Many policies, standards and regulations on artificial intelligence (AI) have been presented or are in the pipeline. A common problem with these is that they are often too abstract, too specific or both at the same time to be applied to real projects. The issues and risks related to the various technologies under the AI umbrella are highly context-dependent and it is difficult to find any particular level of abstraction at which to approach them.
In this talk, I describe a context-aware structure that, in a way similar to Value Sensitive Design, bridges the gap between high-level and actionable requirements and explicit design elements, but also provides ways to assess compliance with these requirements. I describe how this approach can take a positive role in the AI developer-buyer-regulator ecosystem and some directions for further research and development. A demo of an implementation of the above, RAIN, developed by the Responsible AI group at Umeå University is also presented.
