Like everyone else in Machine Learning, we’ve been following the EU AI Act closely at Hugging Face.
It’s a ground-breaking piece of laws that’s poised to shape how democratic inputs interact with AI technology development world wide.
It’s also the final result of in depth work and negotiations between organizations representing many alternative components of society –
a process we’re particularly sensitive to as a community-driven company.
In the current position paper written in coalition with Creative Commons,
Eleuther AI, GitHub, LAION, and Open Future,
we aim to contribute to this process by sharing our experience of the vital role of open ML development
in supporting the goals of the Act – and conversely, by outlining specific ways through which the regulation
can higher account for the needs of open, modular, and collaborative ML development.
Hugging Face is where it’s today because of its community of developers, so we’ve seen firsthand what open development brings to the table
to support more robust innovation for more diverse and context-specific use cases;
where developers can easily share modern recent techniques, mix and match ML components to suit their very own needs,
and reliably work with full visibility into their entire stack.
We’re also conscious about the vital role of transparency in supporting more accountability and inclusivity of the technology –
which we’ve worked on fostering through higher documentation and accessibility of ML artifacts, education efforts,
and hosting large-scale multidisciplinary collaborations, amongst others.
Thus, because the EU AI Act moves toward its final phase, we imagine accounting for the particular needs and strengths of open and open-source development of ML systems can be instrumental in supporting its long-term goals.
Together with our co-signed partner organizations, we make the next five recommendations to that end:
- Define AI components clearly,
- Make clear that collaborative development of open source AI components and making them available in public repositories doesn’t subject developers to the necessities within the AI Act, constructing on and improving the Parliament text’s Recitals 12a-c and Article 2(5e),
- Support the AI Office’s coordination and inclusive governance with the open source ecosystem, constructing on the Parliament’s text,
- Make sure the R&D exception is practical and effective, by permitting limited testing in real-world conditions, combining features of the Council’s approach and an amended version of the Parliament’s Article 2(5d),
- Set proportional requirements for “foundation models,” recognizing and distinctly treating different uses and development modalities, including open source approaches, tailoring the Parliament’s Article 28b.
You will discover more detail and context for those within the full paper here!
