Editor’s note: This article is from the micro-channel public number “Blue Hole Business (ID: value_creation), Author: Guo Zhaofei, 36 krypton release authorized.

Survival or destruction is a problem.If we add the premise of “artificial intelligence to humans”, the answer is even more elusive.
But in Li Yanhong’s eyes, destruction?nonexistent.Artificial intelligence will make people “eternal life”, in the future every ordinary person can achieve the eternal spirit and eternal soul.
Yes, of course you are not mistaken.This may seem like a metaphysical thesis, but its explanation is actually very scientific.
“Every word that everyone says, every thing they do, even your memory, your emotions, your consciousness, etc. can be stored digitally and placed on a network disk or other cloud, your way of thinking can beWhen the machine learns and encounters new problems, it can be used to restore reality through technology, and it is possible to have a dialogue across time and space with future generations. ”
Putting artificial intelligence in the business field is an indispensable opportunity for Baidu, who has made detours in the mobile era.A few years ago, Li Yanhong said that he hopes that Baidu will be an artificial intelligence company in the future, not an Internet company.
In order to accelerate the intelligent commercialization of Baidu’s AI industry, Li Yanhong has made several adjustments in strategy, organization, and personnel in the past year.
In December 2018, the Intelligent Cloud Business Unit (ACU) was upgraded to the Intelligent Cloud Business Group (ACG), which also carries the development of AI to B and cloud services.In May 2019, Wang Haifeng was promoted to Baidu CTO; in September, Baidu Intelligent Cloud and CTO system merged, and Baidu Intelligent Cloud team reported to Wang Haifeng.
In Wang Haifeng’s view, the development of artificial intelligence technology is becoming more and more mature, showing strong versatility, as well as the characteristics of standardization, automation and modularization of industrial mass production, and has the ability to deeply empower all walks of life.Artificial intelligence will continue to penetrate and transform all aspects of people’s production and life like mechanical technology and electrical technology. Artificial intelligence is becoming the core driving force of technological revolution and industrial transformation.
Baidu CTO Wang Haifeng
On May 18, 2020, Wang Haifeng announced the new business architecture of Baidu Intelligent Cloud, which includes three layers: the bottom layer of Baidu brain, the common basic cloud platform, AI middle platform, knowledge middle platform and other middle layers.solution.Baidu’s smart cloud strategy is based on cloud computing and artificial intelligence as the starting point, focusing on the layout of four major tracks: smart finance, smart customer service, smart medical care, and smart city.
Can 20-year-old Baidu seize the opportunity?
Take a detour
Around 2017, Baidu discovered that the AI ​​technologies accumulated in the search and Baidu business systems, such as machine vision, speech, and natural language processing, have been applied in other industries, and AI may “sell money” on a large scale.
Based on its own accumulation, Baidu combines AI with the cloud and outputs AI capabilities through the cloud. This is also the biggest difference between its competitors such as Alibaba Cloud and Tencent Cloud.
Li Shuo is the deputy general manager of Baidu Smart Cloud and the general manager of the smart finance business department. In his impression, he hardly stayed in the office in 2017, either on the way of researching the industry, or talking to different companies about how to treat AI and howLet the AI ​​land.
“I have to admit that everyone really cared about artificial intelligence at that time, whether the new technology could solve the problems in the business.” Li Shuo recalled to “Blue Hole Business”.
Even for AI and industrialization, Baidu has different voices: In the 1990s, artificial intelligence had a low tide. How long can this round last, will it fall into a trough again?Where should Baidu AI industrialization begin?doing what?What not to do?
Li Shuo admits that there was no precise planning at that time, and which step should be taken in response to random changes.He once talked with Li Yanhong long time, the uncertainty is very big, and also made some predictions, Li Yanhong’s attitude is clear, “bold try.”
Baidu did go through many detours.
At the beginning of 2017, Baidu’s landing robot voice dialogue technology has two types of incoming and outgoing calls in corporate call centers.After analysis, Baidu believes that inbound is a service scenario and outbound is a marketing scenario. Any company is eager to get more customers, and outbound must have more market space than inbound.
China Unicom is the first benchmark customer of Baidu Intelligent Cloud.In order to reach the voice service of the robot, Baidu engineers rushed to the Unicom Tianjin Call Center and stayed there for more than a month.Worked with the intercom staff during the day, stayed in the Express Hotel at night, and finally became a platinum member of the hotel.
