1. The era when large companies control AI can academic entrepreneurship be 'outside the army'?
'Netease Intelligence News July 1st news' In the field of modern artificial intelligence, all the topics seem to be inseparable from the three researchers who are connected with the University of Canada.
The first is Geoffrey Hinton, a 70-year-old Englishman who teaches at the University of Toronto. He pioneered a branch called 'deep learning' and is now synonymous with artificial intelligence. The second is the 57-year-old Frenchman Yann LeCun. He worked in the laboratory of Hinton in the 1980s and now teaches at New York University. The third is 54-year-old Yoshua Bengio, born in Paris, grew up in Montreal, and now teaches at the University of Montreal. These three are close friends. They are also collaborators, so that people in the field of artificial intelligence call them 'Canada Mafia'.
However, in 2013, Hinton joined Google and Facebook hired LeCun. Both maintained their academic status and continued to teach, and Bengio established one of the world's best artificial intelligence projects at the University of Montreal. Think it is the last academic purist.
Bengio is not a natural industrialist. He is very humble, has a little hunchback, and seems to stay in front of the computer screen for a long time. Although he provides consulting services for several companies and has been invited to join, Bengio insists Pursue passionate projects, not those that are most likely to be profitable.
'You should have noticed how far his ambitions are, how good his values are,' said his friend Alexandre Le Bouthillier, co-founder of an artificial intelligence startup called Imagia. 'Some people in the field of technology Forgot the human side, but Yoshua didn't. He really hopes that this scientific breakthrough can help society.'
However, Bengio's insistence has turned into a lonely effort. Large technology companies such as Amazon, Facebook, Google and Microsoft are investing in or acquiring new start-ups, and absorbing the top talents in the university to ensure they are in the field of artificial intelligence. Leadership. Pedro Domingos, a professor of artificial intelligence at the University of Washington, said that he asks academic contacts every year to ask if they have students who want to apply for a postdoc. He said that when he last asked Bengio, 'he said, I even graduated I haven't been able to keep them before. ' Bengio is fed up with this situation and wants to stop brain drain.
On a warm afternoon in September 2015, Bengio and his four best-skilled colleagues met at Le Bouthillier's home in Montreal. The conference was originally a strategic meeting of technology transfer companies founded by Bengio many years ago, but Bengio told him The future of the field is deeply anxious. He also thinks it is time to ask some questions he has been thinking about: Is it possible to create a business that can help startups and universities, rather than hurt them – and may benefit the whole society? If so, can this company be competitive in this technology-led world?
Bengio especially wants to hear from his friend Jean-Fran?ois Gagné, an energetic serial entrepreneur who is 15 years younger than Bengio. Gagné earlier sold his co-founded company to a company called JDA Software's company; After working there for three years, Gagné left the company and became a resident entrepreneur of Real Ventures, a Canadian venture capital firm. At the end of the three-hour meeting, when the sun began to descend, he told Bengio and other people present: 'Well, I want to make a business plan.
That winter, Gagné and his colleague Nicolas Chapados visited the Bengio at the small office of the University of Montreal.
He proposes to co-found a startup that provides artificial intelligence to startups and other under-resourced organizations and companies that are unable to build their own AI division. The key selling point of this startup will be that it has the most talented workforce in the world: It will support researchers at Bengio Labs and other top universities to work for the company for a few hours a month while maintaining their academic status. This way, companies can get top talent at a very low price, and universities can stay. Live their researchers, and their main customers will have the opportunity to compete with even more powerful opponents. Everyone will benefit, except for the tech giants.
1
When Bengio and Gagné started their discussion, the biggest technology companies were not involved in the high-profile AI ethics issues – about controversial sales of artificial intelligence in military and predictive policing, and race and other biases. The manifestation in the product - these problems will soon have a negative impact on them. But even so, insiders clearly know that large technology companies are deploying artificial intelligence to expand their strength and wealth.
Understanding this requires knowing that artificial intelligence is different from other software. First, there are relatively few artificial intelligence experts in the world, which means they can control their salary to six digits; this makes it possible to build a team of artificial intelligence experts. A large team is too expensive for everyone except the richest company. Second, artificial intelligence usually requires more computing power than traditional software, which can be very expensive, and good data is difficult to obtain. Unless you happen to be a technical cow, you can access both without limit.
Bengio said: 'Recently, there have been some problems with the way artificial intelligence works... This has put expertise, wealth and rights in the hands of a few companies. ' Better resources will attract better researchers, which will bring more Good innovation, bringing in more income, so that you can continue to buy more resources. He added: 'It seems to be able to grow up.'
