In this paper I conceptualize the narrow scope of some AI models from the perspective of trustworthy AI. Trustworthy AI is useful to explore how different stakeholders can rely on AI models with good reason. It is a matter of living up to the commitments and expectations that different stakeholders have toward AI.
I use the example of NAP4DIVE to consider a potential barrier to trustworthy AI. This EU project has an ethical aim: to develop a complex combination of AI and in vitro models of the blood-brain barrier to replace animal models. A relational account of trustworthy AI implies that reasonable expectations of AI models should be demonstrated for different stakeholders (e.g., regulators) in order for them to accept and promote replacement of animal models.
The narrow scope of such models forms a potential barrier to trustworthy AI. The AI model in NAP4DIVE can only be used to predict nanoparticle behaviour across the blood-brain barrier, and is paired with a specific in vitro model. I distinguish four aspects of its narrow scope: specific socio-technical embedding, question-specificity, limited system representation, and narrow validation. In the paper, I relate and prioritize these aspects from the perspective of trustworthy AI.