After a long time, Baidu engineers found that the exhalation is relatively complicated, the basic conditions are not available, and some scenes cannot be landed.Incoming customer needs and pain points are obvious. For example, it is difficult to get through the phone during peak hours or when the operator is off work. Many problems and demands are similar, which is more conducive to the replacement of labor by machines.
In addition, in some companies’ outgoing scenes, the outgoing calls may be harassment or even fraud.
In the fall of 2017, Baidu decided to temporarily stop outgoing calls and only do incoming calls.Some insiders objected that calling out made more money than calling in. If you stop, you may lose a huge industry opportunity.Li Shuo had also struggled, and finally made up his mind not to touch out for the time being.Looking back now, this is a wise choice, exhaled through the chaos in the industry, and the current supervision in this field is becoming stricter.
In cooperation with China Unicom, Li Shuo also came up with a methodology for AI landing.Unicom’s annual investment in customer service is billions. The goal of the first phase of cooperation between the two parties in 2017 is that the machine replaces 17% of manual customer service, which saves China Unicom a large-scale investment.
“In many industries behind us, we are reusing this law. When our artificial intelligence first contacts an industry or a customer, we must find a business department whose value is measurable to land, and we cannot float in the air.Talking indiscriminately. “Li Shuo concluded that this value paradigm is to build on the basis of its informatization to solve a customer ’s problem, and finally link it with his KPI or a certain business item indicator to find a landing scenario.Through this standard, it is also possible to determine which are of no value.
Since then, Baidu AI has used this method to enter the Internet, finance, industry, security and other fields.Baidu is also constantly enriching the standards for AI commercialization: to have a good prospect, a business model, and a product. Finally, the accounting period. Only when these four conditions are met at the same time is a sustainable business.
In April 2018, Baidu’s smart medical platform, Lingyi Smart, was launched.Huang Yan, deputy general manager of Baidu Intelligent Cloud and general manager of the smart medical division, explained that Lingyi wisdom has launched an integrated solution around the three main medical links of screening, diagnosis and management.
“Based on the multimodal medical AI platform and professional and authoritative medical knowledge platform, a series of smart medical products and applications have been constructed. These include fundus screening products, new coronary pneumonia screening products, clinical decision-making products, and reasonableDrug products, medical record quality control products and chronic disease management products, etc. “Huang Yan added.
Faced with urban governance, Baidu released a “smart city solution”.In the application of smart city management, relying on Baidu AI platform to achieve multi-modal, multi-scenario knowledge capabilities in urban environments.On the one hand, it adapts various types of front-end devices such as cameras, drones, remote sensing satellites, and handheld terminals; on the other hand, it supports semantic understanding of faces, human bodies, vehicles, and urban events.
According to Liu Yawen, vice president of Baidu and general manager of the Smart Cloud Smart City Division, Baidu Smart City participated in the construction of the “urban brain” in Haidian District, Beijing.For example, in the comprehensive management of muck trucks, accurate identification of muck trucks is achieved, and one million pieces of passing data are processed every day. More than 9,000 muck trucks are found during peak hours, with an accuracy rate of more than 95%.
At the same time, with the help of deep learning algorithms, the trajectory of the muck trucks is predicted and analyzed, with an accuracy rate exceeding 60%, and the management ability of the muck trucks is comprehensively improved to effectively solve the problem of muck environmental pollution.
It ’s not “you die and live”
BAT is turning to the industrial Internet, all playing the AI ​​card. Shangtang, Yuncong, Despise, and Yitu are called “AI Four Dragons”, and the competition is not too fierce.
In Li Shuo’s view, it is still far from the stage of “you are dead or alive”, and you are all moving forward.But he also admitted that finance is a highly competitive industry.For example, a common competitive strategy is that when a BAT manufacturer shoots in its own field, the other two are likely to engage in tactical harassment.
Price wars are also common, and extreme cases such as a quotation of 0.1 yuan have appeared more than once.
Li Shuo also played this card, so far only once.
At that time, it was somewhat helpless. Baidu won a project, and there is another project that is closely related to the customer. If it can be obtained, two projects and an AI base will be smooth in the later stage.However, in the bidding and bidding process, the opponent may pay 0.1 yuan, and the bid will be dropped if not followed, and the previous project will also be affected.
“At that time, we can only say that we have done this for 0.1 yuan.” Li Shuo emphasized that the field of artificial intelligence still has to stick to the bottom line. The bottom line is that artificial intelligence brings value to customers, not that the product has no value. Take 0.1 yuan to emphasize itthe value of.
