Knowledge-based Adequacy assessment Approach to support AI adoption
Adopting AI as a technology that solves a particular problem (i.e., meets an architectural driver) is a significant architectural decision. Existing techniques for adequacy assessment of architectural decisions often fail to predict effects of adopting complex technologies such is AI. In this paper, we argue that the reason for this is that they fail to capture the level of knowledge that architects have about a complex technology that they aim to adopt, making the discussion about it difficult and adequacy check prone to mistakes. To solve these issues, we introduce an approach that instructs architects to decompose complex drivers for adopting new technologies according to properties of the technology, and to explicitly assess knowledge that architects have about each of those properties. In order to do so, we present a template that explicitly captures the level of knowledge that architects have about important AI properties, which serve as new requirements exposing the influence of adopting AI on software system. Through evaluation, we have demonstrated that our approach successfully complements existing adequacy assessment techniques and is able to expose influences of adopting new complex technologies on underlying software architecture.