The technology circle is a rapidly changing field. From VR technology, AR technology, artificial intelligence to today's blockchain technology, more and more new technologies and new concepts emerge, and artificial intelligence is gradually getting cold. When the tide recedes, Technology can't find application scenarios, but talents can't keep talents. How far can artificial intelligence go?
For the whole year of 2017, the investment in giants around artificial intelligence and entrepreneurial bets are endless, and the amount of financing continues to set a new record. But whether it is image, language interaction, algorithm, machine learning is not enough to combine with the actual scene, in other words, landing becomes the biggest problem.
On the other hand, the emergence of more and more new technologies and new concepts has also caused artificial intelligence to gradually become cold. In the second half of 2017, with the rise of Bitcoin, the blockchain has become a new hot spot for everyone, and it has occupied almost a circle of investors and media people.
Everyone knows that artificial intelligence represents the future, but in real life, except for certain specific scenes, people can hardly feel the convenience brought by artificial intelligence.
For example, smart speakers can be said to be one of the first products to be created by the wave of artificial intelligence. In 2017, it was once a "hundred box battle", and it was also a place for giants such as Ali, Jingdong and Xiaomi to grab the beach, but Speech recognition, one of the core selling points, is not as good in the experience as the public expects.
First of all, speech recognition requires huge database support, but when we actually use it, we find that there are many expression habits that cannot be understood by the machine. Although many speech recognition companies say their technology recognition rate has reached 95%, or even 98%, in the common situation of dialect, biting, swallowing, etc., most products can not make reasonable feedback. (Interested readers can check the IT ear evaluation article to learn more) Second is the accuracy of sound collection, although the main scene of the smart speaker is the living room and bedroom, but the scene is more complicated when used, the tone of the person speaking, The speed of speech is different. When you collect sound, you will be disturbed by other people, environmental noise, or even the speaker itself. This will cause the accuracy to drop and affect the experience. Last but not least, there is a lack of interactive nature. Almost all smart speaker products require a voice key like "Sesame Open" to wake up, and each time the command is issued, the wake-up words are repeated, which creates a significant distance between the user and the product.
In general, the product form of the smart speaker is indeed novel and attractive, but its essence is still a voice assistant, and it is not fully qualified for the role of the smart home.
Another obvious example is autopilot technology. Everyone is accustomed to classify autonomous driving into five levels of L1-L5. Recently, we also interviewed some companies that drive autonomously (assisted driving). They generally believe that their technology is still in the L3 level, that is, in the highly automatic driving stage. In the case of intense situations, human intervention is still required. How is this technology implemented?
Tesla's identification of the surrounding environment is achieved by 12 long-distance ultrasonic distance sensors Ultrasonic Sensors, a long-distance radar Radar and a forward-facing camera Forward-facing camera installed on the vehicle, while other domestic companies The solution is similar. However, through long-term visits and observations, we also found some of the problems. First, radar or ultrasound will interfere with each other. Imagine that a dozen cars are stuck at the intersection, and there will be strong interference between them, resulting in the collection. The data is biased. Secondly, the forward camera is not omnipotent. Think about the "high-light dog" on Weibo, and think about the various neon lights in the bustling market. These are the biggest challenges for the camera. In a chat with the head of NavInfo's new automatic driving R&D department, he also admitted that the cameras used in autonomous driving are still difficult to accurately judge in the complex environment of low light, backlight, or multiple light sources, even in some cases. Areas without road markings and obvious roadsides are also difficult to identify accurately.
In the two typical cases where artificial intelligence was the fastest growing in 2017, the performance was not satisfactory. Look at other application scenarios, robots, smart logistics, smart cities seem to be too far away from consumers. The major artificial intelligence companies frequently broke out the financing news, and the conference did not open, but rarely heard the case.
Playing the concept but not doing it wellMost of the enterprises in the domestic artificial intelligence outlets such as Shangtang, contempt, Yitu, and Tucson are doing algorithms. Due to the lack of commercialization of technology, AI technical services can only be provided to customers through project customization. That is, simply selling models and selling algorithms rudely, such as face recognition technology services, basic language recognition services, and knowledge mapping projects in the financial field. However, it is unsustainable to use people and algorithms as the core competence of the enterprise. In particular, the algorithmic dividend period in the field of deep learning has become shorter and shorter.
Taking facial recognition as an example, after the education of iPhone X, face unlocking of mobile phones is becoming a standard for more and more manufacturers. But as a consumer, 98% recognition rate and 99% recognition rate have almost no difference in experience. Users who have used face unlocking know that unlocking in mobile (swaying) is the main reason why they can't replace fingerprint unlocking at present.
