Traditional enterprises are transforming into artificial intelligence and relying on big data alone

"Even with the support of big data, if it is not closely integrated with use scenarios, no matter how superb technology consumers are, they will not buy it. Users don't care about AI technology or not, they care more about content. ”

In the past year, artificial intelligence is still the top keyword in the Internet world. Just like AlphaGo upgraded to AlphaZero, dominating all kinds of board games without a teacher, AI is rushing into the future at a speed that humans can't predict, like a wild horse. What role will China play in this?

In July 2017, the State Council issued the "New Generation Artificial Intelligence Development Plan"; In December 2017, the Ministry of Industry and Information Technology (MIIT) issued the Three-Year Action Plan for the Development of the Artificial Intelligence Industry (hereinafter referred to as the "Plan"), proposing to strive to make important breakthroughs in a series of landmark AI products by 2020 and form international competitive advantages in several key areas.

China has an advantage in industrial application

In 2017, a sign of the progress of artificial intelligence was that ImageNet, an image recognition challenge held in the United States for eight years, officially withdrew from the stage of history. Eight years ago, the accuracy rate of AI recognition was only more than 70%, and now the error rate is only 2.7%, far lower than the 5.1% of humans, and the competition is meaningless. Recent reports show that China's medical AI has also surpassed senior doctors in its ability to identify small tumors.

Sooner or later, the AI will conquer more human territory, and in which area will it blossom first?

Dai Wenyuan, founder and CEO of the artificial intelligence company "Fourth Paradigm", told the Science and Technology Daily reporter that there are five prerequisites for the explosion of AI applications: big data (Big-data), feedback data (Response), algorithm (Algorithm), infrastructure (Infrastructure), and business needs (Needs), collectively known as "BRAIN", and any field that meets these five conditions is likely to achieve the explosion of artificial intelligence before 2020.

"We can also see that the commercialization scenarios listed in the plan are concentrated in the fields of transportation, medical care, image, voice, and home, which are also relatively complete scenarios for these five prerequisites." Dai Wenyuan said.

It has been judged that it will take 5-10 years for AI to be implemented on the consumer side, but Ji Yue, a partner of Sequoia Capital China, does not think it will take that long. "For example, Toutiao, as long as users see the content they want to see, users don't care whether there is 'AI technology'." Ji Yue said, "Do you use NetEase Cloud Music?" Before you know it, you're enjoying the convenience of AI. When AI will be realized on the consumer side does not depend on the judgment of experts, as long as the needs of users are actually solved, it will be realized. ”

Sequoia Capital has invested in many Chinese AI companies: Toutiao, Kuaishou, Miaopai, JD Finance, Meituan, Didi, Mobike, DJI, ...... Ji Yue said that AI needs to be closely integrated with the use scenario, otherwise no matter how superb the technology, consumers will not buy it.

Big data is the key to training AI. "What Sequoia Investment values is the application scenario and its own data." In 2015, Didi became the world's second-largest online trading platform after Taobao, with a data scale of 20 billion path planning requests per day, so its AI predicts user destinations with an accuracy of more than 90%.

Dai Wenyuan is extremely confident that China will form advantages in a number of key areas as proposed in the "Plan." "In the field of AI, it is basically a pattern of competition between China and the United States, and China still has a certain lead in the application of the AI industry." "This is because the algorithms and computing power of artificial intelligence can be realized through the platform, and data has become the biggest threshold, and the huge user data is China's advantage," Dai said. China is a country with a large Internet population, and the development of the Internet in China has prepared the conditions for us to study AI. ”

The integration of artificial intelligence and the real economy needs to be deepened

"China's development stage is condensed, showing leapfrog and parallel development, which may bring more opportunities to Chinese entrepreneurs." Ji Yue said that in China, there are many innovative companies that combine several stages as one (including AI), while in the United States, IT giants have accumulated deeply, and emerging companies need to find a gap to cut into, and from this perspective, China's entrepreneurial opportunities are greater.

Transfer learning is considered a breakthrough point where China may catch up with the United States in the field of AI. Dai Wenyuan, who created the largest machine transfer learning system in China, "Phoenix Nest", is the world's third most cited paper in the field of transfer learning, and Yang Qiang, the chief scientist of 4Paradigm, is the world's first.

Transfer learning refers to the ability to apply knowledge to different domains to solve the problem of insufficient data in the target domain. Because of the problem of applying AI to small data, China's research has attracted the attention of Google and Facebook.

The plan proposes that the integration of artificial intelligence and the real economy needs to be further deepened. Dai Wenyuan said that this requires the adjustment of the organizational structure of enterprises and the embedding of AI into business management. Dai Wenyuan said that the production and operation of traditional enterprises are made by manual decision-making and machine execution; The significance of AI lies in machine decision-making and machine execution. If you want to become an intelligent enterprise in the future, the threshold is also very high, including machine learning algorithms, hundreds of millions of dimensional system architecture, and even more cognitive level.

"Many traditional enterprises will say that they have big data and can do AI. In fact, not necessarily, the right big data plus the right machine learning produces artificial intelligence, and the wrong big data plus machine learning does not. Dai Wenyuan said that the core system of enterprise artificial intelligence will be another core of the enterprise after the core system of trading and the core system of the Internet.

The plan specifically emphasizes "supporting AI enterprises and financial institutions to strengthen docking and cooperation". Dai Wenyuan commented that the financial sector has sufficient data and computing resources, making it one of the most ideal industries for AI to land. "Machines are already helping people do a lot of business analysis and decision-making in the financial field." "For example, in our cooperation with a bank to solve the problem of transaction fraud, the number of fraudulent transaction rules written by machines reached 2.5 billion, and the number of rules written by senior business experts was up to 1,000. In the end, the accuracy of the machine in fraud prediction was increased by 7 times compared with humans. ”

Yang Yuanqing, chairman of Lenovo Group, also expressed optimism about China's AI development. "The future business competition must be a competition of business models, and the degree of intelligence of the business model determines the competitiveness of enterprises." Yang Yuanqing said, "If more emphasis can be placed on the digital economy and intelligence, then in the future, Chinese enterprises will be one step ahead of other countries." ”