To regulate the vision of AI, at present, more than 100 different AI ethics guidelines have been published around the world, and there are other guidelines that may be oriented to the principles of specific fields such as autonomous driving or data ethics. These efforts collectively provide a complementary understanding from the perspective of related fields. In terms of the overall vision of AI development, people seek to build human-centered AI, or trustworthy AI, friendly AI, AI for SDGs, beneficial AI, etc. Many people choose the term responsible AI, beneficial AI, and these different perspectives are not competing with each other, but complement and improve each other.

In 2019, China published “the Governance Principles for the New Generation Artificial Intelligence--Developing Responsible Artificial Intelligence”, which emphasize harmony and human-friendly, fairness and justice, inclusion and sharing, respect for privacy, safety and controllability, as well as shared responsibility, open and collaboration, and agile governance. In 2021, China published the "Ethical Norms of New Generation of Artificial Intelligence" reflects how to implement the above principles into the entire life cycle of AI, including the management, research and development, supply and use phases. There are different focuses on implementing these ethical principles at each phase, and when doing research and developing AI, we should be genuinely committed to enhancing safety and security and avoid bias and discrimination. In terms of management, we should strengthen risk prevention; In terms of use, we emphasize the avoidance of misuse and abuse, and prohibit malicious use. That's why we're discussing this issue throughout the life cycle, not just making a checklist – we must embed these ethical principles throughout the lifespan of research, applications, and uses of AI.

Therefore, the overall vision of China's “Governance Principles for the New Generation Artificial Intelligence”is that we believe AI should serve as enabling technology to contribute to the achievement of the United Nations Sustainable Development Goals (UN SDGs). Based on existing English publications, our research shows that, in Computer Science and AI communities have begun to contribute to different aspects of UN SDGs. The vast majority of efforts are currently focused on AI for health and AI for education related research. Only a very small proportion is related to SDG 16 (peace, justice and strong institutions), which is terrible. However, the good aspect is that the proportion of research related to this goal is growing. On the other hand, although many SDG pillars may not help us with obvious economic benefits directly, they are still indispensable and serve as must to the world. For example, the protection of life on land and below water. That's why I think that from the perspective of government and social infrastructure, we have to step up our efforts to achieve these goals.

I think the vision of today's session has many things to do with our next generation. We must influence the world in which our next generation lives in a careful way – and we hope to create a peaceful future for our next generation. So that's why UNESCO Recommendation on the Ethics of Artificial Intelligence is such an important world effort, and we need to ensure that it is used to protect our next generation and build a trusted, inclusive and better world for all through strategic policy and technological research on AI.

An example I would like to give here is that sometimes while you may want to use AI to facilitate the achievement of SDGs, you may be taking the wrong approach. Using emotion recognition in class is a very problematic design, although this is related to AI for Education. While using AI to protect the vulnerable groups from online and school bullying, I think it's a good idea. However, whether we can effectively detect school bullying through the automatic detection and recognition technology of AI, and at the same time protect the privacy of students and teachers is a topic worthy of in-depth study and practice.

Responsible AI governance practices should have governments, media, civil society, users, and service providers, etc. involved. Stakeholders must be very active in interacting with the entire lifecycle of AI and embed ethics within it. My second example is that although you want to use AI in a good way, let's say you want to serve a lot of people well, so you build large machine learning models. On the other hand, they are very energy-intensive, inefficient and costly in this regard. Are they really contributing in a good way? To this end, we have created carbon efficiency observatories for AI to analyze the carbon energy efficiency of representative large-scale AI machine learning models (AI Carbon Efficiency Observatory). This effort is to inform our AI community to minimize their negative impacts on climate actions while AI provides the necessary services.

For SDG 16 (Peace, Justice and Strong Institutions), it is important to note that ethical discussions and designs for autonomous weapons are essential. We should conduct a discussion on AI and peace under the framework and guidance of Goal 16, and I believe that AI should be used to reduce misunderstandings and increase mutual trust. I think AI enabled Cultural Interactions are essential to build the bridge for Live in Peacefully and Harmony. I think that's the way we should use to remove many of the barriers that prevent us from understanding each other, by avoiding misunderstandings and learning from each other.

Lethal Autonomous Weapons Systems (LAWS) are highly relevant to autonomous decision-making. I think the problem is not just limited to lethal autonomous weapon systems themselves, but should address in greater details on how people use AI-based autonomous decision-making techniques to assist and semi-replace their decisions of their own. So we should be discussing the ethics of AI autonomous decision-making, not just focusing on lethal autonomous weapon systems, so that we can see the overall risks posed by such applications in broader views.

All the military AI right now may not be safe and reliable enough, while AI has been trying to be used in almost all aspects of the military field. For example, the adversarial attack technique on neural network models that is now commonly used. These adversarial approaches are likely to be attacked in most military AI applications, which means that no military AI application is sufficiently secure, hence not trustworthy, and not responsible. This is also an issue that we must address together to advance SDG 16.

I think maybe I don't have time to start a discussion about how we should apply AI to facilitate cross-cultural interactions today, but I'm still willing to mention that we've built a cultural interaction engine based on AI to link cultural heritages around the world (Cultural Interactions Engine). UNESCO's World Heritage List can help us see the commonalities and differences of different societies and cultures, so that we can truly connect with each other through cultural commonalities, and be curious about each other because of cultural differences, so as to understand and learn from each other. Finally, not only should we humans live in harmony, we should also take into account biodiversity and share the future with other creatures on Earth with the help of artificial intelligence.

Introduction