On artificial intelligence and how to develop,teach and govern it

1. What is AI?

Before discussing other aspects of AI, we should first define it based on its characteristics. AI is a technology imitating human intelligent behaviors (such as perception and cognition). It relies on artificially designed hardware systems (such as integrated circuits) and trained by artificially designed models (such as deep neural networks, etc.) with data and rules provided directly or indirectly by humans. It is to note that Humans play the crucial role throughout the entire process of the research, development, deployment, application, operation, and management of this technology.

Via technological advances triggered by the digital revolution, AI continues to drive innovation through integration with big data and greater computing power. A prime example is the launch of ChatGPT in November 2022. ChatGPT is a powerful tool that leverages large models with trillions of parameters and a suite of advanced technologies (such as word embedding, self-attention, human feedback reinforcement learning and value alignment). It has quickly garnered a large number of users with its unexpected ability to imitate human in the generation of texts, audios and videos, and this has expanded the application of AI to an unprecedented breadth and depth. In August 2023, McKinsey & Company released a comprehensive global survey report on AI, which confirmed the explosive growth of generative AI (GenAI) tools. According to the survey, individuals across regions, industries and seniority levels are using GenAI both for work and outside of work: 79% of all respondents say they have had at least some exposure to GenAI, either for work or outside of work; and 22% say they regularly use it in their own work (McKinsey & Company, 2023).

The powerful capabilities of GenAI have been both awe-inspiring and unsettling. Given its rapid development, some important questions about AI warrant exploration: What is AI's relationship with humanity? Can AI be placed under effective control? Will AI terminate the very existence of humanity? In March 2023, over 1000 prominent figures from the technology and business sectors, including Turing Prize winner and AI expert Yoshua Bengio and Tesla CEO Elon Musk, signed an open letter expressing concerns about the potential risks of AI to society and humanity and appealing for a pause in the development of AI large models until the risks can be reliably managed (Future of Life Institute, 2023). However, throughout history, from manual to mechanical weaving, from horse-drawn carriages to automobiles and trains, from manual to automated production lines, and from the abacus to the computer, humans have consistently developed new tools to enhance their physical and mental capabilities, and therefore to expand the productivity of human society. This is a historical process that is not subject to human will; therefore, rather than stagnating, the development of AI technology has accelerated since the launch of GPT.

In October 2018, during the ninth collective study session of the Political Bureau of the 19th CPC Central Committee, General Secretary Xi Jinping pointed out that the information revolution has significantly amplified human cognitive abilities, leading to a qualitative leap in productivity. He emphasized that “AI is a strategic technology at the forefront of this round of scientific revolution and industrial transformation, and, like the ‘head goose’, possesses a significant catalytic effect” (Xinhua News Agency, 2018). Just as mechanical and electrical technologies once propelled the mechanization and electrification of the economy and society, AI, as a new revolutionary general-purpose technology, is driving the evolution towards a smarter economy and society. Consequently, as an emerging advanced force of production, the developments in AI represent an inevitable progression that we must accept and embrace.

In recent years, the integration of large AI models with a suite of advanced technologies, such as word embedding, multi-head self-attention and reinforcement learning from human feedback, has promoted AI development to new heights and reshaped our way of production and our daily life. One notable innovation is Model as a Service (MaaS). Using pre-trained general-purpose models and fine-tuning them with specialized data and knowledge, we can finish a vast array of downstream tasks. The application of large models has also created new jobs, such as GPT engineer. However, AI development is still not mature. As noted in an OpenAI report:

Despite its capabilities, GPT-4 has similar limitations as earlier GPT models. Most importantly, it still is not fully reliable (it ‘hallucinates’ facts and makes reasoning errors). Great care should be taken when using language model outputs, particularly in high-stakes contexts, with the exact protocol (such as human review, grounding with additional context, or avoiding high-stakes uses altogether) matching the needs of specific applications. (OpenAI, 2023)

The risks associated with AI applications stem from two primary sources: the inherent imperfections in the technology (which can lead to factual, ethical and logical errors) and the possibility of malicious use and overdependence on AI tools.