Even with fierce competition, Li Shuo believes that Baidu has an advantage in the financial industry.More and more companies provide financial products to C-end users, especially the mobile payment of Ali and Tencent makes financial institutions very anxious. Baidu does not grab business with financial institutions in the payment field and insists on technology empowerment cooperation. Some joint-stock banks and city commercial banksWelcome, cooperation continues to deepen.
In more industries, it seems difficult to see the differences in AI capabilities of giants. They all advertise that they have AI capabilities ranging from chips, algorithms, computing power to data.Ali has the Bodhidharma Academy, and Tencent has three major artificial intelligence laboratories, such as AI Lab, Youtu, and WeChat AI.
Wu Tian, ​​vice president of Baidu Group and deputy director of the National Engineering Laboratory of Deep Learning Technology and Applications, admitted to “Blue Hole Business” that the threshold of basic AI algorithms has been greatly reduced, and some of their own algorithms have been built based on deep learning frameworks or platforms.It becomes a lot easier.Baidu’s AI core competitiveness and barriers are mainly manifested in core technology accumulation, continuous application polishing, and multi-technology integration innovation.
In terms of multi-technology integration and innovation, for example, Baidu “Digital Man” can broadcast news, serve as a customer service, and report the weather. Behind this is a comprehensive AI technology solution, including speech recognition, speech synthesis, visual understanding and synthesis, natural language processing and other technologies.The facial expressions generated need to be natural, and at the same time the lips must match the language, including gestures, movements, etc., to coordinate with the appropriate language content.
Li Yanhong interacted with the digital person “Xiaopu” at the 2019 Baidu AI Developer Conference
Baidu Brain has become a software and hardware integrated AI production platform, including computing power and data platform, AI basic base deep learning platform, general AI capabilities, scenario-oriented custom training platform, easy deployment and integration platform for terminal use, plus AISafe escort escort has opened up the entire process of AI industrialization.
“Baidu Brain took shape in 2010, and Baidu AI has systematic technological advantages,” Wu Tian added.
Baidu’s deep learning platform PaddlePaddle is also worthy of attention.In 2013, Baidu began to develop the propeller. After three years, it was open source for developers. In 2018, it was upgraded to an open source deep learning platform.
The deep learning framework is an operating system in the intelligent age.Globally, the more influential deep learning frameworks are Google’s TensorFlow and Facebook’s PyTorch.
Wu Tian explained that the positioning of the deep learning framework is the underlying basic software, just like the OS, which can be understood as an operating system in the intelligent era.Deep learning frameworks and platforms connect hardware downwards and undertake applications upwards to help developers quickly and easily train applicable AI models, which are ultimately used in various AI business scenarios.More importantly, the deep learning framework will form a symbiotic and win-win AI ecosystem. Comparing the powerful Wintel alliance back then, one can imagine its value.
In Wu Tian’s view, TensorFlow and PyTorch were built earlier for the open source ecosystem, and have greater influence among developers, especially researchers.The accumulation of flying propellers over the years has formed some characteristics and leading technological advantages, will continue to build a developer ecosystem, and is committed to better supporting industrial development.
Can’t just learn fur
The AI ​​Zhongtai and Knowledge Zhongtai under the new business architecture of Baidu Intelligent Cloud are accelerating the commercialization of Baidu AI.
Wang Haifeng believes that the industry is undergoing intelligent upgrades. Enterprises need AI to achieve intelligence. At the same time, AI must be deeply integrated with enterprise application scenarios to better meet enterprise needs.But at this stage, most enterprises lack AI basic capabilities and platforms that support AI development and application.
The core capabilities of Baidu ’s AI platform include an AI capability engine and an AI development platform, which can help enterprises build autonomous capabilities for AI development and application, intensively manage enterprise AI capabilities and resources, and plan overall planning for an enterprise ’s intelligent upgrade.
State Grid Shandong Electric Power Company built its own AI platform based on Baidu’s BML machine learning platform to achieve rapid business applications.For example, by using the transmission platform visualization platform built by the AI ​​middle station, it can realize the safety inspection in various complex outdoor scenes.In the running scene, AI face recognition is applied to realize functions such as hall entry recognition, face swipe and electricity.
Baidu Knowledge Central Taiwan is a full-cycle, intelligent enterprise knowledge empowerment platform based on Baidu ’s accumulated knowledge graph, natural language processing, multimodal semantic understanding, intelligent search and other AI core technologies accumulated over the years.
