The AI boom has been fueled by an extractive approach to data. Models are trained on data scraped from the open internet, with no regard for consent, provenance, or fair compensation. This talk unpacks why that status quo - a "poisoned well" of low-quality, unrepresentative data that excludes billions of people, coupled with practices that flout the rights of the individuals and communities whose data is being used - is unsustainable all around. On one hand, data producers who might otherwise want to share their data hold back for fear of free-for-all exploitation, while data consumers are left frustrated by "trash in, trash out" datasets that fail to serve the diverse needs of people around the world.  

In rejection of this status quo, Mozilla Data Collective is building a different kind of data-for-AI ecosystem: data producers retain ownership of their datasets, set their own terms of use (free, conditioned, or compensated), and decide who gets to benefit from their data; while data consumers gain streamlined, transparent access to high-quality, contextualised datasets, with provenance backed by the communities who are sharing the datasets. The platform is designed to serve a wide range of partners: nonprofits, enterprises of all sizes, universities and researchers, and others who wish to share and download data on fair terms. 

The talk will focus on tangible illustrations of what it's like to practice "defiant optimism", a conviction that AI's promise doesn't have to come at the cost of human agency, and that a multilingual, multicultural, multimodal tech future is achievable if people and communities are given real control over how their data is created, shared, and governed.

Veuillez noter que ce webinaire se déroulera en anglais.