Mediation Challenges and Socio-Technical Gaps for Explainable Deep Learning Applications

Deep learning experts often lack awareness of social implications: The study found that DL experts "did not spontaneously engage into considerations about the social meaning of machine learning models that they build," suggesting a significant gap between technical development and real-world impact on users and stakeholders.

There's an expectation that "mediators" will solve the explanation problem: Rather than building explainability into their work, participating experts expected that intermediaries would eventually bridge the gap between technical and social meanings—a potentially unrealistic assumption that could delay necessary changes in how AI systems are developed.

Current research incentives misalign with explainability requirements: The paper concludes that "current research incentives and values guiding the participants' scientific interests and conduct are at odds with those required to face some of the scientific challenges involved in advancing XAI," indicating systemic barriers to responsible AI development.

GDPR's right to explanation reveals deeper social challenges: The regulation's requirement for explanations has exposed that explainability is not just a technical problem but a social one requiring coordination between AI developers, organizations, and users—areas where current practices fall short.

Responsibility gaps exist between technical and deployment stages: The study highlights a "socio-technical gap" where technical experts assume responsibility ends at model development, leaving critical questions about user understanding and fairness unaddressed until deployment.