In the past year or two, the enthusiasm of artificial intelligence has been obvious to all. From the academic world to the industry, from entrepreneurs to investors, the arguments about the upcoming AI are emerging one after another. At the same time, the valuation of startups in the AI ​​sector has also risen, and the high valuations carry the optimistic expectations of entrepreneurs and investors on AI and the company's development prospects. But behind this high expectation, is there enough strong logic to support it? Many investors who have “entered†can't even say why.
Near the end of the year, some investors began to “cool down†the AI, guiding the industry to take a cool look at current AI technology and commercial applications. If the investor's point of view is not enough to serve the public, then the industry predecessors who have fought in the AI ​​field for more than 20 years may be able to give us some more realistic inspiration from a technical perspective.
Recently, Love Analysis interviewed Zhang Lianyi, Chairman of Jietong Huasheng, and discussed his current development level of the various technologies of artificial intelligence and the status quo of the industry application, as well as the views on the development trend of the whole industry.
Photo: Zhang Lianyi, Chairman of Jietong Huasheng
Zhang Lianyi graduated from the Department of Environmental Engineering of Tsinghua University in 1989. Since 1990, he has been committed to commercializing Tsinghua's OCR technology. In 2000, he and Jie Shinan, a researcher at the Institute of Acoustics of the Chinese Academy of Sciences, jointly created Jietong Huasheng, which mainly provides speech synthesis technology. In June 2016, Jietong Huasheng officially landed on the New Third Board.
Today, this 16-year-old company has built dozens of technologies covering speech recognition, speech synthesis, voiceprint recognition, semantic understanding, OCR, handwriting recognition, face recognition, fingerprint recognition, machine translation, and data mining. Azimuth artificial intelligence platform provides AI technology and solutions for customers in many industries including finance, telecommunications, energy, transportation, education, medical, government, automotive, IT Internet. (See the analysis of love article "AI pattern for 3 years, can Jietong Huasheng become the decathlon champion? | Company research")
Zhang Lianyi believes that artificial intelligence technology, including voice, image recognition and semantic understanding, is still not perfect. It is the improvement of the acceptance of AI technology by the whole society, which has led to a surge in the demand for commercialization of various technologies. However, some companies and industry people have over-promoted AI technology, which is actually misleading the public. This misleading will not be conducive to the future development and application of AI technology. In addition, he also pointed out that from 2011 to the first half of 2016 is the five years of AI technology enlightenment, and the next 3-5 years, is the stereotype of the AI ​​industry pattern.
Love analysis and excerpt some of the exciting content to share with you.
AI technology is not perfect, application and commercialization depends on industry requirements
Q: Is OCR an artificial intelligence technology?
A: Before the artificial intelligence and human-computer interaction were separated, they are now combined. I think that artificial intelligence can be understood as two levels. Human-computer interaction is a shallow level. Let the machine be able to see and listen like a human being, and let the machine have the same thinking and reasoning ability as human beings. OCR belongs to a relatively traditional technology in the field of image recognition. It belongs to human-computer interaction and is a shallow level of artificial intelligence.
Of course, although it is more traditional, it is not fully mature. In recent years, with the development of deep learning technology, some breakthroughs have been made. The accuracy of recognition in the case of blur, line, overlap, background color, etc. obvious improvement. In addition, from the needs of developers, the demand for OCR technology in various industries is still very strong, including text recognition on the bottle, ticket identification, express delivery identification in the logistics industry, and so on. Therefore, OCR technology is a traditional technology in the field of artificial intelligence.
Q: What are the sources of OCR technology in China?
A: OCR technology began to rise in 90 years. The main sources are Tsinghua (represented by Jietong Huasheng and Wentong), Chinese Academy of Sciences (represented by Hanwang Technology (002362)), Motorola (represented by Shanghai Hehe Information), and Singapore National The institute also has two representative companies in China.
Q: Does image and speech recognition now require manual proofreading?
A: All need it. Some of the outside world's nearly 100% accuracy speech recognition demonstrations are actually specially trained, and the actual usage scenarios will not have such high accuracy. Of course, all walks of life have different requirements for voice recognition. For example, medical, legal and other fields are not allowed to make mistakes. Then the technology company will set up a call center at the rear to conduct manual proofreading. For example, Nuance has a team of several thousand people. In the proofreading work, either special data training is strengthened to reduce the error rate. In actual use, for some small errors, the user can perform self-calibration.
