How to train the upcoming generation of AI researchers? Design principles and results from the Ethos+Tekhne initiative as a case study

<ns3:p> A lasting challenge for the next generation of AI researchers and professionals will be to shape a global AI infrastructure capable of accommodating stakeholder heterogeneity, while addressing technical constraints. It becomes crucial to provide early-career researchers and professionals in AI with the tools and the knowledge to understand the complexity of this challenge, and with the technical skills to partake in it. Here, we present core design features and results from a novel training initiative combining AI and ethics. This initiative is called “Ethos+Tekhne” because it centers on the discussion of two main questions: what kind of AI we want ( <ns3:italic>ethos</ns3:italic> ), and how to achieve it concretely ( <ns3:italic>tekhne</ns3:italic> ). Its programme was designed around three key principles: pluralism, narrative, and balance. In terms of pluralism, the training combined several types of theoretical and practical learning activities, as well as speakers from a variety of fields, including non-academic sectors and citizens. With regard to narrative, the training followed a programme-wide thread of thematic days. Finally, with respect to balance, on-site participant group size was limited to 30–40 individuals, in order to offer each participant the space and time to actively engage in discussions, while supporting online attendance to promote open access. Moreover, the programme schedule aimed to balance activity and break time to encourage participants to exchange their thoughts in informal settings. The hallmark of success of this initiative has been its regenerative nature: some first-edition participants led the organisation of the second edition, and some second-edition participants are currently organising the third edition. Content-wise, three axes of action in ethical AI development emerged from participants’ discussion: <ns3:italic>AI for Good</ns3:italic> , <ns3:italic>Good for AI</ns3:italic> , and <ns3:italic>Good AI.</ns3:italic> We report our rationale, programme design, and results, in the hope that these could provide material for future training programmes in AI, and potentially other technologies. </ns3:p>