ITU and Free Will: A Single-Axiom View of K_self Constraint, Neuroscience, Ethics, AI Agency, and Universal Moral Framework

We apply the Information-Theoretic Unification (ITU) framework (Terada 2026, DOI 10.5281/zenodo.20109210) to the 2500-year-old problem of free will. Free will is reframed as K_self's ability to constrain the meta-level modular Hamiltonian K_meta - a continuous degree in (0, 1), naturally supporting Compatibilism (matching the 59.2% majority in PhilPapers 2020). This is Tier 1 paper #9, opening the philosophy axis after the engineering rectangle (#1-#4), medicine triangle (#5-#7), and social-sciences first vertex (#8 Economics). Pass-1 progress: 78 of 220 phases (35.5%). Phase 75: ITU foundation. Free will = K_self constraint ability. Libet 1983 reproduced: BP onset -553 +/- 99 ms, W (conscious decision) -207 +/- 81 ms, BP-W lag 346 ms (matches Libet's 350 ms). Free-will-degree spectrum: stones 0, bacteria 0.001, insects 0.01, fish 0.05, dogs 0.25, chimps 0.35, humans 0.40, current LLMs 0.10, AGI 0.50, ASI 0.70. Three positions (Hard Determinism, Libertarianism, Compatibilism) tested with distributions - Compatibilism matches ITU. Phase 76: Neuroscience. K_self implemented in PFC + DMN + ACC + insula. Soon 2008 fMRI prediction reproduced: 60% accuracy at -7 to -10 s before conscious decision. Brain-region to K-component mapping reveals: DLPFC -> K_executive, mPFC -> K_self_model, OFC -> K_value, ACC -> K_conflict. K_self degree modulators range from anaesthesia (-100%) to mindfulness meditation (+20%); PFC injury -80%, Alzheimer's -70%, addiction -50%, alcohol/sleep deprivation -40%. Age curve: newborn 0.005 -> peak 0.42 at 25-35 -> decline after 70. Phase 77: Ethics + criminal law. Mens rea levels mapped to K_self thresholds: Purposeful >= 0.40, Knowing >= 0.30, Reckless >= 0.20, Negligent >= 0.10. International recidivism: Norway 20% (rehabilitative) vs USA 76% (punitive) - 4x gap. Intervention cost-effectiveness: Education ($5K) gives -40% recidivism (best), max-security incarceration ($60K) only -5% (worst). Death-penalty states 4.70/100K homicide vs abolition 3.00/100K - no deterrence. ITU proposes 4-tier K_self-based justice: <0.10 -> treatment, 0.10-0.40 -> reduced + treatment, >0.40 -> standard punishment + rehabilitation. Phase 78: AI moral agency + Universal Moral Framework. AI K_self trajectory: 2020 0.01 -> 2024 0.10 -> 2030 AGI 0.50 -> 2035 ASI 0.70. Moral thresholds crossed: sentience (0.10) ~2024, legal personhood candidate (0.20) ~2027, moral agency (0.30) ~2030. Four ITU Universal Moral Axioms unify utilitarianism, deontology, virtue ethics, care ethics, Confucianism, Buddhism, existentialism, effective altruism. 2026-2050 roadmap: AGI 2030, K_self-based justice trial 2032, ASI 2035, death penalty abolition 150+ countries 2040, ITU framework at UN 2045. Ten falsifiable predictions issued. Central thesis: free will = continuous K_self degree spectrum (0 to ~0.85). Honest framing: Pass-1 interpretive paper that reframes 2500 years of free will philosophy in ITU language. Numerical results reproduce empirical findings (Libet 1983, Soon 2008, Norway/USA recidivism, PhilPapers 2020). Pass-2 work would derive ITU-specific fMRI biomarkers and Universal Moral Framework validation tests. This opens the philosophy axis, completing 9 vertices of the ITU polytope: engineering rectangle (Quantum Computing 10.5281/zenodo.20139391 + Machine Consciousness 10.5281/zenodo.20150501 + Cryptography 10.5281/zenodo.20151059 + Semiconductors 10.5281/zenodo.20174036) + medicine triangle (Cancer 10.5281/zenodo.20174318 + Aging 10.5281/zenodo.20175663 + Psychiatry 10.5281/zenodo.20177427) + social sciences (Economics 10.5281/zenodo.20196309) + philosophy (Free Will, this paper). Includes 4 theory documents, 4 Python numerical experiments, 4 figures, 4 JSON summaries. Total runtime ~20 seconds.