Licensing in Artificial Intelligence Competitions and Consortium Project Collaborations

Platforms are emerging that allow data scientists, software and hardware engineers to collaborate through organisational boundaries to develop systems of Artificial Intelligence (AI). Such collaboration involves the exchange of assets representing Intellectual Property (IP) of the collaborators. The tension between permitting access and protecting IP is thus one of the critical challenges faced by organisations willing to innovate through collaboration. Licensing is a common way to address the issue, but the influence of the licensing rules on the intended form of collaboration is still unclear. In this paper, we identify and analyse the rules that are used to regulate IP exchanges in two common forms of collaboration: a) competitions where one customer benchmarks and selects among multiple suppliers and b) consortium projects where multiple parties collaborate to product joint results. Due to our interest in AI, we have chosen to analyse the terms and conditions of competitions hosted on Kaggle™ a leading online platform for Competitions. For consortium projects, we have analysed the DESCA Consortium Agreement template. DESCA is often used for European projects, an increasing number of which are used to fund AI innovation projects. We have applied In Vivo Coding and Concept Coding coding techniques to highlight rules applicable to IP exchange. We structured the findings in the form of tree graphs consisting of interdependent textual phrases to extract, group and compare the terms and conditions of IP sharing in each collaboration form and how they relate to the characteristics of the studied collaborations. The results indicate that each form of collaboration has its own set of rules that address comparable concerns but have different content. Practitioners, both platform providers and collaborators, can utilise our results to implement licensing for IP exchange that fits the desired type of collaboration. For researchers, our results represent a step towards the automation of license generation and enforcement.