Europe and Open Source AI: Beyond Licenses

Report by Mitchell Baker: “..An asset is “open source” or not based on its license terms. An open source asset can have a wide variety of impacts: public benefit or private enrichment; sustainable or a “flash in the pan;” research achievements or products that people use, to name a few. An open source asset’s impact depends on other layers of activity. There is no canonical artifact that sets out these layers of activity; the framework below is distilled from three decades of practice.
The first layer is the license, which determines whether or not something is actually “open source.” The definition of “open source AI” is under discussion, so the License Terms section below recommends a way to move forward despite the ambiguity. A second layer is the technology itself – is there a clear enough vision of what a project aims to build and the success criteria? A third layer is the team, which includes the people who build the project, the reasons they choose to participate, and what they need to give their best. A fourth layer is the collaborative process by which that technology is developed – is it a process in which multiple stakeholders in the EU share a vision, a community sense of ownership and a commitment to the whole? A fifth layer is the finished product – can the research and development advancements be turned into products that are used regularly and productively? A sixth layer is the organization that holds responsibility for the technology and/or products.
Two things about frontier AI are unlike the open source software experience this paper draws on: the capital required to build the largest models, and the structural challenges of the underlying hardware, chips, and compute. How frontier-scale training is financed in a public-benefit model is a genuinely open question, and this paper does not resolve it. The focus of this paper is different: whatever methods and amounts of funding are allocated, the likelihood of bringing the open source AI projects to successful impact depends on the variables discussed here. This paper will limit itself to the objectives for open source software and AI…(More)”.