Value Sensitive Design (VSD) is an established method for integrating values in technical design. It has been applied to different technologies and recently also to artificial intelligence (AI). We argue that AI poses a number of specific challenges to VSD that require a somewhat adapted VSD approach. In particular, machine learning (ML) poses two challenges to VSD. First, it may opaque (to humans) how an AI system has learned certain things, which requires attention for such values as transparency, explainability, and accountability. Second, ML may lead to AI systems adapting themselves in such ways that they ‘disembody’ the values that have been embodied in them.
In order to address these, we propose a threefold adapted VSD approach. 1) integrating the AI4SG principles in VSD as design norms from which more specific design requirements can be derived. 2) distinguishing between values to be promoted by the design and values to be respected by the design in order to ensure that the resulting design does not only do no harm but also contributes to doing good.`And 3) extending the VSD process to encompass the whole life cycle of an AI technology in order to be able to monitor unintended value consequences and to redesign the technology if necessary. We illustrate the new VSD for AI approach with an example use case of a particular SARS-CoV-2 contact-tracing app.