Professor Paul H Cleverley of Robert Gordon University, Aberdeen, has questioned claims of openness in China's AI development, specifically regarding the GeoGPT geoscience system from Zhejiang Lab. GeoGPT, showcased at last month's World AI Conference, is promoted as open but would not meet the standards of a UN-recommended model openness framework.
According to Professor Cleverley, GeoGPT releases model weights, which are built mainly on Alibaba Qwen, but does not release training data or application source code. Its governance committee also answers directly to Zhejiang Lab. He noted that GeoGPT users route their data through China's legal jurisdiction, which could create dependency and sovereignty risks.
Professor Cleverley also stated that similar concerns arise with the US private-enterprise model, suggesting that no country's AI ecosystem is as open as its diplomacy implies. He advocates for shared openness standards, such as the UN-recommended framework, as a basis for international cooperation.
Separately, Dr Claire Jenkins of Wollaton, Nottinghamshire, proposed that Britain could pursue a distinctive path in AI development. She suggested focusing on AI that is trustworthy, transparent, and directed towards the public good, rather than solely commercial advantage. Dr Jenkins highlighted the NHS and public institutions as strengths that could support this approach, citing recent concerns raised by the Medicines and Healthcare Products Regulatory Agency's national commission into AI regulation in healthcare.