'Mind the Gap! Pathways Towards Unifying AI Safety and Ethics Research' arXiv 2512.10058 — diagnostic paper on the parallel safety-vs-ethics research-track divergence and reunification proposals
The Mind the Gap arXiv paper (2512.10058) diagnoses the structural divergence between AI safety research and AI ethics research tracks — both addressing alignment-related concerns but operating in parallel with limited cross-citation and disagreement on basic definitions of 'alignment'. The paper proposes specific pathways toward methodological and institutional unification.
The substantive piece is the methodological-divergence diagnosis. Pre-Mind-the-Gap the AI safety and AI ethics research communities operated largely in parallel — safety focused on technical methodology (RLHF, constitutional AI, interpretability, formal verification), ethics focused on sociotechnical analysis (fairness, accountability, transparency, justice). Both communities addressed alignment-related concerns but reached different conclusions about what alignment means and how it should be evaluated.
The competitive read for the H2 2026 to 2027 alignment-research direction is that the field needs institutional infrastructure for safety-and-ethics reunification. The sociotechnical critique of RLHF and the shared-failures analysis both reflect the parallel-tracks divergence — technical work that doesn't fully address sociotechnical concerns produces results that ethics researchers don't accept as alignment achievements. The Mind the Gap pathways proposal addresses this directly.
arXiv — Mind the Gap! Pathways Towards Unifying AI Safety and Ethics Research (2512.10058) → · arXiv — AI Alignment Strategies from a Risk Perspective →