Smooth natural language generation, knowledge system coverage in all fields, and smooth human-computer interaction interface

Accelerating the Development of Artificial Intelligence (Technology Perspective)

Reporter Wu Yuehui and Gu Yekai

The Central Economic Work Conference proposed to vigorously promote new industrialization, develop the digital economy, and accelerate the development of artificial intelligence.

On October 2023, 10, China released the Global Artificial Intelligence Governance Initiative during the 18rd Belt and Road Forum for International Cooperation, systematically expounding China's plan for AI governance in three aspects: AI development, security and governance.

In recent years, with the rapid development of global artificial intelligence technology, it has become an important force to promote the accelerated development of science and technology and industry, and has a far-reaching impact on economic and social development and the progress of human civilization. What is the current status of the development of artificial intelligence technology, what are its applications, and what are the future trends? The reporter interviewed relevant experts.

The ability of AI to handle complex tasks has been greatly improved

At present, artificial intelligence technology has entered the practical stage and is profoundly changing the production and life of human beings.

"In the past 70 years of development, artificial intelligence has gone through three stages: instilling rules, instilling knowledge, and learning Xi from data. In recent years, the rapid development of artificial intelligence large model technology in the world, the basic model on which it is based is based on 'big data + large computing power + strong algorithm' training, which is a typical embodiment of the third stage of artificial intelligence development. Huang Tiejun, president of Beijing Zhiyuan Artificial Intelligence Research Institute, said.

At present, all kinds of artificial intelligence models are in rapid development, and many high-tech enterprises around the world have devoted themselves to the construction of artificial intelligence models.

"At present, a relatively mature technical framework has been formed around the artificial intelligence model, but the product and ecology are still being developed. Zeng Dajun, deputy director and researcher of the Institute of Automation, Chinese Academy of Sciences, said, "Overall, the technological development process of artificial intelligence large models is more rapid than any previous artificial intelligence technology, and its influence is unprecedented." ”

The emergence of large artificial intelligence models has opened up a new imagination space for the realization of general artificial intelligence, and greatly improved the ability of artificial intelligence to deal with complex tasks.

"For example, chatbots based on large language models with artificial intelligence can achieve high-quality information integration, translation, and simple problem solving and planning. Zeng Dajun said, "This kind of robot has attracted attention mainly because it has initially possessed some characteristics of general artificial intelligence, including smooth natural language generation, knowledge system coverage in all fields, general processing models across task scenarios, and smooth human-computer interaction interfaces." ”

However, there are still limitations to the current AI large model capabilities.

"On the one hand, due to the loopholes in the structure and mechanism of the AI model, there is a risk of being maliciously attacked, and on the other hand, there are still defects in the knowledge expression and learning Xi mode of the AI model, resulting in common sense errors and fabricated content in its answers. Zeng Dajun said, "Artificial intelligence scholars are tackling these problems. ”

Artificial intelligence is accelerating to a new stage of comprehensive application

"I am a first-year computer science major, and I want to take an artificial intelligence course, what do I need to prepare?" "You need to learn Xi basic mathematical knowledge, programming languages, learn Xi machines, learn algorithms Xi, and pay attention to technology trends...... This conversation is not between teachers and students, but between students and artificial intelligence.

In August 2023, Zhejiang University, together with Higher Education Press, released the "Zhihai-Sanle" education vertical model, which can provide services such as intelligent question answering, test question generation, Xi navigation, and teaching evaluation on the basis of corpus and professional instruction datasets such as core textbooks, field papers and dissertations, which has been applied in many universities.

"We divide these textbooks into sentences, paragraphs, and chapters to 'feed' the large model, and these high-quality corpora will synthesize the probabilistic associations between words to inspire students. Wu Fei, a professor at Zhejiang University, said.

Industrial Quality Inspection, Knowledge Management, Code Generation, Voice Interaction...... At present, China's artificial intelligence is deepening from single-point application to diversified application, from general scenarios to industry-specific scenarios, and accelerating towards a new stage of comprehensive application. In particular, with the breakthrough of artificial intelligence models and the rise of generative artificial intelligence, artificial intelligence can better deal with complex problems in production and life, and provide more advanced tools and means for various industries to achieve product and process innovation.

