Bridging the Communication Gap: Evaluating {AI} Labeling Practices for Trustworthy {AI} Development

Artificial intelligence (AI) is becoming integral to economy and society. However, communication gaps between developers, users, and stakeholders hinder trust and informed decision-making. To make the behavior of AI models more transparent, high-level AI labels have been proposed, drawing inspiration from systems like energy labeling. While AI labels can already inform on performance trade-offs, for example with regard to predictive model performance and resource efficiency, the practical benefits and limitations of this communication form remain underexplored. Our study evaluates AI labeling through qualitative interviews along key research questions. Based on thematic analysis and inductive coding, we firstly identify a broad range of practitioners with diverse use cases and requirements to be interested in AI labeling. Benefits are primarily seen for bridging communication gaps and aiding non-expert decision-makers. However, our interviewees also mentioned limitations and suggestions for improvement. In comparison to other reporting formats, the reduced complexity of labels was acknowledged to benefit fast knowledge acquisition without deep technical AI expertise. Trustworthiness was found to be strongly influenced by usability and credibility, with mixed preferences for self-certification versus third-party certification. Our insights specifically highlight that AI labels pose a trade-off between simplicity and complexity, address diverse user needs, and nudge interviewee priorities toward sustainability. As such, our study validates AI labels as a valuable tool for enhancing trust and communication in AI, offering actionable guidelines for their refinement and standardization.