'Human society has quickly entered the era of wisdom, what is the core driving force of this era, productivity - is calculation!' At the 2018 Artificial Intelligence Computing Conference hosted by the Ministry of Information and Electronic Engineering of the Chinese Academy of Engineering, Academician of the Chinese Academy of Engineering Wang Endong, chief scientist of Inspur Group, gave this view.
Wang Endong said that a country's GDP has a significant positive correlation with its computing power. The top 5 countries in terms of global GDP are almost in line with the top 5 global server shipments. Today's top 10 market capitalists, such as Apple. , Amazon, Google, Facebook, Alibaba, Tencent, and so on, without exception, are among the top companies in the world in terms of server purchases – which means they are investing heavily in computing power. Said that computing power is productivity.
'Calculation' 'Algorithm' 'Data', known as the 'Troika' that pulls artificial intelligence, in the machine learning 'algorithm' continues to break through, the huge 'data' explosive growth today, 'calculation' can become artificial intelligence The booming power engine is highly anticipated.
In fact, reviewing the history of artificial intelligence, it is not difficult to find that computing power plays a key role in it. 'Turing invented the computer first, then invented artificial intelligence. It can be said that there is no artificial intelligence without calculation, and artificial intelligence makes the computing power. With the driving force of progress, there is a direction of development. 'Wang Endong said.
Since the introduction of artificial intelligence in 1956, it has gone through three stages: The first stage was in the 1960s and 1970s, when artificial intelligence tried to realize the logical reasoning of machined logic through computers, but it was finally difficult to achieve. The second stage was In the 1970s and 1990s, computer power had made great progress over the past few decades. At this time, trying to solve problems by establishing a computer-based expert system, but because of the small amount of data and too limited to empirical knowledge and rules, it is difficult Building an effective system. The third stage is the development of deep neural network technology in recent years, and it has gradually entered a period of rapid development.
'Why did artificial intelligence have a 30-year development stagnation between the second and third phases?' The American Academy of Engineering academician, UCLA professor Cong Jingsheng threw this question at the conference.
In his view, in recent years, artificial intelligence has been able to break out again. On the one hand, the Internet, informationization, and digitization have brought big data. According to statistics, 90% of all data obtained by human civilization are the past two. The number of data generated by the world will reach 44 times today by 2020.
How much data is generated, stored, interconnected, and processed? It relies on computing. This is what Cong Jingsheng calls 'the other side': the improvement of computing power. In the 1980s, the computers used by people, It can execute 2 million to 3 million instructions per second, and now it can have 100 billion to 200 billion instruction operations per second.
From this perspective, it is calculations that light up artificial intelligence. Cong Jingsheng said, 'Because of these computing powers, today's artificial intelligence is everywhere.'
Of course, artificial intelligence in turn puts more demands and challenges on computing. For example, the demand for computational power of artificial intelligence is far beyond the performance growth rate of Moore's Law.
In other words, we need more computing power.
The conference released the "2018 China Artificial Intelligence Competency Development Report", which mentioned that with the passage of time, artificial intelligence will be used more and more in emerging economies and the digital economy -
From now until 2020, smart city technologies such as face recognition, speech recognition, natural language processing and other technologies such as biometrics and vehicle identification, smart communication, and intelligent street lights will be the most typical application scenarios for artificial intelligence; and 2020~2025 The technologies related to smart manufacturing and smart home will mature and become the most typical artificial intelligence application scenarios; in 2025 and later, related technologies and policies such as smart medical, automatic driving, and intelligent assistants will be formed to promote artificial intelligence applications in these industries. Achieve explosive growth.
The report also mentioned that the main challenges that hinder the development of artificial intelligence computing are four aspects: First, the development of computing power has not yet reached the demand; Second, the amount of available data is limited; Third, from the laboratory to the actual application process, it also faces Many challenges and problems; Fourth, it takes time from application scenarios to providing comprehensive industry solutions.
Gao Zhongqi, director of the Second Bureau of the Chinese Academy of Engineering, said that although the development of artificial intelligence applications in China is very fast, compared with developed countries, especially the United States, we still have significant gaps in the field of artificial intelligence core technology of hardware algorithms.
In his view, although the development of application terminals has been far ahead of the hardware architecture, it is now difficult for computing platforms to meet the ever-increasing computing needs of artificial intelligence. How to strengthen the underlying architecture and improve computing power has become artificial intelligence. The key issue of development.