The In-Between - A Theory of Relational Governance for Autonomous and Superintelligent Systems

Current governance frameworks for AI systems rest on an assumption that acceptable behavior can be specified in advance and enforced through external constraints. This paper argues that this assumption fails categorically once systems operate in open environments, exercise delegated judgment, and exceed human cognitive oversight. Guardrails—rule-based, boundary-enforcing mechanisms—produce a context gap (finite policies cannot cover open-ended situations), a responsibility gap (dispersed delegation dissolves accountability; cf. Matthias, 2004), and a displacement mechanism (compliance artifacts substitute for normative judgment, producing what Meyer & Rowan, 1977, termed decoupling between formal structure and actual practice). The paper develops an alternative: governance through the In-Between—the relational space where alignment is continuously tested rather than statically prescribed. Rooted in enactivist philosophy (Varela, Thompson & Rosch, 1991) and responsive regulation theory (Braithwaite, 2006), the In-Between is a structural invariant that transforms across scales: an emergent relational field in dyadic interaction, a minimal institutional governance space in organizational contexts, and a calibration interface that a future superintelligent system might maintain voluntarily for epistemic self-correction. Three governance modes address three distinct regimes. Control (guardrails) suffices in bounded environments where behavior is specifiable. Orientation (Führung)governs agentic systems in open environments where a human can still lead—through ownership of delegated action, interruptibility, and the capacity for costly revision. Calibration becomes necessary beyond a categorical ceiling: the threshold at which human cognition can no longer recognize deviation as deviation, rendering orientation impossible. The paper introduces operational tools: three pressure points with falsifiers for diagnosing governance theater, the D/C/K framework (Discrepancy throughput, Contestation capacity, Commitment revisability) for measuring whether governance functions or merely exists, and a conditional argument—drawing on Russell’s (2019) beneficial AI framework—for why an autarkic superintelligence might rationally maintain external friction as an epistemic safety mechanism. The theory is presented as a falsifiable architecture with explicitly marked boundary conditions. Its strongest claim—that the safest future superintelligence may be onethat chooses to maintain the In-Between—is conditional on assumptions about objective uncertainty and temporal goal integrity. These assumptions are stated, nothidden. The paper concludes with an honest assessment of open gaps: empirical validation, the accountability problem, the transition zone between orientation andcalibration, and manipulation resistance under asymmetric intelligence.