◎ Reporter Ni Lin

With the release of the Global AI Governance Initiative, AI governance has once again attracted global attention. As a strategic force leading the scientific and technological revolution and industrial transformation, the wide application of artificial intelligence is profoundly changing the economic and social structure and reshaping people's lifestyles.

At the same time, AI will also bring about problems such as changing the employment structure, impacting law and social ethics, and violating personal privacy, which will have a profound impact on government management, economic security and social stability, and even global governance.

At present, how to make artificial intelligence pick up its original intention and balance its security and development? A reporter from Science and Technology Daily recently invited Shen Changxiang, academician of the Chinese Academy of Engineering, Wang Haifeng, chief technology officer of Baidu, and Deng Zhidong, director of the Visual Intelligence Research Center of the Institute of Artificial Intelligence of Tsinghua University, to discuss and interpret the development of artificial intelligence.

Original meaning: artificial and intelligent, which is more important

Reporter: The definition of artificial intelligence first appeared at the Dartmouth Conference in 1956. From the perspective of your practice, what is artificial intelligence?

Shen Changxiang: Artificial intelligence refers to the use of human design and programming to enable intelligent machines to complete a variety of complex tasks, thereby replacing traditional human roles.

Wang Haifeng: Artificial intelligence is the science of researching and developing theories, methods, technologies and application systems that simulate, extend and extend human intelligence. At this stage, artificial intelligence has become a strategic technology leading a new round of scientific and technological revolution and industrial transformation, which is driving the fourth industrial revolution and bringing human society into the intelligent era.

Deng Zhidong: From a practical point of view, AI has the ability to simulate humans to solve various complex tasks.

Reporter: With the emergence of ChatGPT and other products, some people believe that the role of humans is gradually replaced by intelligence, and some people think that intelligence is just a tool to assist humans. What do you think about the trade-off between "human" and "intelligent" in AI?

Shen Changxiang: Intelligent machines are intelligent products evolved from human intelligence, which can provide personalized services according to human preferences and needs. All actions and behaviors of intelligent machines are programmed and controlled by humans, and it is impossible to completely detach them from human control.

Wang Haifeng: The development of artificial intelligence should be "people-oriented" to improve human well-being. With the rapid development of artificial intelligence technology represented by large language models, Wenxin Yiyan and ChatGPT can talk to people, assist in creation, and improve people's work efficiency and life experience. Essentially, it is still a productivity tool, replacing humans with some of the work and creating more jobs. In the future, humans will act as managers, and artificial intelligence will assist people, collaborate with people, improve work efficiency, and stimulate creativity. All sectors of society should have a correct understanding and reasonable expectations of AI, and promote the development of AI to better serve mankind.

Deng Zhidong: "Artificial" and "intelligent" are inseparable. Artificial intelligence refers to the human intelligence realized by machine simulation, and the relationship between the two should be called the relationship between artificial intelligence and humans. In the short term, artificial general intelligence is an auxiliary tool that helps humans accomplish challenging tasks. In the long run, interactive AI will promote the "integration" of humans and agents, and move towards a new world of "human-machine integration".

Positioning: Growth product or infrastructure

Reporter: Artificial intelligence has had an important impact on the development of related industries. Do you think of AI as a growth product or infrastructure?

Shen Changxiang: AI was originally a stand-alone product, and later it was integrated into a device with various functions. A system is formed when devices are connected to each other and work in harmony. The convergence and interaction of these systems forms the framework of the AI infrastructure that provides us with a variety of products and services.

Reporter: In the field of artificial intelligence infrastructure construction, what aspect do you pay special attention to?

Shen Changxiang: Security is of paramount importance. Artificial intelligence derives major risks such as artificial intelligence self-errors, malicious attacks on intelligent hardware, and loss of control of robotic intelligent actors. It is necessary to strengthen research and prevention against these problems, and build a safe and credible guarantee system composed of laws, strategies and systems to ensure the safe, stable and sustainable operation of AI infrastructure.