Bengio's earliest contact with artificial intelligence heralded the rise of large technology companies. In the 1970s, he grew up in Montreal and especially liked science fiction, like Philip K. Dick's "Do Androids Dream of Electric Sheep?". Bengio majored in computer engineering; when he was a graduate student at McGill University, he accidentally read a paper by Geoff Hinton. He was shocked, which reminded him of the science fiction story he loved when he was a child. 'Hey, Scorpio! This is what I want to do,' he later recalled.
Years later, Bengio and Hinton and LeCun became important figures in the field of deep learning, a field involving computer models called neural networks. But their research is full of false beginnings and frustrated ambitions. Theoretically speaking Deep learning is attractive, but in practice no one can make it work well. 'For years, at machine learning conferences, neural networks were not popular, and Bengio insisted on his neural network. "Working around," recalls Mozer, a professor at the University of Colorado. "I thought, poor Bengio, he was too fluent."
In the late 20th century, researchers began to realize why deep learning was not effective. Training high-level neural networks requires more computing power than before. In addition, neural networks require good data to learn, and before the rise of the consumer Internet, They don't have enough data to learn.
By the end of the 20th century, everything had changed. Soon, large technology companies began applying the techniques of Bengio and his colleagues to achieve commercial milestones: language translation, speech recognition, face recognition. At the time, Bengio's brother Samy also Working at Google, he is also an artificial intelligence researcher. Bengio wants to go to Silicon Valley with his brother and colleagues, but in October 2016, he, Gagné, Chapados and Real Ventures launched their own startup: Element AI. Matt Ocko, managing partner of DCVC, which invests in the company, said: 'In the past five years, Bengio has no substantial ownership on any artificial intelligence platform, except Element AI. ' 'He supports this home with his own reputation the company. '
In order to win customers, Element AI relies on the star effect of its researchers, its investors' reputation, and commitment to provide personalized service more than large technology companies. But the company's executives also stand on another perspective: In this era, Google competes to sell artificial intelligence to the military. Facebook has a scandal that affects elections. Amazon is swallowing up the global economy, and Element AI has positioned itself as a less greedy, more ethical artificial intelligence company.
2
Responding to the decision of Google employees to provide artificial intelligence to the Pentagon, it can be seen that the position of technology companies using military intelligence technology in the military has become a moral touchstone. Bengio and his co-founders in the early days Swearing that artificial intelligence will never be built for the military purpose of the attack.
But earlier this year, the Korea Research Institute of Korea's research university announced that it would cooperate with the defense department of Hanwha, a large Korean conglomerate, to establish a military system. Although Element AI has contact with Hanwha, Bengio signed a Opened the letter and boycotted the Korean Institute until it promised 'will not develop autonomous weapons that are not controlled by humans.' Gagné also cautiously wrote to Hanwha, emphasizing that Element will not cooperate with companies that develop independent weapons. Gagné and scientists soon Guaranteed: Korea Science and Technology Institute and Han Wah will not do this.
Autonomous weapons are neither the only ethical challenge to artificial intelligence nor the most serious challenge. New York University professor Kate Crawford focuses on the social impact of artificial intelligence. She writes that all the 'distress' of artificial intelligence will become a threat to the future, Distracting attention from existing problems, because 'sex discrimination, racism and other forms of discrimination are being embedded in machine learning algorithms'. Because artificial intelligence models are trained based on data provided by engineers, data Any deviation in it will have a bad effect on a given model.
Tay is an AI chat bot released by Microsoft on Twitter to understand the way humans talk. It quickly began to publish racist remarks, such as 'Hitler is right'. Microsoft apologized and revoked Tay and said Efforts are being made to address discrimination in the data. One of Google's artificial intelligence features is to help users find their corresponding artistic characters based on selfies, and this feature matches African Americans to slaves based on stereotypes, and Asian Americans. Matching geisha for oblique eyes. This may be due to the fact that artificial intelligence is overly dependent on Western art during training.
Problems like this are due to global prejudice, but this does not help the field of artificial intelligence. Because of the diversity of artificial intelligence in the broader computer science community dominated by whites and Asian men. It is even less.
Element AI's employees are 33% women, 35% of leadership positions are women, and 23% of technical positions, which is higher than many large technology companies. Element AI employees come from more than 25 countries. However, the company did not classify employees by ethnicity. During my visit, I saw mainly whites and Asians, especially high-rises.
