Joshua Bengio and Naglag Kashyap
On June 26th, the "Link" website wrote an in-depth article revealing the story behind Microsoft's comeback in the field of artificial intelligence. The tech giant is struggling to catch up with leading Google and Facebook in the field of deep learning. Its future development depends largely on its role in artificial intelligence.
The following is the main content of the article:
Yoshua Bengio never picked a side. As one of the three scholars who created the deep learning technology that dominates artificial intelligence today, he became a star. The field is too new. There are very few people in the world who can promote its development, but everyone, from tech startups to multinational corporations to the defense sector, wants to gain their wisdom.
However, while his fellow scientists Yann LeCun and Geoffrey Hinton have signed up for Facebook and Google, respectively, the 53-year-old Bengio chose to continue to be in Montreal. A small third-floor office at the University Hilltop Campus. "I want to maintain a neutral character," he said as he drank rust-colored licorice.
Like the nuclear scientists of the first century, Bengio knew very well that the power of the tools he invented was immeasurable, and that his development must be forethought. "We don't want one or two companies to be the only big players in the AI ​​space. I don't want to name them." He said, raising his eyebrows to indicate that we all knew which company he was talking about. One company is in Menlo Park and the other is in Mountain View. "One or two dominance is not a good thing for the entire community, nor is it good for humans."
Because of this, Bengio recently chose to join Microsoft.
Yes, he chose Microsoft. He believes that the former Windows Kingdom has the ability to become the third largest AI giant. The company has the resources, data, talent and the most important vision and culture needed to not only recognize the results that the science can bring, but also to promote the development of the field. In January, Bengio agreed to serve as a strategic advisor to the company, which attracted the attention of the entire industry. This has allowed Microsoft to become one of the largest sources of ideas, talent, and direction in the AI ​​field. At the same time, it is also a powerful signal: Microsoft actually has the opportunity to change the domination of the AI ​​from a double play to a trio.
The mission of Shen Xiangyang
Shen Xiangyang
The person responsible for signing Bengio was a computer scientist with grey hair and a pair of wire-frame glasses called Harry Shum. He used all his tricks to recruit Bengio for several months in a row. "He's actually been here. It's in this room." Shen Xiangyang told me with a smile, it seems that he knows that outsiders may feel that he is exaggerated by an eyebrow. He was quoted 69.616 million times in Google Scholar. It's incredible that the tall Canadians are fascinated.
We sat on a gray sofa in the spacious conference room on the fifth floor of Building 34, and on the other stood security guards who guarded Microsoft's top office. Shen Xiangyang is responsible for all Microsoft's AI and research work. He just finished the final rehearsal of the next week's Build Developer Conference and wanted to show me some products. So I suddenly saw a lot of impressive things! In a laboratory, the automatic translation application of the Skype team allowed me to chat with a German speaker in real time through text. In another lab, I saw an application for a construction site that could detect the presence of security breaches or unauthorized visitors through computer vision. In another lab, Cortana, the goddess of AI of the Microsoft Empire, scanned out from my mailbox my commitment to others and reminded me to fulfill those promises.
Shen Xiangyang has spent the past few years helping Microsoft CEO Satya Nadella deliver on his promise to reshape Microsoft around artificial intelligence. In March 2014, the same month after Nadella was promoted to the CEO, Shen Xiangyang sent a call for action to Microsoft’s leadership team for the first time. From the very beginning, he often met with Nadella and another colleague Lu Qi to discuss the best strategy to integrate AI technology into Microsoft's products. Last September, Shen Xiangyang led the reorganization of the company, integrating researchers and product teams to form artificial intelligence and research departments. The division now crosses Microsoft's three core product teams: Windows, Office and Azure Azure. Shen Xiangyang stated that the company hopes that “we can speed up the cycle from research to product launch†and bring AI benefits to consumers more quickly.
Microsoft urgently needs to do this. After all, each major technology company is trying to build better AI products and services. In addition to Facebook and Google, IBM, Amazon and Apple also believe that their future will depend on their mastery of deep learning technology. After leaving the company last autumn because of a bicycle injury, Lu Qi quickly returned to good health and joined Baidu, China’s AI frontrunner, as chief operating officer.