Predicting the path of a typhoon in the next 10 days used to take 3000 hours on 5,10 servers for simulation, but now based on the pre-trained Pangu meteorological model, more accurate prediction results can be obtained within 4000 seconds......

"The rapid development of the generative AI industry driven by AI models has brought huge innovation opportunities in many fields such as scientific exploration, technology research and development, artistic creation, and business management. Wang Endong, academician of the Chinese Academy of Engineering, said.

Driven by both supply and demand, technological innovation has begun to move from laboratory research to industrial practice on a large scale, and the process of artificial intelligence industrialization has been accelerating. According to incomplete statistics, as of October 2023, China has released more than 10 artificial intelligence models, and scientific research institutes and enterprises have become the main force in development.

According to Shang Haifeng, President of Huawei Hybrid Cloud, innovative technologies represented by artificial intelligence are accelerating the reshaping of various industries.

Zhao Zhiyun, Director of the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology, said: "Artificial intelligence technology is continuing to evolve in the direction of pursuing higher precision, challenging more complex tasks, and expanding the boundaries of capabilities. Scenario innovation has become a new path for AI technology upgrading and industrial growth. ”

Liu Jun, senior vice president of Inspur Information, believes that in the future, artificial intelligence needs to further deepen application scenarios and empower specific industrial links. "This process is difficult to be completed by a single manufacturer alone, and requires more in-depth collaboration in the industrial chain and innovation ecosystem. Liu Jun said.

More general-purpose artificial intelligence is expected

Experts said that the third stage of development of artificial intelligence, represented by the artificial intelligence model, will have a long period of development dividends in the future, and will become an important driving force for a new round of scientific and technological revolution and industrial transformation.

The Institute of Automation of the Chinese Academy of Sciences has made a judgment on the evolution of artificial intelligence models, and Zeng Dajun introduced his views: the application and innovation ecology is undergoing drastic changes or at least has the potential for drastic changes, artificial intelligence models promote the rapid development of decision-making intelligence, the miniaturization of artificial intelligence models and the need for field specialization are very urgent, and more general artificial intelligence is expected to be realized.

Zeng Dajun said: "The artificial intelligence model is like the prototype of a human brain, which realizes various intelligent capabilities by 'feeding' various data. Artificial intelligence models are redefining the interaction between humans and computers, and are expected to become the main interface for human-computer interaction. ”

Zeng Dajun emphasized the development of miniaturization and domain of large AI models. He said that the challenges of computing power and energy consumption of existing AI large models will prompt a lot of work to develop in the direction of domain specialization, lightweight small models or mixed and matched large and small models, especially in the fields of finance, education, medical care, transportation, etc., and strive to reduce the cost of large models.

Huang Tiejun believes that artificial intelligence will develop from information intelligence to physical intelligence, and visual and embodied artificial intelligence models will be the next flashpoint. "Big data is the expression of the world, and the language cognitive model trained from it can support information services, and the language model can improve the intelligence level of autonomous driving, robots and other entities, but it also needs the development of visual, auditory, embodied, interactive and other technologies. ”

Huang Tiejun told reporters that the current emergence of intelligence is only a static emergence, and it does not yet have the dynamic emergence ability of the human brain. "In the future, it is expected that artificial intelligence with real dynamic emergence ability will be realized through brain-like intelligence. ”

Zeng Dajun believes that the large AI model is expected to develop into a more general AI. "In the near future, the artificial intelligence model will go beyond the information domain, combined with hardware facilities, and develop into concrete intelligence that interacts with the physical and human world, and gradually narrows the gap with the real 'general artificial intelligence'. ”

"It is expected that the degree of intelligence will continue to improve in the future, and the driving and impact on various industries will be more profound, which is difficult to match with other technologies. Huang Tiejun said.

In Zeng Dajun's view, the integration and innovation of decision science and artificial intelligence can collide with a lot of sparks.

"Artificial intelligence, especially with the blessing of automatic reasoning of knowledge, can realize the leap from the known to the unknown. Zeng Dajun further explained, "Higher-level decision-making functions, such as the research and judgment of the industrial situation, risk assessment, etc., this kind of decision-making from the unknown to the unknown, machine intelligence may play a great auxiliary role." ”