Bengio said he was “shame” about the current situation and tried to solve the problem and expanded the scope of recruitment and provided funding for students from disadvantaged groups. At the same time, Element hired a vice president of personnel, Anne. Mezei, she prioritizes diversity and inclusion. In order to address the ethical issues that may exist in her products, Element is hiring ethicists as researchers to work with developers.
However, ethical challenges still exist. In the early research, Element AI developed some products based on its own data. For example, the Q&A tool was based on internal shared files for training. Operations Director Martel told me that because Element AI managers are not sure how to Ethical use of artificial intelligence for facial recognition, they plan to experiment with their own employees by installing cameras that capture their facial expressions with the permission of employees. Managers will investigate the feelings of employees and improve Their understanding of the moral dimension. 'We want to solve this problem by ourselves,' Martel said.
Bengio believes that scientists' work is to continue to explore the discovery of artificial intelligence. He said that the government should be more active in regulating this area, while making the distribution of wealth more equal, investing in education and social security networks to alleviate the artificial intelligence Inevitable negative effects. University of Washington professor Domingos said: 'I think Bengio believes that artificial intelligence can be ethical, and his company can become an ethical artificial intelligence company. ' 'But frankly, Bengio is a bit naive. Many techniques The staff are a bit naive, they have this utopian view. '
Bengio disagrees with this description. 'As a scientist, I believe we have a responsibility to engage with civil society and the government,' he said, 'influencing thoughts and hearts in the way we believe.'
3
Nonetheless, Bengio may have influence over any other scholar. He has trained the next generation of researchers. (One of his sons has also become an artificial intelligence researcher; the other has become a musician.) Although he is the co-founder of Element One of them, but Bengio admits that he has not spent a lot of time in the office; he has been focusing on the frontiers of artificial intelligence research, which are far from commercial applications.
While technology companies have been focused on getting artificial intelligence to do its job better—identifying patterns and drawing conclusions—Bengio wants to skip these foundations and start building machines that are more inspired by human intelligence. Hesitant, not willing to describe what it might look like. But we can imagine that in the future, machines can not only move products in the warehouse, but also navigate in the real world. They not only react to commands, but understand and sympathize Humans. Not only do they recognize images, they also create art. To achieve this goal, Bengio has been studying how the human brain works.
Although Bengio believes that artificial intelligence like humans is possible, he is impatient with the moral concerns that Elon Musk and others emphasize that machines can transcend humans. Bengio is more interested in human construction and use of artificial intelligence. Moral choice, 'One of the biggest dangers is that people treat artificial intelligence in an irresponsible way or in a malicious way – I mean for their personal benefit,' he said in an interview. Other scientists also have Bengio feels like this, however, with the continuation of artificial intelligence research, it is still funded by the world's most powerful governments, companies and investors. Bengio's university lab is largely funded by large technology companies. .
In a discussion about large technology companies, Bengio told me, 'We hope that Element AI can be as big as these giants.' When I questioned whether he would continue to maintain the concentration of wealth and power he condemned, he replied : 'The idea is not just to create a company, to be the richest person in the world. It is to change the world, change the way business works, make it less concentrated, and make it more democratic.'
(Selected from: Fortune Compilation: NetEase Smart Participation: Li Qing)
2. China's artificial intelligence enterprises have exceeded 4,000;
Science and Technology Daily News (intern reporter Tang Fang) 'Artificial intelligence enterprises showed explosive growth from 2015 to 2016. As of May 8, 2018, there were 4,040 artificial intelligence enterprises nationwide, and Zhongguancun became the highland of artificial intelligence innovation in China. On June 30, Yu Jing, deputy director of the Software Division of the Beijing Economic and Information Technology Commission, on behalf of the committee, released the White Paper on the Development of Beijing Artificial Intelligence Industry for the first time.
For the first time, the white paper has clarified the base of Beijing's artificial intelligence industry. The relevant organizations of the early stage of the committee have conducted a large number of research and special discussions, and comprehensive data collection, IT Orange, Blue Ocean and other investment companies.
In recent years, Beijing has emerged numerous products and enterprises in the field of artificial intelligence. For example, today's headlines have developed the world's first downloading of the vibrating APP in the first quarter of 2018. 'Beijing has formed an industrial cluster of artificial intelligence that covers the whole industry. This is related to its advantages in related policies, innovation and entrepreneurial atmosphere, capital environment, software R&D and patent protection. 'Yu Jing's 22nd China International Software with the theme of 'focusing on artificial intelligence and enjoying a better future' When the white paper was released at the 4th Global Software Industry Development Summit of the Expo, he said.