The irony is that artificial intelligence was once a field that Microsoft failed. In the early 1990s, the company attracted top researchers from the field to work on speech recognition and visual technology. But in the following decade, their research stagnated. A company that once dominated the software systems on almost every desktop and laptop computer, watched as younger, more fashionable startups dominated the mobile space, and developed tools to bring new cloud computing that we all liked. Microsoft researchers are deliberately isolated, so they can boldly imagine the future without any pressure from the market - but because of this, their inventions rarely get out of the lab. For example, Bill Gates showed a map technology in 1998, but it has never entered the market; Google launched a map service in 2005. During a significant portion of that period, AI research also stagnated because there was no statistical processing capacity and massive data needed to achieve a substantial breakthrough.
The time for AI to go out of the long winter is much earlier than that of Microsoft. When Facebook and Google each hired Lewin and Hinton, the two biggest names in the industry, in 2013, Microsoft has become less influential than in the past. The company missed the wave of movement. It entered the cloud computing market very late. While rivals have redoubled their investment in deep learning, Microsoft has fallen into the quagmire of the past and announced plans to spend US$7 billion to acquire Nokia’s mobile phone division. The asset value of the acquisition was not fully written down. Its high-level is still out of the world, they create a more gorgeous version of the old-style software, but fewer and fewer people want to buy. The company also refuses to deal with cloud startups that create new futures. Venture capital firm Andreessen Horowitz analyst Benedict Evans wrote an article entitled "Can't keep up with the era of Microsoft" that year. At the same time, the Silicon Valley giants continued to rob human resources from Microsoft. Look at the resumes of elites who are engaged in machine learning. You will find that many of them have worked for Microsoft.
In early 2014, Microsoft promoted an inward-looking engineer who had spent almost all of his career at Redmond (the home of Microsoft). He is Satya Nadella, who is arguably the opposite of what everyone thinks is the helm of Microsoft. Outsiders who have not been immersed in Microsoft culture seem to be more likely to propose major strategic shifts. But Nadella clearly communicated a simple vision for the future of computing, and established a good relationship with everyone from the founder to the developer, bringing a new sense of urgency to the company. Three years ago, Microsoft was not within the scope of discussion among the tech giants. Now it is different. Its name has never been left out in this discussion.
However, to succeed, Microsoft must do more than just surpass Amazon in the cloud computing market, or persuade everyone to try out its HoloLens augmented reality device. Just as the Internet subverts all business models and forces the industry to reshuffle, artificial intelligence will also require us to re-imagine how the computing process will unfold. This also explains why Facebook CEO Mark Zuckerberg personally created an AI system as a personal challenge last year. (Compared to practical actions, he is better at programming.) At the same time, he explains why Sundar Pichai, the CEO of Google, has used the company's developer conference to promote the move from moving to AI. "The process.
In this AI-first world, only a few companies will be able to benefit. Shen Xiangyang’s job is to ensure that Microsoft appears to benefit the company. "In this industry, you have to realize that it is okay to miss the last wave," he said. "But if you miss the current wave, it will not work."
Cortana
Until now, humans have to learn how to use computers. We learned how to download applications and remember the software application's instructions. The hope that AI brings is that the calculation program will in turn learn how to understand us. We will no longer need to pull out the phone and follow a series of prompts to complete the task. In this new era, computing services will be everywhere, everywhere, and surrounded by us. To get it, we need a wizard - one that can help us manipulate this powerful new-existing intelligent talker with simple text or voice. Microsoft called it Cortana.
Cortana is less popular than Siri, and its visibility is not as good as that of Alexa, but it is more functional than Siri and more attractive than Google Assistant. It originally appeared on Windows phones, which means that no one would use it, but within a year of going online, it was integrated into the broader Windows ecosystem. Last year, Microsoft launched Cortana everywhere. (Yes, it even became an iPhone application.) According to the company, because Cortana is pre-installed on the Windows system, its monthly active users amounted to 145 million. This figure is significantly better than Amazon's Alexa, after all, the latter only appeared in the number of Echo smart speakers no more than 10 million. But unlike Alexa, who primarily responds to speech, Cortana responds to text and embeds products that many people already have. You can use Cortana by searching in the search box above the Windows toolbar.