Q: What is the current accuracy of speech recognition in the industry?
A: The mobile APP and telephone channels are 16K and 8K channels respectively. Now the 16K channel recognition rate of the mobile APP is about 95%, and the 8K telephone channel recognition rate is only 85%, of which 60-70%.
Q: What are the difficulties in voiceprint recognition?
A: Single-channel voiceprint recognition is also possible, but cross-channel voiceprint recognition is not enough, such as from the telephone channel to the mobile phone, from the mobile phone to the microphone, etc., it is difficult to identify.
Q: How do you view current face recognition technology and the market?
A: I understand that the technical requirements for face recognition are linked to industry applications. Many startups now use face recognition in the financial and security fields. If used in company attendance, more than 98% of the recognition accuracy is sufficient, but if used in the financial field, such as brush face payment, even if it is 99.99%, there will still be 0.01% error.
Therefore, I believe that the application of artificial intelligence in the financial field, single technology has unbearable risks, technology can not guarantee 100% correct, only through a number of technical integration, multiple verification, can be guaranteed to be foolproof. Now the accuracy of biometric recognition is not very high compared to pattern recognition. Only iris recognition is the highest, so Jietong will use voiceprint, face, fingerprint and license identification together.
Q: What are the current difficulties in face recognition technology?
A: The biggest difficulty in face recognition is light. Light is too strong and too dark to ensure good recognition. Advanced technology will still be affected by environmental factors. Just like the academician Zhang Wei of Tsinghua University, he has not been optimistic about the unmanned car, because the unmanned car mainly depends on the vision, and the vision will be greatly restricted under the conditions of darkness, wind, rain and smog.
Q: What do you think of the unmanned car?
A: Of course, the unmanned car is very hot now. I personally think that the unmanned car is in some specific areas, such as the unmanned track, that is, it is open in a certain section, and no one is allowed. This is ok. However, in order to open on the daily road, in addition to the problems encountered by the camera vision scheme mentioned above, there are also problems of identification of people, things and scenery. In the past, Tesla had an accident, at least indicating that under certain circumstances, the unmanned vehicle still could not identify objects and environmental conditions well. In addition, the definition of responsibility for traffic accidents is also a big problem.
Of course, some multi-sensor fusion solutions are now used in the industry to reduce visual limitations, but I would like to say why we must pay for human life, whether it is in the car or outside the car. We still don't rely on technology to achieve convenience and use our own life as a bet price.
Q: There is still a point of view in the industry. There is no driverlessness. There is also a high accident rate. After drones are mature, it can improve safety to a certain extent. What do you think of this view?
A: It is right to improve safety after the technology matures, but the responsibility of human accidents is relatively easy to define, and the words of unmanned vehicles are not easy to define, its risks are unpredictable, and human risks are predictable. . Of course, unmanned vehicles as a technical pursuit is worth studying and exploring. It may be realized one day, but I think it is still far away. In addition to the sensors on the car, the related supporting infrastructure has to keep up. It still takes time, and the technology that is foreseeable is not yet available.
AI is booming thanks to increased industry acceptance, and the AI ​​pattern will be finalized in the next 3-5 years
Q: What do you think of the development stage of artificial intelligence?
A: Jietong is a model that was transferred in July 2011. It is just a complete five years until June 2016.
I think there are two stages in the development of artificial intelligence industry. The five years before June 2016 are the enlightenment stage. The players in these 5 years are basically the companies that have done this before, such as Jietong Huasheng, Yunzhisheng, Sibichi, contempt, Shangtang, Xiaoyi Robot (300024), Haixin Kejin, proud Yintong, Zhongke Xinli, listed companies have Science and Technology News (002230), Hanwang Technology.
In the second half of 2016, Baidu, Tencent, Ali, Sogou, Huawei and other "aircraft carriers" also came in, and the real battle of artificial intelligence began. Therefore, the five years after this is the stage in which the industrial structure is gradually taking shape. At present, most of the customers are trying their best, instead of using technology, so the demand in many single areas is very fragmented and has not yet formed. In the next two or three years, the industrial pattern will gradually take shape, and the pattern will be gradually determined after four years.
By then, artificial intelligence will definitely produce a giant like BAT, because the times create heroes.
Q: Do you think that the current artificial intelligence is overheating?
A: It is really overheating. This industry is indeed rising, so you can't underestimate the entire industry, but you can't overestimate its technology. The development of artificial intelligence technology is not due to the recognition accuracy of 95% and 97% of speech recognition that everyone sees, but benefits from the understanding and tolerance of artificial intelligence in the whole society.