The reporter learned that among the more than 4,000 artificial intelligence enterprises in China, there are 1,070 Beijing artificial intelligence enterprises, accounting for 26%; the total number of artificial intelligence companies that have obtained venture capital in the country is 1,237, of which 431 are in Beijing, accounting for 35%; Beijing has 56.9% of the artificial intelligence enterprise financing stage before the A round, indicating that more than half of the company is still in the initial stage, with innovative development potential.
According to the white paper, Beijing's top five areas of artificial intelligence are smart medical, smart home, smart city, smart retail and unmanned. Tiantan Hospital, Peking Union Medical College Hospital is in the forefront of exploring smart medical care; Xiaomi Audio, Jingdong Dingdong Audio has become a popular smart home product in the market; Smart Retail has emerged as a distinctive retail enterprise such as Jingdong, Meituan, and convenience bee.
3. Stanford develops the AI program 'rediscovering' the periodic table of elements;
After about a century of exploration and experimentation, human scientists have compiled the great scientific achievement in the history of chemistry, the periodic table of elements into the current form. Now, physicists at Stanford University in the United States have developed an artificial intelligence program, using only a few After an hour, 're-discovering' the periodic table of elements.
'We want to know if artificial intelligence is 'intelligent' to be able to independently discover the periodic table of elements. ' The project leader, Professor Zhang Shouyi of Stanford University said in a press release, 'Our team proved this.'
Zhang Shouqi and others reported in the new issue of the Proceedings of the National Academy of Sciences that their artificial intelligence program Atom2Vec based on Google's natural language processing technology can learn to distinguish different atoms by analyzing multiple compounds in an online database. There is no human intervention throughout the learning process.
Zhang Shouyi explained that the characteristics of a word can be derived from other words that appear around it. For example, the word 'king' often appears with the 'Queen', while the 'man' often appears with the 'woman'. They put this The idea is applied to the atom, and the artificial intelligence program is given a known compound name, including sodium chloride, potassium chloride, and water.
By analyzing the names of these compounds, artificial intelligence programs have found that potassium and sodium have similar properties and can be combined with halogen elements to form compounds. 'Like 'king' and 'Queen' are similar, potassium and sodium are similar. , ' Zhang Shouyi said. (According to Xinhua News Agency)
4.DeepMind artificial intelligence learns to convert flat images into 3D scenes;
According to foreign media reports, DeepMind, a subsidiary of Google's parent company, has recently developed an artificial intelligence technology that renders the entire scene in 3D after observing 2D planar images. Some artificial intelligence researchers are currently trying to teach machine learning like humans. Instead of looking at the world in pixels, we look around our environment and make assumptions about everything.
The DeepMind team trained an artificial intelligence to guess what things look like from its unseen perspective. DeepMind scientists have proposed a Generative Query Network (GQN), a neural network designed to teach artificial intelligence how to Imagine what the scene of the object would look like at different angles. Basically, artificial intelligence observes the 2D planar image of the scene and then tries to recreate it. In this case, it is important that DeepMind's artificial intelligence does not use any artificially labeled input. Etc. It only looks at the three images and starts to predict what the 3D version of the scene looks like.
Researchers are working hard to achieve 'complete unsupervised scene understanding'. At present, artificial intelligence has not yet accepted real-world image training, so the next step will be to render realistic scenes from photos. In the future, DeepMind's GQN-based artificial intelligence may only use photos. Can generate almost exactly the same on-demand 3D scene as the real world.
5. Japanese technology company launched AI security camera Face recognition to prevent thieves
Recently, Japanese telecom giant NTT East and Japanese technology company Earth Eyes have jointly launched an intelligent AI security camera called AI Guardman. This camera is specially designed for thieves and can effectively capture thieves.
The camera uses open source technology developed by Carnegie Mellon University to identify human limbs through AI technology and refine it to the fingers. Calculate clues from different people by reading limb expressions. When this camera captures When suspicious action, the clerk will be notified immediately, and the image and location message will be sent together.
This camera can scan 144°, the angle is relatively large, it is very suitable for installation in supermarkets and other places.
The device is priced at $2,200 and requires an additional $41 per month for data processing.
At present, a shop or supermarket in Japan is currently suffering from theft loss of between 18,000 and 32,000 US dollars. I hope this camera can effectively suppress the losses suffered by Japanese supermarkets. TechWeb