Technical Readout - Columbia Convening on Openness and AI
Key technical takeaways from a convening focused on exploring what "openness" means in the AI era.
Overview
On February 29, 2024, Mozilla and the Columbia Institute of Global Politics co-hosted a gathering of over 40 experts and stakeholders in AI to explore the concept of "openness" in the AI era. This diverse group included representatives from leading AI startups, companies, non-profit AI labs, and civil society organizations.The convening aimed to help develop a better framework for what “open” means in the AI era, drawing inspiration from the pivotal role that open source software has played in technology, cybersecurity, and economic growth over the years.
This work is particularly timely: as AI development shifts from research labs into customer-facing products, there has been increased usage of proprietary AI models — raising concerns about negative impacts for innovation, competition, and accountability.This brief attempts to distill the technical dimensions of openness in AI that participants focused on during the convening. It emphasizes how openness in AI — defined by some participants as the broad public availability of key artifacts and documentation from AI systems — can help AI developers and deployers.
This brief summarizes different conceptual approaches for openness, explores a framework for approaching openness across the AI stack, and provides an appendix that goes deeper into details about openness in different components of the AI stack. Finally, this brief also provides a summary of areas suggested for future exploration.