Connect data of different sources and forms at the bottom layer, and the upper layer supports different business scenarios.The intelligent application of knowledge platform for enterprises includes intelligent knowledge base, industry knowledge graph, enterprise search, intelligent recommendation, intelligent question answering, and knowledge push.The “financial digital man” jointly launched by Baidu and Shanghai Pudong Development Bank has the role of knowledge center.
At present, AI is gradually landing in the Internet, finance, retail and other industries, and industries such as industry and agriculture are also continuing to explore.
Wu Tian analyzed that the problem of industrialization of AI at the beginning was due to mutual incomprehension.After several years of exploration, there have been some improvements, but the penetration rate of AI in the entire industry is not high, and even the penetration rate of a large number of industries is very low. The breadth and depth of AI landing needs to be improved.
It is mainly manifested in four aspects: first, AI has a lot of technology accumulation, the technology stack itself is deep and wide, and it will encounter various problems in specific industries, and it can only meet the needs of different scenarios in a multi-technology deep integration mode.To better solve practical problems, this requires a short period of precipitation and accumulation.Second, there is a shortage of compound talents who understand both AI algorithms and industry applications, or the consciousness and ability to look at problems from this perspective need to be vigorously strengthened.Third, the entire chain of AI technology is relatively long, and it needs to open up the ecological chain, such as from data, hardware, technology to system integration. This is not something that can be done by an enterprise. It needs to be refined by the social division of labor and gradually run in.Fourth, data and industrial security issues need to be resolved quickly.Especially in the financial and other industries, a large amount of personal privacy data is involved. The degree of digitization and intelligence of public safety-related institutions and enterprises has not reached the most perfect state, and the data shows decentralization, regionality, etc., to a certain extentAffect the further development of the role of AI basic data platform.
In addition, for Internet giants, including Baidu, have long understood C-end users, and AI industrialization has landed. Compared with traditional IT vendors, process management is inadequate.
Many To B companies regard Huawei’s “iron triangle” as a rule.Ren Zhengfei once described it as follows: one is to let people who can hear the gunfire make decisions; the second is to form a customer-oriented “iron triangle” combat unit with a working group composed of account managers, solution experts and delivery experts; the third is front-lineCombat should be changed from single-man combat by account managers to small-group combat.
Li Shuo bluntly said that many of his team’s talents were joined from Huawei, IBM, and Accenture. They are indeed learning from Huawei, learning at the business and delivery level, to be as reliable as traditional IT companies.However, we cannot learn to learn by hand, and only learn the fur. According to Huawei’s Audio-Technica, Li Shuo respects two-wheel drive.
On the one hand, in the face of market opportunities, sales and solution teams can drive product technology teams.Previously, Baidu had no business-driven genes, and all were driven by product technology.However, the rules are also very clear. Business-driven business should be responsible, from beginning to end, to ensure the maturity and delivery of products.
The other is product-driven. Especially in the field of artificial intelligence, it is difficult for ordinary sales to understand where artificial intelligence can generate value and where it cannot.For example, in the field of industrial quality inspection, Li Shuo initially led the product manager to convince customers to try.
Li Shuo believes that the reason why the Iron Triangle decision cannot be adopted in the process of AI industrialization is that in the face of many uncertain scenarios and risks, the three functional roles can easily vote 2: 1 to vote out of an opportunity.He encouraged the team to dare to take responsibility, either as sales and solution, or as product technology. When the two sides could not dispute, a person took the table and said that I did it, so that I can go faster.
At present, Baidu is unwilling to disclose the income scale of AI industrialization, but it can also be seen from its financial report. Its AI and cloud are difficult to completely separate.In March 2019, Baidu presented the cloud business report alone for the first time in the fourth quarter of 2018, with revenue of 1.1 billion yuan in this quarter, more than double the same period in 2017.In 2018, Baidu Cloud ranked fifth in the domestic public cloud, and became fourth in the first quarter of 2019.
Wang Haifeng believes that artificial intelligence is bringing human society into the era of intelligence.After the mobile Internet, Baidu is undergoing the test of a new proposition.

Tags : 36 krypton startingAutopilotBlack technologycapital marketcareducationEntrepreneurshipForefrontgameGo publicInformationInternet celebrity economyInternet of ThingsInternet of VehiclesinvestmentNew businessNew economyOccupationreal estatestockTechnologythe Internetunmanned