It turns out that people's attitude towards artificial intelligence is that they are not satisfied with the wrong word. You have to understand when I speak. But these years have slowly developed, people found that although there are typos, but I can tolerate and try, you try the voice, he tried the image, so now the demand is swarming. It is this tolerance that greatly promotes the application of artificial intelligence technology.
Of course, this kind of tolerance and trial is still more in the commercial field. Some voice companies started to make 2C products, but when they first came out, they were hot for a few days, and no one used them. In contrast, the use of voice technology in the field of intelligent customer service is a relatively correct direction. From the current stage, the commercialization of the vertical sector will go faster.
Q: How do you think that some artificial intelligence will defeat and replace human beings?
A: After Alpha Go won, someone asked me if artificial intelligence would defeat humanity. I think this problem depends on what the machine beats humans. If it is said in some respects, then after the birth of the calculator, the machine has already defeated humans.
Regardless of what others say, I understand that artificial intelligence is a human partner, and it was born to help human beings and become human assistants. It is not impossible to replace human beings, but there is a premise. Can we humans first understand what is going on in our lives?
Artificial intelligence is a human being. The reason why human beings can recover a car that has been broken is because people are proficient in the overall structure of the car, but humans have little knowledge of their own brains, and the core of AI is precisely in the brain. Therefore, to create an AI that transcends humanity is based on the fact that human beings have sufficient knowledge of themselves.
The future development of artificial intelligence must be accompanied by the deepening of human understanding of itself. When this understanding reaches a certain level, we can talk about whether artificial intelligence can defeat human beings. It is still too early to discuss this issue. Now we should pay more attention to how these technologies can serve the industry, serve the public, reduce the workload and pressure, improve work efficiency, reduce costs, and bring some entertainment and convenience to the public. .
Q: How do you think that artificial intelligence will take the college entrance examination and take this project?
A: This is actually one of the things I don't agree with the most.
Chinese parents have two heart diseases, one is worried that their family is sick and the other is worried about their children's education. Education in China is a solidified cramming education. It is no exaggeration to say that it is to ruin children. True education should be enlightened and guide you to discover the essence of things. And we are indoctrinated education.
Letting artificial intelligence participate in the college entrance examination can't prove anything. Even if artificial intelligence is successful in the college entrance examination a few years later, it can only prove one thing. China's exam-based education has reached a point where it can't be added. Because whether it is answering a question or writing a text, it is not difficult technically. Just take a large amount of question bank and let the machine learn and train. This process has not created anything and has not made some beneficial changes to China's education.
So our company can explore the application direction of artificial intelligence, such as helping children write, practice Mandarin and spoken English, that is ok, but don't exaggerate its application in education. Enterprises must serve the industry and cannot mislead the industry because such misinformation is a mistake.
Q: How do you view the high valuation of current AI companies?
A: First of all, we need to be sure that there are still some very good companies in this field.
But from a technical point of view, the threshold in this field is not as high as expected. Many companies are based on foreign open source technologies, and there are not many original technologies. Since it is open source, everyone can learn, not to say that there is a high threshold. For example, in the past two or three years, a lot of companies have made a lot of voices in China.
So the current high valuation is on the one hand because the market is really up, and on the other hand does not rule out some packaging ingredients in it.
I have never believed in unicorns, because artificial intelligence contains a variety of technologies, and it is impossible to monopolize by a single technology. This is an era of turbulence, and innovation will continue. Enterprises certainly have to build core technologies, but it is unlikely that they will rely on core technologies to monopolize the industry.
Artificial intelligence companies still have to do it in an open and shared mindset. The key to an artificial intelligence company's long-term development is whether it can correctly understand artificial intelligence, not the lead of a single technology. If you can't understand artificial intelligence correctly, you may be alive, but how far you can climb can be a question mark.
Love Analysis is an Internet research platform focused on innovative enterprise research and evaluation. Love analysis takes enterprise value as the research core, and conducts long-term follow-up research on innovation fields and benchmarking enterprises in a unique product form, serving enterprise decision makers, practitioners and investor user groups. Pay attention to the love analysis public number ifenxicom and get important information in time.
Aloe Oil,Jojoba Oil,Olive Oil,Cosmetics Essential Oil
Ji'An ZhongXiang Natural Plants Co.,Ltd. , https://www.zxnaturaloils.com