How experts in China and the UK view AI risks and collaboration

Recent advances in artificial intelligence (AI) offer many important benefits for society, the economy, and scientific progress. An important factor in these advances is the development of open source AI: AI technologies whose source code and data are freely available for others to use, study, modify, and distribute. Open source AI is essential to accelerate innovation through collaborative development, reduce redundancy, and democratize access to AI capabilities. Being accessible to all, it facilitates economic development and social progress, for example by allowing anyone to adapt and refine high-performance AI models for specific tasks. In addition, global collaboration between researchers, developers, and users on open source AI enables collective progress on shared AI projects and fosters the development of guidelines and best practices for transparency, accountability, and ethics. Finally, by promoting transparency and accountability through accessible code and data, it helps identify and address biases, errors and ethical concerns in AI and enables users to understand how the technology works.

Both China and the UK are active members of the open source community and pioneers in the field of AI. This history has led them to emerge as leaders in open source AI. For example, UK-based Stability AI is the developer of many popular open source generative AI tools used to create images, audio, 3D models, and code. China has produced some of the world’s most successful open source extended language models (LLMs), including Qwen (Alibaba) and Yi (01.AI). These open source AI projects compete with proprietary (or closed) AI where developers limit public access to the underlying technology.

However, open-source AI presents unique challenges. First, unlike proprietary AI, where developers can monitor what users do with their technology, once developers make open-source AI publicly available, they have little to no control over how others can use their technology. Malicious actors can tamper with open-source AI to remove protections, manipulate results, and generate inaccurate information. Additionally, malicious actors can use the technology for dangerous and illicit purposes, such as conducting cyberattacks, spreading disinformation, committing fraud, creating contraband, and engaging in other illegal activities. Second, unlike proprietary AI, where the developer is responsible for the technology, there isn’t always someone responsible for open-source AI projects. As a result, the technology may have known bugs or security vulnerabilities that no one fixes. Similarly, open-source AI may be provided without any warranties. For example, users may not know whether developers trained an open source AI model on poor quality or illicit data. Finally, open source product development practices can create unknown risks, such as if attackers surreptitiously attempt to introduce malicious code or data into an open source project.

Managing AI risks is a matter of global concern, and the UK and China have remained at the forefront of this effort, even as they both seek to support the development and use of AI by their respective companies. The UK hosted an AI Safety Summit in 2023, which was attended by many countries, including China. The summit concluded with the Bletchley Declaration, in which participating countries resolved to maintain an inclusive global dialogue that engages existing international forums and other relevant initiatives and contributes openly to broader international discussions. President Xi Jinping subsequently reiterated this call for mutually beneficial cooperation on common interests, including AI, in remarks in San Francisco during a bilateral meeting with the US President. The Chinese Ministry of Foreign Affairs also issued a statement in October 2023 calling for global collaboration to promote the sound development of AI, share AI knowledge, and make AI technologies available to the public under open-source conditions.

Despite these high-level government statements, it is unclear whether the UK and China can deliver on their aspirations for closer cooperation on AI. To understand the feasibility of such partnerships, it is important to better understand whether the concerns and priorities of AI experts outside government align, and what their experiences of collaboration have been to date. This report seeks to provide insights into these issues.

Read the report.