Technically, GenAI needs to overcome issues such as lack of interpretability, data dependency, low energy efficiency and high-level carbon emission. Application-wise, it faces the challenges of reliability, trustworthiness and security. In regard to governance, while numerous principles have been proposed, tools for governance are still lacking. In addition, technical standards are necessary, and regulations should be agile in responding to AIs rapid development. It is critical to overcome the impact of geopolitics and push for global co-governance. In the context of education, defining the academic discipline of AI—by nature, a general-purpose technology—is an important challenge. The lack of qualified educators with robust theoretical foundation and necessary practical experience is also an important issue that needs to be addressed.

In summary, AI is a revolutionary general-purpose technology and an irresistible advanced productive force. After nearly 70 years of development since it was first proposed in 1956, AI has penetrated into human production and life with unprecedented breadth, depth and speed, creating new working modes, jobs and lifestyles. However, despite significant breakthroughs, AI is still in its infancy. Its applications will continue to expand and unfold at a faster pace, yet the process also brings with it varying degrees of ethical, security and other risks. The historical task we are now facing is to fully unleash the potential of AI in a manner that benefits all of humanity while ensuring that AI development is under control.

2. How to develop AI?

Facing an advanced yet immature technology like AI, it is essential to balance development and governance. Development goes first, and governance is for healthy development.

In terms of AI development, borrowing the term ‘deep learning’, we need to pursue deep innovation and deep application. For deep innovation, we must reinforce foundational research and seek breakthroughs in core algorithms, thereby laying a robust theoretical groundwork for AI's evolution through in-depth exploration of its fundamental theories and methodologies. We must vigorously develop a new type of intelligent model driven by both data and knowledge (theory) and enhance AI's explainability by combining big data and professional knowledge. We must attach more importance to developing brain-like computing and explore new computing models and algorithms to elevate AI's cognitive capabilities. For deep application, we must actively integrate AI into the real economy. By developing industrial models, AI technology could match to the specific needs of various sectors to enable industrial upgrade and innovation. We should promote AI for science, using AI technology to tackle complex scientific problems and advance scientific breakthroughs. We must actively develop the MaaS platform, encourage open-source models, and construct shared AI infrastructures to lower the threshold of AI application and promote the adoption and evolution of AI technology.

In terms of AI governance, we must develop supportive technologies such as privacy computing, trustworthy computing and unbiased algorithms to ensure AI's compliance with ethical, legal and social standards. We must also promote data sharing, while strictly protecting user privacy and data security. In pursuing innovation, we must focus on overcoming the challenge of explainability to lay the foundation for achieving AI trustworthiness.

In summary, pushing for deep integration of AI and real economy should be the central task and overarching goal. We must promote deep innovation of AI technology, particularly in AI algorithms. On this basis, we should foster a holistic innovation ecosystem with the alignment of algorithms, data, computing power and application scenarios, promote the integration of hardware and software, coordinate all relevant units and systems, and synchronize of standard-setting and testing-verification, in order to accelerate the evolution of AI-powered advanced productivity. Furthermore, we must break through the boundaries between disciplines, industries, ownership structures, regions and countries to align with the development needs of AI as a general-purpose technology; nurture comprehensive and open platforms for innovation and application; and forge new relations of production that support the growth of emerging productive forces.

3. How to teach AI?

The evolution and application of GenAI present a myriad of challenges for both educators and learners. The transformation of society and economy powered by AI productivity is an unstoppable historical process. In this process, education plays a fundamental, overarching and pioneering role. As such, education must be at the forefront of intelligentization. The objective of AI education is to nurture qualified citizens and innovative leaders who are well-equipped to thrive in the intelligent era. They must possess moral ethics, a solid scientific foundation, robust learning abilities and innovative ability in the era of intelligentization. AI education is, essentially, an educational transformation driven by the new generation of information technology, with AI at its core. It demands not only the teaching and learning of general and specialized knowledge of intelligent technologies, but also calls for the construction of an educational system that integrates digital thinking, methods and skills. The system includes digital campuses, classrooms and facilities that blend reality with virtualization; digital and intelligent teaching programmes that encompass general and specialized knowledge; digital cultivation processes that coordinate standardization and individualization; and digital learning modes that combine theory and practice, teaching and learning, and teachers and students.

