The centrality of good-quality data to train present-day AI systems is unprecedented. But the decades-old legal frameworks of IP protection and copyrights present a major bottleneck to equitable progress. Whether or not we completely do away with copyright laws, it's clear that major reconsiderations are required.
This session will explore the following themes and events:
Copyright and "Fair Use" laws across the globe
A survey of AI companies based in countries with differing levels of leniency regarding free use of data for AI training
Chinese companies Deepseek and Qwen releasing major models to the open with far lesser resources
Other companies releasing leading opensource models obscuring their data sources
How big entities get away with using large pools of "copyrighted" data — showing it's the smaller players that actually get impacted by regulations
The recent example of Facebook doing the same, now coming to public awareness
The Ghibli-style art generation backlash, sparking debate over artistic rights and years of effort being instantly style-transferred
What it means for open source and models for the commons
The talk aims to bring to public discourse the state of IP regulations, what "fair use" actually means, and how regulations don't really deter the bigger players anyway - in the context of training AI models. It attempts to shed light on the urgency of this matter — especially from the perspective of democratizing access to, and building AI technologies.
how IP protection laws stifle AI advancements, and its drastically higher impact on lower-resourced players
the violations bigger entities make regardless of legal regulations
the debatable nature of IP rights
the urgency of a reformation in copyright laws and the definitions of "fair use"