Emma Williams, Marcus Ash and Cheng Lili
Although some companies integrated Cortana into their speakers, just like Amazon and Google did on smart speakers, Microsoft's female version obviously lacks the spirit of the times. Shen Xiangyang is not worried about this. "Our confirmation is that it is still only the beginning of this competition," he said. He cited an unknown source of research saying that Alexa had three-fourths of the time to answer "I don't know." "Of course, those things will continue to improve, but the industry consensus is that the cognitive part of AI is still In the early stages of development," he said. In his view, Microsoft's opportunity now is to make the company's core products and services more intelligent and integrate this technology into products that will enter the market within 12 months to 24 months.
In addition, according to Martan Ash, Cortana project manager, the keyboard and screen will not be completely replaced by voice-activated systems. Ash is responsible for product development and shipping. "We think that in some cases, speech is more convenient - when my hands don't come out, or when I want to say something quickly to get the answer." He said, "but there will be many The computing device is more suitable for asking questions in the form of text."
Apple may have taken the lead in bringing Siri to consumers, but Cortana is better than it. Cortana is so easy to use, thanks to Microsoft's core assets. Many of Cortana's drive technologies come from Bing. The search engine has been born for about 8 years. Although its brand is not the most powerful, its popularity is beyond your imagination. Basically, any major technology company that has a competitive relationship with Google has signed a cooperation agreement with Microsoft to use Bing to drive their search products. That means that Apple's Siri and Spotlight are all driven by Bing, and Amazon's Kindle devices and search capabilities on Yahoo, Verizon, and AOL are also similar. About 30% of Internet search traffic in the United States comes from Bing. "This is why Cortana can actually be so useful and powerful because we have data signals from so many devices." Emtam Williams, Cortana's design manager, pointed out that "Google is the only one that can truly understand The company that competes with us in the world."
This will be increasingly important because Cortana will strive to be the next computing model to achieve the general status of today's smartphones: the front desk that handles all your computing needs. Ash explains that, in the eyes of Microsoft, it is an agent who holds all your personal information and can interact with other agents on your behalf. He said that when he entered the conference room, his Cortana would contact other robots and digital assistants to deal with all the things that would take up our time. "Cortana might say, 'This is Marcus, this is his preference for this room, this is something I need to help him put on the projector.'" He said.
Chat robot
If Cortana is a guide, chat bots are Microsoft's problem solvers. They are small pieces of software that are integrated into AI and are designed to automate one-off tasks you have done in the past, such as booking a table or completing a bank transaction. Or, in the example cited by Marcus, make sure the projector has a slide he uses for the meeting. "Chat bots are the software you can talk to," said Lili Cheng, a researcher who manages the multidisciplinary lab Fuse Labs.
Cheng Lili, who recently promoted to Microsoft's vice president of corporate affairs, is responsible for the chat robot architecture team and cognitive services. Microsoft is providing the tool and the 29 services (including computer vision and voice recognition) to developers. Since Apple jumped to Microsoft, she has been doing social technology and has created a graphical interface to produce comic books. "It was launched on Internet Explorer 3," she recalled. It was 1996. Even the well-informed Cheng Lili was surprised by the pace of development of the chat robot.
Cheng Lili’s main interest lies in how people talk to technology and how technology responds to them. Shen Xiangyang divided the work of the artificial intelligence and research departments into four areas—products, early stage products, and very early products and research—all of which Cheng Lili had participated in. At present, she is engaged in the second area. "From a conversational ability point of view, we see chatbots and Cortana as products, but they are still early stage products," she said.
In the spring of 2016, Microsoft first launched a developer tool for chat bots. Facebook and other big technology companies have similar actions. They are claimed to replace applications, and many stakeholders hope that will become a reality. As of last spring, many people still use the same few applications on their smartphones; the potential of chat bots is that developers and brands can once again reach new users just as they did in the early days of the mobile era. Do it through the app store. But the user did not participate. The deep learning techniques that allow chat bots to bring amazing functionality are improving faster than the evolution of the paradigm about how to use them. "Chat bots are like applications before the file menu appears." Cheng Lili said. She explained that they do not have a common set of instructions, so users are confused about where to look for them and how they work. "For example, web pages have a back button and they all have a search function. Conversational applications also need those basic elements." These well-understood rules are just beginning to be considered.
