During this year’s Mozilla Festival, we organized a session titled “Open AI systems: will they challenge the concentrations of power?”. Our aim was to explore the emerging new field of open, driven by AI and ML research: the development of open AI systems like Stable Diffusion or BLOOM models and the proliferation of open training datasets like LAION.
Zuzanna, who leads our research work on openness and AI, presented our work on mapping mechanisms for enabling openness of different parts of AI systems. And we were joined for the roundtable by Aditi Surana (University of Edinburgh, DCODE network), Isaac Johnson (Wikimedia Foundation) and our fellow, Nadia Nadesan.
During the session, the speakers discussed how openness will develop and with what impact in the space of AI systems. Will it serve to democratize the tools or make responsible use more difficult?
There was a consensus that traditional understandings of openness will not suffice, unless sufficient attention is also paid to decentralization and democratization of these technological systems. Our speakers pointed to the importance of a design justice approach, participatory governance mechanisms and community-grounded design, and the importance of a public interest perspective.
You can view the Miro board used during the session to present our work on openness and AI.