A coalition of civil society organisations and academic researchers, including the Center for Democracy and Technology (CDT), Mozilla, and acclaimed academics from esteemed universities like Stanford and Harvard, have written a letter to US Secretary of Commerce Gina Raimondo. The letter, dated 25 March 2024, was submitted in response to a public consultation conducted by the National Telecommunications and Information Administration (NTIA) operating under the Department of Commerce. The consultation explores the risks and benefits of open-source AI models.

The coalition's letter emphasises three crucial points regarding openness and transparency in AI models. Firstly, it underscores the significant societal advantages of open-source AI models, including their ability to drive innovation, enhance understanding of AI safety, and facilitate compliance with privacy and civil rights regulations. Next, it argues for evidence-based policies that consider the marginal risks posed by open models compared to closed ones and existing technologies; the coalition believes that a thorough evaluation is needed before imposing broad restrictions. Lastly, the letter advocates for addressing well-defined risks through tailored solutions rather than imposing overarching limitations to avoid hindering economic growth and scientific dialogue.

The coalition calls for several actions to support these arguments. They recommend increased research and development support for open-source AI and highlight the importance of collaboration with the open-source community to establish improved testing standards. Additionally, they propose establishing a robust interagency process involving key agencies responsible for competition, civil rights, and scientific research. This collaborative effort aims to ensure comprehensive AI reports and effective utilisation of export controls.

Why does it matter?

The tech industry's debate over open-source AI continues. Entrepreneur Elon Musk, shortly before the launch of an AI open-source chatbot model called Grok1, sued OpenAI, which he accused of violating the agreement that GPT-4 would remain open-source, further highlighting the complexities and implications of this issue.

Knowledge slaveries: Is Bottom-up AI a possible solution?

The constant use of AI services generates vast amounts of data reflecting human thoughts and emotions, leading to concerns about privacy and control over personal knowledge, potentially resulting in a state of 'knowledge slavery'. To counter this, retaining ownership over thinking patterns, including those derived from AI, is crucial. Bottom-up AI, advocated by Diplo's Executive Director, Dr Jovan Kurbalija, offers a possible solution, empowering individuals and communities while ensuring privacy and data protection. Leaked documents from Google suggest that open-source AI could rival proprietary models, offering similar quality at lower costs and with greater efficiency. However, challenges remain in ensuring data quality, particularly in data labelling processes. While the full adoption of bottom-up AI is uncertain, exploring alternatives to the prevailing AI paradigm is essential for informed decision-making and to mitigate potential disruptions from reliance on larger AI service providers.