To promote the comprehensive and profound transformation of the educational paradigm, we must put ethics first, strengthen the knowledge base, focus on practical application, think out of the box, and advocate open-source practices. First, we should encourage discussions about AI ethics, guiding students to explore the ethical, security and privacy challenges that AI technology may bring, and thus nurture their sense of responsibility and ethical consciousness. Second, in the face of rapidly evolving science and technology, it is crucial to establish a robust scientific foundation and foster the ability to learn independently. Third, we can use intelligent tools effectively in AI education and incorporate AI knowledge and skills with various disciplines to help students master AI technology through hands-on experience and enhance their practical skills and innovative spirit. Besides, we can adopt teaching methods and content tailored to the needs of students of different ages to ensure the effectiveness of AI education. Finally, as AI development is an open process, we must embrace open-source principles in AI education to promote equitable and universal access to educational resources, encourage active student engagement, and foster an interactive learning environment.

In summary, AI education is not just about creating new courses or majors; it requires a shift towards intelligent education. It is not just about using AI tools or teaching AI knowledge; it also encompasses improving the scientific literacy of citizens in the intelligent era. This includes fostering AI literacy with critical thinking at the core, creating a digital education system that catalyzes a holistic and deep transformation in the educational paradigm, and cultivating specialized talents in digital and intelligent applications, thereby advancing the sustainable development of the digital economy and society. AI education should comply with the inherent laws of socioeconomic transformation driven by digital productivity as well as the intrinsic laws and missions of education itself. It must adhere to the principles of putting ethics first, strengthening the knowledge base, breaking through traditional confines, combining theory with practice, and encouraging teacher-student interaction. The fundamental mission of AI education is to cultivate new talents for the intelligent era and use ‘artificial’ intervention to ensure that ‘intelligence’ is not only useful, powerful, and efficient, but also benevolent and compassionate, thus achieving the vision of AI with human values for sustainable development.

4. How to govern AI?

The healthy, ethical and sustainable development of AI calls for the establishment of an effective AI governance system, that should, above all, be a development-oriented system. In promoting AI development, it is important to balance safety, regulation and innovation, and avoid constraining AI's advanced productive forces within the confines of current productive relationships. Second, AI governance should be grounded in ethical principles. The UNESCO Recommendation on the Ethics of Artificial Intelligence adopted at the 41st general conference of UNESCO in 2021 defines the values and principles of AI ethics, and ensures AI's fairness, transparency and explainability (UNESCO, 2022). Third, AI governance should adopt a categorized and tiered approach based on the level of risks. Risk management strategies are necessary to delineate the boundaries of AI use, specifying where and for whom and by whom it should not be employed. Fourth, AI governance must be process-oriented (including planning, design, development, production, service, application, management and maintenance) with responsibilities borne by relevant persons, and at the same time, the rights and interests of all people must be protected equally. Fifth, AI governance should be an open cooperative system. In the current geopolitical environment, maintaining an open cooperative governance system is not easy, but critically important. The advancement of science and technology requires a global perspective. It is necessary to ‘prioritize both the secure and controllable advancement of AI and the promotion of open innovation, and safeguard cybersecurity and strengthen international exchanges and cooperation within an open environment’ (Xinhua News Agency, 2019). Sixth, AI governance should be a global initiative that involves all stakeholders. It is imperative to strengthen broad-based dialogue on AI development and governance, with the United Nations setting the agenda and the participation of multiple stakeholders, in order to forge a clear consensus, encourage collaborative innovation and governance, and embed AI with human values for sustainable development.

In summary, AI stands as advanced productivity. We should ensure its safety while promoting the rapid and healthy development of AI. To achieve this, a comprehensive strategy must be deployed across technology, application, education and governance to harness AI's full potential and mitigate its potential risks. By implementing such a strategy, we can ensure that AI serves as a catalyst for the advancement of human society, thus achieving the goal of AI for good of all people and the planet.