Does One Framework Fit All? Testing Governing Knowledge Commons in AI-Powered Drug Discovery

Abstract The implementation of advanced artificial intelligence (AI) models throughout the drug discovery and development process is made possible thanks to the massive amounts of data on which these models are trained. However, this data is difficult to interpret, compile and access and thus are typically not available to other companies and/or individuals. Moreover, despite the possibility under patent law, to provide protection to AI technologies and drugs created with the assistance of AI (AI-assisted drugs), the data generated within the AI-powered drug discovery process may be shared and accessed by the broader scientific community, since there is the need to include small-scale innovators and public institutions in a scenario where their contribution has been increasingly marginalized in favour of AI training undertaken by big companies. These technical and legal issues result in a societal dilemma that this paper defines as a ‘vicious cycle’ comprised of uncertain proprietary regimes, scarce sharing of data, ineffective AI training, and lack of safe and effective drug candidates. To break this cycle, this paper suggests adopting the Governing Knowledge Commons (GKC) Framework, which is structured according to a series of general questions (‘variables’) that not only find application in case the knowledge resource is completely ‘commonized’, but also where there are co-existing spaces of property and non-property in its governance. Such framework should be tested, focusing on AI platforms in drug discovery, by assessing to what extent AI platforms are being managed, as knowledge resources, and being able to manage further knowledge resources (training data and generated outputs) in a distributed and communal way. Ultimately, this assessment of governance of AI platforms through GKC testing, which in this paper will focus on the well-known AlphaFold AI platform, can contribute identifying promising cases of knowledge commons within drug discovery.