Wang Haifeng: I have long been concerned about the Xi platform and large models of deep learning. The deep learning Xi platform is equivalent to the operating system in the intelligent era and is at the key core position of the artificial intelligence technology system. Deep learning Xi presents the characteristics of standardization, automation, and modularization, which promotes artificial intelligence to enter the stage of industrial mass production. With the popularization of deep learning Xi technology and the breakthrough of pre-training technology, large models have become the mainstream direction of artificial intelligence development, with the characteristics of good effect, strong generalization, and standardization of R&D process, and are becoming a new foundation for artificial intelligence technology and application.

Deng Zhidong: From the basic large language model of text unimodality to the native multimodal large language model, it is necessary to solve the API access problem of high-level artificial intelligence general large model, and promote the construction of "self-oriented" artificial intelligence infrastructure, including artificial intelligence chips and cloud services, and the ecological construction of open source frameworks. In the field of visual intelligence, it is necessary to lay a good foundation, promote the construction of artificial intelligence infrastructure at the national level, make up for shortcomings, and enable the development and application of large models from foundation and pedestal to vertical models.

Reporter: Enterprises may face dual needs when using artificial intelligence. Can you talk a little bit from experience about how companies can maximize the value of AI in this situation?

Shen Changxiang: Enterprises need to consider the needs of industrial digitization, digital industrialization, and social intelligence, and make full use of artificial intelligence technology to improve production efficiency and service quality. At the same time, we should attach great importance to the cultivation and introduction of talents in the field of artificial intelligence, and promote technology research and development to maintain a leading position in technology. We should also strengthen cooperation with other industries, and constantly explore new business models and application scenarios to provide more convenient and efficient services for the society.

Wang Haifeng: Baidu is not only an AI infrastructure builder, but also a leader in AI technology innovation. Artificial intelligence technologies represented by the PaddlePaddle Deep Learning Xi Platform and Wenxin Large Model have become an important infrastructure in the intelligent era, enabling manufacturing, energy, agriculture and other fields, and driving the development of new technologies, new products, and new business formats. The implementation of the large-scale model industry can adopt the model of "intensive production and platform application", that is, enterprises with comprehensive advantages in algorithms, computing power and data can encapsulate the complex process of model production, and provide services for thousands of industries through a low-threshold and high-efficiency production platform. The Wenxin model supports the development of a large number of AI native applications, reconstructs existing products, and conceives new products "from 0 to 1", building an AI native application ecology and promoting high-quality economic and social growth.

Deng Zhidong: The development of the general artificial intelligence industry, represented by multimodal large language models, is still in the early stage of the industry. Different enterprises should have different emphasis, and the leading artificial intelligence enterprises should benchmark the foundation and base large model products with the world-class level to shorten the international gap of high-performance AI chips. Small and medium-sized enterprises should actively carry out the research and development of artificial intelligence value and application products, focusing on the diversity of application scenarios, the speed of implementation and the innovation of business models.

Risk: Everywhere, who is more worried

Reporter: Is the risk of artificial intelligence widespread at present? Please explain with a specific case.

Shen Changxiang: AI risks are closely related to cybersecurity. Network security is the foundation of the development of artificial intelligence, and its risk degree = vulnerability × threat degree, in which the vulnerability comes from the lack of offensive and defensive security concepts in Turing computing principles, the lack of protective components in the von Neumann architecture, and the lack of security governance services for major engineering applications. This makes cyberspace extremely vulnerable, with an increased risk of cyberattacks. On May 2021, 5, Colonial Pipeline, the largest U.S. oil product pipeline operator, was forced to shut down its smart oil supply network in the states along the eastern coast of the United States after a ransomware attack, and the U.S. government declared a state of emergency in 7 U.S. states and Washington, D.C.

Wang Haifeng: From the perspective of previous industrial revolutions, the technologies that drive scientific and technological revolutions and industrial transformations, whether they are mechanical technology, electrical technology, information technology, or current artificial intelligence technology, will bring great value to human beings, but also bring certain risks. However, with the advancement of science and technology, human beings have enough ability to control risks and give full play to the value of new technologies.

