI, like Musk, who pointed out that, if you do not control the study, the bad guys will be able to get the technology before it is guaranteed to be safe. "if you have a button that can do bad things," Bostrom says, "you won't want to make it public, to give anyone the. on the other hand, if the OpenAI decided to keep research, not to let the bad guys come into contact, Bostrom and suspected OpenAI and Google or Facebook no difference.
he does say that the non - profit nature of OpenAI will bring change - though not necessarily. The real power of the project, he says, is that he does have a control over the giants such as Google and Facebook. "This reduces the likelihood of super intelligent being monopolized. He has reduced the possibility that a group can have a better AI than the others.
but, as the philosopher said in a new theory, a OpenAI like this is willing to share its successful organizations, its primary effect is to accelerate the progress of AI, at least in the short term. And in the long run it can also advance technical progress, if it is in a mutual aid, the open degree of choice than the commercial operation of the higher. "
", however, a research grant with a charitable purpose can be carried out in an open manner to promote technological progress. "
is like Xerox PARC as
1 Brockman at the beginning of the month, the 9 AI fellow in his San Francisco apartment. The project was so new that they didn't even have a white board. They bought a few white boards, and then began to work (can you imagine the picture?).
Brockman said OpenAI will begin to explore reinforcement learning, a way to make the machine learning tasks, by allowing the machine to repeat, to track which method can bring the best results. But another major goal is "unsupervised learning" - creating machines that can really teach yourself without human guidance. Today, deep learning is driven by a careful annotation of the data. If you want to teach a neural network to identify a picture of a cat, you have to give some examples to the system ". And the learning process is supervised by human beings. But like many other researchers, the neural networks that OpenAI wanted to create don't need to be labeled so that they can learn.
Brockman said: "if you have a very good unsupervised learning, the machine will from the knowledge of the Internet Learning -- like people, by around will learn knowledge -- or by reading books. "
he conceived the OpenAI, is the 70s science and technology research laboratory PARC Xerox modern incarnation. < / strong > like Parc most open and undisturbed study promote from the graphical user interface, laser printer to the object oriented programming language, Brockman and staff hope further in-depth before is considered to be the realm of science fiction. Yes, PARC is Xerox, but it also supports a number of other companies, including the most famous apple, because such as Jobs, who are involved in it. In Brockman, OpanAI wants everyone to take part in it. < p > this month to be the culmination of this power, Brockman and company seized several other famous researchers, including Ian Goodfellow, another Google brain team of former senior researcher. "The special thing about PARC is that they get a bunch of smart people together and let them do what they want to do.". Brockman said, you need a common goal, there is no central control. "
gives up control is the core of the open source. If enough people are committed to this common goal, the end result will be more than you secretly study results. But if AI became so powerful as we thought, things would change. We must make sure that the new AI will follow the same principles of mutual assistance, which will give birth to the principles of the world. Musk, Altman and Brockman put their hopes in the wisdom of the group. But if they are right, the group is not entirely human in the future. Wired