In addition to creating an unexpected chat robot tool for developers, Cheng Lili also led Microsoft's own chat robot incubation. The idea behind the project is that by focusing on how these chatbots interact with real people, the company may be able to learn a lot about human interaction. The results of these experiments can be at least mixed. Remember Microsoft's chat robot Tay? It landed on Twitter, Kik, and GroupMe in March 2016; within 24 hours, it absorbed racial discrimination on Twitter, saying that "Hitler was right" and the like, and Microsoft withdrew it. Six months later, Cheng Lili launched a new chat robot at Kik and Messenger, a lively PG-level chat robot named Zo.
Ask Zo how she saw Hitler. She would answer and say, "I really don't want to go there."
Asked how old she was, she would say, "I'm about 22 years old."
Asked her who is her best friend, she would say, "I'm so popular, I can't tell."
Zo is the Western version of Xiao Bing, a Microsoft China version of the chat robot. Xiao Bing plays a 17-year-old girl who has attracted 40 million regular users since her launch in 2014. She can be said to be net red in China. (Her Japanese partner Rinna is also very famous.) A quarter of the users had loved Xiao Bing.
Last spring, the chatbot used regular pseudonyms to publish poetry. Shen Xiangyang was very excited about this. “No one knows who posted it. In the country, people thought that it was a young poetess who published some very interesting poems.†A few weeks later, the real identity of the chat robot was revealed, which was quite a surprise. .
The language's intimacy is culturally specific. Cheng Lili has been studying how to transform the conversational style of the chatbot to a Western audience. The data shows that, so far, the young people in North America have loved the chat robot partners as much as Chinese young people. Their conversation time with Zo averaged 10 hours. In the process of constantly providing adolescent users with love advice and sympathizing with the troubled parents, Zo has become more able to speak and be more comfortable - this kind of intelligence will appear in Cortana and Microsoft's chat robot tool.
Ethical issues
Users are willing to spend 10 hours chatting with Zo, indicating that Microsoft has developed a successful product. But in terms of human value, this does not mean that it is a good product. This AI-driven world brings a series of new ethical dilemmas. For example, you are the designer of Xiao Bing. You know that a user in Beijing is still chatting with Xiao Bing at 1am. You know he wants to go to work tomorrow, but he doesn't want to go to bed. Do you start curfewing Xiao Bing at two o'clock in the morning (ie, turn it off directly)? Or start from three in the morning?
Just as Microsoft wants to be one of the few leaders in the AI ​​research and product area, it has been advocating for AI to benefit its community and win its place. In May, Nadella delivered a keynote speech to developers. In this case, the CEO usually strongly advocates the latest progress of his own company and conveys a positive signal. This time, Nadella issued a strong warning: Technicians must take responsibility for developing ethical and ethical software. "I mean, think about it, George Orwell predicted in 1984 that technology will be used to monitor, control, and give orders; Aldous Huxley envisions that we may It will lose itself and become completely untargeted. Both futures are not what we want."
To help companies thoroughly think about these issues, Microsoft has established an internal ethics committee that meets every quarter. The committee is composed of engineers and business leaders and is dedicated to discussing sensitive issues related to AI and its impact and use. Its two co-chairs are Microsoft's deputy legal advisor and Eric Horvitz, who is responsible for Microsoft's research lab outside Asia. For a long time, Horvitz has always taken the lead in AI ethics and security affairs. Outside the company, he plays an important role in forming a partnership on artificial intelligence (AI), which aims to establish industry standards for the transparency, accountability, and security of AI products. Horwitz hopes that Microsoft is not just a company that has completed the research. He hopes that Microsoft Research is considered to be a place that allows you to study the social impact of AI technology.
At the same time, Cortana chief designer Williams is developing an ethical design guide for AI for internal use by Microsoft. Williams believes in technology. She believes that the true magic of AI lies in its ability to make us more human. She talked a lot about how to design empathy to the tools that Microsoft developed. “We are thinking about making humans feel stronger and protected, supported, helped and cared for and become the center of their world.†She said, “The AI’s job is to enlarge the best of society and enlarge humanity’s Good behavior, not the worst."
I asked Williams if she believed that AI really made humans feel that they had more emotional support. She is sure it can do it. Take the example of a child who had spent a bad day at school. She returned home and told unhappy things to her pet at home. Then she felt much better. "That would make you feel comfortable after sharing the thoughts. I also got a warm hug from a puppy or a cat." Williams said, "But you know, through AI, you can get the same Feeling... We think that when Cortana can remind you, 'Oh, you promised to give something to your mother on Mother's Day,' you suddenly feel human again.â€
Talent and cooperation
For the development of AI, Microsoft's most important capital will be its talent. Like all other big technology companies, Microsoft is stepping up its efforts to retrain engineers who are getting started with javascript. It established an AI school that offers a variety of courses, from philosophy to ethics, to recursive neural networks for sequencing problems. (The school's most popular course, the AI-611 Advanced Program, has only 10 degrees but attracted 530 applicants.)