Deng Zhidong: The development of general artificial intelligence has exceeded most people's expectations. As long as artificial general intelligence develops, its risks will inevitably be widespread. From law, ethics, privacy, to employment, etc., all fields, tracks, and industries are facing challenges and risks of rewriting and reshaping.

Reporter: In specific fields such as autonomous driving, smart medical care and face recognition, are the risks of artificial intelligence more prominent?

Shen Changxiang: Technologies such as autonomous driving, smart medical care, and facial recognition will threaten human life and property safety once they go wrong. IT system design can not exhaust all logic, the risk of exploiting vulnerabilities to attack by using logical flaws has always existed, the traditional "blocking and killing" is difficult to deal with unknown malicious attacks, and the idea of finding vulnerabilities and patching is not conducive to overall network security, so it is necessary to reduce the risk of specific areas of artificial intelligence by changing from passive defense to active immunity.

Governance: whether security and development can go hand in hand

Reporter: With the continuous development of artificial intelligence technology, issues such as data privacy and security, and energy consumption are becoming increasingly prominent. Do you think that in response to these problems, should we develop first and then govern or govern while developing?

Shen Changxiang: It is necessary to pay attention to the balance between governance and development. It is necessary to build a new model of active immunity trusted computing, create a parallel architecture with computing components and protection components as the core, and build a protection framework for active immunity supported by a trusted security management center, so as to realize trusted dynamic access control of subjects, objects, operations, and environments, and ensure that architecture, resource allocation, operation behavior, data storage, and policy management are credible. In this way, it can achieve the protection effect that attackers cannot enter, unauthorized people cannot get important information, steal confidential information and cannot understand it, the system and information cannot be changed, the system work cannot be paralyzed, and the attack behavior cannot be relied on, and the security of artificial intelligence is maintained in an all-round way.

Wang Haifeng: It is necessary to adhere to the principle of attaching equal importance to development and security, and support and encourage innovation and development while ensuring security. Only by continuing to develop artificial intelligence technology can we find and solve problems in development and grasp the initiative in scientific and technological competition.

Reporter: Judging from your experience, how should the industry or enterprises prevent potential security problems of artificial intelligence?

Shen Changxiang: It is necessary to analyze and study scientifically and rigorously, adhere to independent innovation, grasp the core competitiveness, and grasp the active advantage. Based on domestic software and hardware equipment with trusted computing functions, and with the support of complete laws, strategies and systems, a complete trusted computing product chain should be established as soon as possible, so as to create a new ecosystem of secure and trustworthy artificial intelligence and build a solid foundation for the development of the digital economy.

Wang Haifeng: Correctly understand the potential security problems of artificial intelligence. The development and application of any technology will bring potential security problems, and artificial intelligence, as a strategic technology driving scientific and technological revolution and industrial transformation, has attracted more attention. Technical security problems can be solved by technical means; On the one hand, the government should support independent and controllable AI technologies and applications, and on the other hand, it should establish and improve relevant laws and regulations to promote the development of AI technologies and applications under appropriate standards and clear standards.

Deng Zhidong: There are still many uncertainties in the development of artificial intelligence on security issues. Industries or enterprises need to look at as they go, go deeper, and be careful to deploy for the benefit of mankind.

Reporter: AI governance has become an important issue for international competition and cooperation. What are your views on promoting international cooperation on AI governance?

Shen Changxiang: We should conscientiously implement the Global AI Governance Initiative, start from the three aspects of innovation and development, security and governance, comprehensively safeguard China's AI infrastructure, and actively go abroad to build a community with a shared future in cyberspace with global partners.

Wang Haifeng: The research in the field of artificial intelligence itself is open and collaborative, and AI governance is one of its research hotspots, and all parties are very concerned. Experts and scholars from academia and industry around the world can carry out cooperation and exchange and contribute wisdom through open source communities, project cooperation, academic conferences, seminars and other ways.

Deng Zhidong: The governance and international competition and cooperation of AI have gone beyond academia and industry, and stronger international cooperation is needed in order to achieve international consensus and build binding, consistent and balanced international norms as soon as possible. (Source: Science and Technology Daily)