AbstractCrowdAI (https://www.crowdai.org) is an open source platform for hosting challenges on Open Data for Open Science. It aims to engage a community of machine learning enthusiasts and researchers to help solve problems from big and small research groups in academia, humanitarian organizations, and also industry partners who share the spirit of open data and open research. It aspires to empower many more community members to both understand and use many complex but powerful machine learning paradigms to solve their own problems. It envisions to become an open source repository of machine learning solutions for a diversity of problems in different domains. In this paper, we introduce the core vision of CrowdAI and the problem we aim to solve. We continue by explaining the basic primitives of the CrowdAI ecosystem and the proposed integration with external ecosystems like the federated Knowledge Graph from the Swiss Data Science Center. We briefly discuss the lessons learned from some of the challenges we ran on plant disease classification (with PlantVillage), reinforcement learning for musculoskeletal models (with Stanford and the NIPS conference), and genotype to phenotype prediction (with OpenSNP). We finally conclude by providing a future outlook on the direction where CrowdAI is headed, and how it can play a potentially important role in catalyzing the use of collective intelligence for developing publicly accessible artificial Iintelligence across numerous domains of research.