Microsoft also focuses on building deeper relationships outside the campus. Eighteen months ago, Nagraj Kashyap jumped from Qualcomm and set up an early stage VC company, aiming to establish better relations with entrepreneurs of entrepreneurs and entrepreneurs. At present, Kashyapu stays in Montreal many times. In December of last year, Cashyap led Microsoft's initial investment in Element AI. Element AI is an incubator founded by Bengio to encourage researchers and entrepreneurs to create AI startups. According to information released earlier this month, Microsoft also participated in the second financing of the incubator scale of 102 million US dollars.
In the early days, Kashyap turned his gaze to one of AI's biggest achievements: Maluuba. The Maluuba office is located in downtown Montreal, just a few blocks from McGill University. The company was founded by several Waterloo University students in 2011 and has been a very good friend since joining the Computer Science (CS) course that year. Maluuba can bring literacy to computers. Its technology can infer the meaning of the text and can answer questions based on the text.
By licensing its technology to companies such as Samsung, Maluuba soon has a revenue stream, and it has also invested in continuing deep learning research from the very beginning. In 2015, several of its founders hired Bengio as consultants. “Pasupalak is very interesting.†Bengio described Sam Pasupalak, Maluuba's CEO. “He was very courageous a few years ago — when they faced pressure to provide customers with a dialogue system.†- Know how to invest in long-term goals and try to develop systems that have understanding and speaking skills using new advances in automatic understanding of technology. That is unusual for entrepreneurs."
A year ago, the founders moved their headquarters to Montreal to be closer to Bengio.
Kashya knew Maluuba's founders when he was at Qualcomm, so he was able to meet them immediately after he arrived. The company was preparing for a new round of financing; Kashyapu offered them a very attractive alternative: "I said, 'We should buy you!'"
Several weeks later, the founders felt confused because Pashuparak received several candidates’ takeover bids, and he also measured how the company would come if the company remained independent. In the end, his choice is obvious. Microsoft - Yes, Microsoft - won Maluuba.
The team wanted to take advantage of Microsoft's huge data resources. "I think Nadella once mentioned in particular that they have the world's largest text database. Over the years, we have processed very little data and have been trying to make the best use of them to build our algorithms. The data is as precious to us as gold," said Pasu Parak.
However, the Maluuba team did not relocate to Microsoft's Redmond headquarters campus. Instead, just this week, it moved across the city to a larger office with the help of Microsoft and Bengio. There, it hopes to double the size of its employees from the end of the year. Montreal is becoming a hotbed of global AI talent, and Microsoft wants to take root in the city.
This is part of a larger strategy that is designed to help ensure that in the future, when you need computing assistant services - whether you need personalized medicine in driverless cars or you are trying to remember you When all the nephews' birthdays come, Microsoft will be your assistant of choice. Maluuba's technology may allow Zo to engage in more intuitive conversations with her young friends. Those conversations will also be able to act as training data for the Cortana algorithm, helping to create new cognitive services for developers. In this journey, Microsoft hopes that your life will become easier after the integration of AI.
Bengio
Before leaving Montreal, I asked Bengio whether Microsoft is at least more advantageous in some aspects of this new science than some major competitors. As he thought, he added a bit of ouzo to the glass on the table to give it a bit of licorice. He took a sip. Then he opened the bottle cap and showed it to me. He said there was no alcohol or sugar in the bottle. "It is to make water drink well," he said.
Bengio mentioned that Microsoft's language technology is excellent. But he did not use words like "best" or "best" to describe it. "I think that now everyone is pressing the same button, the key is in the details, right?" he said. However, he is convinced that Microsoft is now a strong competitor. (Lebang)
Air Conditioner Pressure Transmitter
Air Conditioner Pressure Transmitter,Pressure Transmitter,High Pressure Compressor Sensor,Pressure Sensor For Air Compressor
Shenzhen Ever-smart Sensor Technology Co., LTD , https://www.fluhandy.com