Challenges in Openness Standards for AI Models

Ambassador Zheng Zeguang Celebrates Open AI Models

Ambassador Zheng Zeguang rightly celebrates openly released AI models, highlighting the progress made by Chinese labs like Qwen, DeepSeek, and Kimi in this space. These systems have led to significant competition that benefits everyone, especially where models can run on modest hardware in developing countries (The future of AI hinges on openness and cooperation. China and Britain can gain much by working together, 30 July). However, his claim that openness is a defining feature of China’s AI development deserves closer scrutiny.

GeoGPT: A Model of Open Science with Caveats

Take GeoGPT from Zhejiang Lab, showcased at last month's World AI Conference as an example of jointly governed open science. Despite being promoted to countries as open, it does not fully adhere to international standards. GeoGPT releases only model weights built mainly on Alibaba Qwen, whose licenses are not Open Systems Interconnection (OSI)-compliant. Moreover, no training data or application source code is released. Its governance committee answers directly to Zhejiang Lab itself.

Comparisons with Global Standards

In contrast, domain rivals offer more comprehensive openness. The European Space Agency's Earth Virtual Expert (EVE), designed for earth sciences, releases its training and evaluation tooling. Similarly, the US non-profit Ai2’s OLMo publishes everything needed for reproducibility. These differences are crucial as simply having a free hosted service does not equate to true open access.

Ambassador Zheng's claim extends beyond China, highlighting that no country’s AI ecosystem is entirely open. The U.S. private-enterprise model shares similar concerns. For the UK to gain from cooperation with these countries, it should advocate for shared openness standards like those recommended by the UN. This approach ensures that any joint work adheres to international best practices.

Implications for British AI Development

Dr Claire Jenkins argues that the UK cannot afford to be a passive consumer of AI technologies shaped by geopolitical interests (The future of AI hinges on openness and cooperation. China and Britain can gain much by working together, 30 July). Instead, Britain has unique strengths in its National Health Service (NHS), public institutions, and ethical tradition.

These provide an opportunity to develop AI that is both innovative and ethically sound, focusing on the public good rather than commercial gain. For instance, AI can reduce administrative burdens, improve diagnosis, and widen access to care. However, public confidence depends on robust governance, meaningful human oversight, and ensuring patient data serves the needs of citizens, not corporations or geopolitics.

The Medicines and Healthcare Products Regulatory Agency's (MHRA) national commission has highlighted these issues. The UK should avoid imitating either Silicon Valley’s dominance or China's state-led approach. Instead, it could offer a unique path: AI that combines innovation with democratic accountability, scientific excellence, and social justice.

This form of leadership would align with Britain’s traditions and set an example for others to follow (The future of AI hinges on openness and cooperation. China and Britain can gain much by working together, 30 July).

Conclusion

In conclusion, while China and the US might claim their ecosystems are open, none fully meets international standards. The UK should press for shared openness standards like those recommended by the UN to ensure that any collaboration is based on real partnership and not just the language of openness. By focusing on ethical AI development, the UK can lead in a way that benefits its citizens while setting a global example for transparent and responsible AI practices.

Source: The Guardian