Mozilla Data Collective seeks to build AI’s data economy around trust

Mozilla Data Collective, founded by CEO E.M. Lewis-Jong, is addressing critical gaps in AI data collection by prioritizing community ownership, consent, and fair value exchange. Current AI models often rely on indiscriminate web scraping, which perpetuates biases, underrepresents marginalized languages and cultures, and raises legal and ethical concerns. The collective’s approach contrasts with traditional methods by empowering data creators—such as those contributing Hazargi literature from Afghanistan or Mada oral histories from Cameroon—to determine how their data is used, whether through open sharing, attribution requirements, or compensation. The initiative builds on Mozilla’s Common Voice project, which successfully gathered over half a million volunteer contributions across 300+ languages for speech datasets. However, the rise of generative AI exposed tensions: contributors questioned whether their open data contributions benefited opaque AI ecosystems. Mozilla Data Collective now offers customizable licenses, allowing creators to restrict use by geography, purpose, or seek payment, ensuring sovereignty without outright access restrictions. Today, the collective hosts hundreds of curated datasets, including underrepresented linguistic resources like Romansh newspapers from Switzerland. These datasets would otherwise be inaccessible through commercial channels. The model aligns with growing global scrutiny of AI data practices, offering a scalable alternative to extractive models that centralize power and value in tech companies. Lewis-Jong emphasizes the need for ‘clean, abundant, contextualized, consentful datasets’ to build ethical AI systems. The collective’s governance structure reinforces this mission by decentralizing control, ensuring diverse communities—from Afghanistan to Cameroon—shape AI’s future. By addressing bias, consent, and compensation, the project aims to redefine AI’s data economy on principles of trust and equity.