Title: What Must Be Preserved? Mapping Theories of Consciousness to Engineering Requirements for Mind Preservation

Abstract

The engineering of consciousness preservation (whether through whole-brain emulation, cryonic revival, or brain-computer interfaces) proceeds largely without reference to theories of consciousness. This is a remarkable omission. The choice of what must be preserved in a brain depends entirely on what generates consciousness, and the major theories disagree on this question by many orders of magnitude in their engineering implications. I evaluate eight major theories of consciousness (Integrated Information Theory (IIT 4.0), Global Neuronal Workspace Theory (GNWT), Higher-Order Thought theory (HOT), Predictive Processing / Free Energy Principle, Recurrent Processing Theory (RPT), Biological Computationalism, Orchestrated Objective Reduction (Orch OR), and Attention Schema Theory (AST)) against nine preservation-relevant criteria, and derive specific engineering requirements from each theory's core postulates. The theories split 4-3-1 on substrate independence. Required information content ranges from approximately 1-10 TB under AST to physically impossible under Orch OR. Required compute spans from 10^15 FLOPS to formally uncomputable. Despite these disagreements, all eight theories converge on three requirements: temporal dynamics must be preserved, integration across components is necessary, and feedforward-only architectures are ruled out. Cross-theory risk analysis reveals that biological preservation (cryonics with future revival) is the only strategy compatible with all eight theories, while gradual bio-hybrid replacement is the strongest non-cryonic option (compatible with 7 of 8). I identify a "deflation paradox": theories most favorable to preservation deflate consciousness to a functional property, while theories that take phenomenal experience most seriously make preservation hardest or impossible. The primary bottleneck for consciousness preservation is not engineering but theory.