Shangtang Technology has cooperated with OPPO mobile phone to provide facial recognition unlocking solution for its flagship R11s. However, after our actual test, R11s can scan the face and unlock it, we actually use a one-inch photo to successfully unlock the mobile phone. Think about how much safety you have. (Interested readers can check the IT ear evaluation article for more information.) Of course, what parameters and anti-shake technology is used by the smart phone's front camera is beyond the control of the artificial intelligence solution company, so the result of such an experience Can't completely blame the solution provider.
But from another angle, consumers can't have time to understand the complexity behind this technology. Everyone will only feel that the technique of unlocking the face is not easy to use. Slowly, those who dare to try it will lose their enthusiasm, or even accept it. This technology. Just like the VR market two years ago, the nine-nine-nine VR device did give more people the opportunity to experience VR for themselves, but it also kept many people away from VR, didn't they?
Grab the talent but can't keep the talentAlthough artificial intelligence has been around for a long time, there are not many professionals in this field in China. According to the “European Think Tank: The Core AI Talents Map of Major Chinese Enterprises†released by Yiou.com, 218 artificial intelligence Chinese executives accounted for 85% of the doctoral students graduated from well-known universities, and these people basically covered the domestic You can think of all the well-known companies of artificial intelligence.
But with these people, can you do a good job? By sorting out the public information, we found that: Baidu chief scientist Wu Enda resigned to establish Deeplearning.ai; Bai Jin senior vice president Wang Jin resigned to set up Jingchi Technology, focusing on the development of unmanned taxi technology solutions; Yunzhisheng CMO left to create artificial intelligence education company Xiansheng Education; Cao Xudong, the founder of Cao Xudong, founded the automatic driving company Momenta; Dr. Ding Peng, who is deeply involved in the creation of Ge Ling, left the company to establish the artificial intelligence medical company DeepCare; five people from Intel, such as Wu Gansha, left the company to establish the autopilot company. Ni Kai, Vice President of Super Automobile, founded Heduo Technology...
Liu Qingfeng, chairman of Keda Xunfei Co., Ltd., publicly stated that "artificial intelligence is currently the most needed talent, especially in the field of industrial applications."
The importance of talent to the industry can be seen. Almost every artificial intelligence company that has been established for more than three years will have an event of “team leaving, second startupâ€, and it is precisely these “bottoming†incidents that have caused many companies to stagnate in product iterations.
According to the data report, artificial intelligence enterprises are now more favored for middle and high-end talents with more than 5 years of work experience, and the wages of the corners are generally 2-3 times of the current salary level. The top talents leaving the company and the mid-to-high-end talents have been dug into the corners of the domestic artificial intelligence industry.
How far can the tide go back to artificial intelligence?Along with the fiery bit of bitcoin, the blockchain has recently become a new favorite of attention, and large-scale financing incidents have gradually emerged. For a time, everyone used “fried coins†as talks. The artificial intelligence technology that has been speculative for one year seems to be cold.
The “cold encounter†mentioned here is not to say that the industry is dying, and more refers to the industry pattern formed during the year. From a resource point of view, in the past year, the top 50 artificial intelligence companies in China have received almost 80% of the financing in the market, and the resulting high concentration of resources is beyond the reach of later entrepreneurs. From a technical point of view, speech recognition, face recognition, image processing, assisted driving and other related fields have also formed a situation of super-strong or multi-super strong, which is difficult for small and medium-sized teams to match. From an application point of view, companies such as Shang Tang and Yun Cong have already cooperated with many partners. The data behind this will be a prerequisite for technical iteration, and a few streets for those who only do research in the laboratory. In this way, aside from the top giants such as BAT, Microsoft, Google, etc., these artificial intelligence companies on the head alone have already confirmed the so-called "two-eighth rule". How far can other companies go?
The current situation of the entire domestic artificial intelligence company is like this: First of all, there is money, and there are tens of millions of financings everywhere. Secondly, there are talents. The above has mentioned the current situation of talents. Many companies also cooperate with well-known universities to cultivate talents. Finally, there are some technologies. According to official data, the recognition accuracy of speech has reached 99%, and the recognition rate of images is over 95%. But what is the life experience that artificial intelligence brings to us in a year?
Perhaps 2018 is a year of technology, and it is also a year of going to falsehood.
Guangzhou Winson Information Technology Co., Ltd. , https://www.barcodescanner-2d.com