LAION wants researchers and public institutions to have the computing resources needed to develop advanced AI models openly. Its petition proposes an internationally funded research institution with a supercomputer, arguing that access to powerful models should not rest with a small number of large companies.
A public resource for open AI
The initiative calls for 100,000 state-of-the-art AI accelerators to support training fundamental open-source AI models. LAION presents this infrastructure as a way to strengthen technological independence, encourage innovation and protect democratic principles.
Members Christoph Schuhmann, Huu Nguyen, Robert Kaczmarczyk, and Jenia Jitsev argue that systems such as GPT-4 are too consequential to be controlled exclusively by a few companies. Their petition warns that educational institutions, government agencies and countries could become reliant on corporations that operate with limited transparency and public accountability.
The proposed institution is modeled on the CERN project. Its funding would come from the international community, particularly the EU, the UK, Canada and Australia. The petition says experts in machine learning and supercomputing would oversee computing resources, alongside oversight by democratically elected institutions in participating countries.
Why LAION opposes an AI pause
LAION supports rapid progress in AI research and opposes the “AI pause” proposed by some researchers and business leaders. Its concern is that a pause could constrain public research while corporate or state actors continue making advances out of public view.
In that scenario, researchers and society could have fewer opportunities to examine the safety of advanced AI systems. LAION’s argument is that open research can make scrutiny more possible by enabling a wider community to study how these systems are built and what they can do.
The petition connects that case to the broad reach of foundational AI models. It points to applications across scientific research, education, government and small and medium-sized businesses, and says access to these systems should be as open as possible.
Access to models, not just interfaces
The proposed platform would let researchers and institutions around the world train and refine advanced models such as GPT-4, investigate how they work and explore uses for the public good. LAION says services such as OpenAI provide programming interfaces rather than the models themselves, a limitation that can restrict research opportunities.
LAION also advocates open-source models that work with multiple kinds of data: audio, video, text and program code. The organization says this approach could expand academic research, improve transparency and support data security.
Openness is only one part of the proposal. The institution would have security measures comparable to those of a biological research laboratory, with multiple security levels and internationally recognized experts on staff. It would also be expected to communicate its research results transparently to both the scientific community and society.
Petition and LAION’s background
At the time described in the article, the petition had more than 2,000 signatures, around 22 percent of its 10,000-signature goal. Most signatories were from the US and Germany, with about 1,000 from the EU. The collection was set to continue for more than two months.
LAION is a German non-profit association whose work promotes publishing AI models, datasets and code. It is known for the LAION-5B dataset, which contains links to images used to train models including Stable Diffusion and Imagen. The dataset has faced criticism because some links point to copyrighted or private material not intended for AI training.
Founded in Germany in the summer of 2021, LAION has founding members from around the world who work on its projects remotely. Its petition extends the organization’s focus on open AI into a proposal for shared computing infrastructure, international governance and public communication of research.