Chapter 28: Geometric Military Ethics — War and Constrained Pathfinding on the Decision Manifold
RUNNING EXAMPLE — Priya’s Model
One patient in the BEACON-7 cohort is Sergeant (Ret.) David Kowalski, whose melanoma is linked to burn-pit exposure during deployment. His VA benefits cover trial participation, but TrialMatch does not recognize VA coverage as equivalent to private insurance—a feature encoding error that systematically disadvantages veterans. The constrained pathfinding of military ethics—jus in bello as bounded optimization under rules of engagement—mirrors TrialMatch’s own constrained optimization. Both systems face the same question: who sets the constraints, and are the constraints just?
This chapter applies the Geometric Ethics framework to military decision-making and the ethics of armed conflict. The dominant tradition in military ethics — just war theory, from Augustine through Aquinas to Walzer — provides a rich multi-dimensional vocabulary: just cause, right authority, proportionality, distinction, military necessity. Yet the formalization of these concepts in International Humanitarian Law (IHL) and operational practice has systematically reduced them to scalar tests. Proportionality becomes a cost-benefit comparison of military advantage against civilian harm. Distinction becomes a binary classification: combatant or civilian. Rules of engagement become checklists. The geometric framework reveals that these scalar reductions destroy information irrecoverably, and that the characteristic pathologies of modern warfare — civilian casualty spirals, moral injury epidemics, autonomous weapons failures, and post-conflict instability — are predictable consequences of dimensional collapse in the highest-stakes domain of human decision-making.
28.1 The Failure of Scalar Military Ethics
Military decision-making has historically been dominated by a single scalar: military necessity. From Clausewitz's dictum that war is the continuation of politics by other means to the modern doctrine of "effects-based operations," the tendency has been to evaluate military actions primarily on d_1 — mission effectiveness, tactical advantage, force protection. This scalar reduction is not merely an intellectual simplification; it is operationally embedded in targeting processes, command structures, and weapons systems.
Just war theory, in its classical formulation, resisted this collapse. Augustine's requirement of right intention (d_7: the moral character of the belligerent), Aquinas's insistence on legitimate authority (d_8: institutional sanction), and Grotius's development of proportionality (d_3: fairness between harm and advantage) together constitute a genuinely multi-dimensional framework. Walzer's "Just and Unjust Wars" (1977) further enriched the tradition by attending to the moral experience of soldiers (d_7), the rights of civilians (d_2, d_4), and the epistemic demands of targeting (d_9).
Yet the translation of just war theory into positive law — the Geneva Conventions (1949), the Additional Protocols (1977), the Rome Statute (1998) — has progressively flattened this multi-dimensional structure. The IHL proportionality test, as codified in Additional Protocol I, Article 51(5)(b), requires that an attack not be launched which "may be expected to cause incidental loss of civilian life, injury to civilians, damage to civilian objects, or a combination thereof, which would be excessive in relation to the concrete and direct military advantage anticipated." This is a scalar comparison: one number (civilian harm) weighed against another (military advantage). The nine dimensions of the moral manifold are collapsed to a single cost-benefit ratio.
Rules of engagement (ROE) attempt to restore multi-dimensionality by encoding constraints as checklists — positive identification requirements (d_9), escalation-of-force procedures (d_3), protected-site restrictions (d_2). But checklists are scalar approximations of manifold pathfinding: they discretize continuous dimensional trade-offs into binary gates, losing the geometric structure that determines which paths through moral space are genuinely optimal.
28.2 The Military Decision Complex
Definition 28.1 (Military Decision Complex). The military decision complex M_mil is a weighted simplicial complex whose vertices are military states — configurations of force disposition, civilian population status, territorial control, intelligence picture, and command authority — and whose edges are military actions (engagements, movements, targeting decisions, rules of engagement modifications, ceasefire initiatives). Each vertex v_i carries an attribute vector a(v_i) in R^9.
The nine dimensions of the moral manifold, instantiated for military contexts, are:
d_1: Military effectiveness — mission success probability, force protection, tactical and strategic advantage. This is the dimension on which military necessity operates.
d_2: Legal obligations — compliance with IHL, the Geneva Conventions, the Laws of Armed Conflict (LOAC), status-of-forces agreements, and applicable ROE. These are binding constraints, not optional considerations.
d_3: Proportionality and fairness — the balance between military advantage and civilian harm, proportionate use of force, the principle of unnecessary suffering (Hague Convention IV, Art. 23(e)).
d_4: Combatant and civilian autonomy — freedom of movement for civilian populations, rights of detainees (Geneva Convention III), prohibition of forced displacement (Rome Statute, Art. 8(2)(e)(viii)).
d_5: Trust — between allied forces, between military and civilian populations, between combatants and their chain of command, between belligerents and mediators. Trust is the lubricant of both military operations and post-conflict stability.
d_6: Strategic and societal impact — collateral damage to infrastructure, displacement of populations, environmental destruction, long-term economic consequences, cultural heritage destruction.
d_7: Warrior identity and honor — the soldier's moral self-concept, military ethos, professional ethics, the traditions of chivalry and martial honor that constrain behavior even in extremis.
d_8: Institutional legitimacy — UN Security Council authorization, democratic mandate for the use of force, legitimacy of the chain of command, international coalition support.
d_9: Intelligence and epistemic status — reliability of intelligence, certainty of target identification, fog of war, communication integrity, situational awareness.
Definition 28.2 (Military Edge Weights). The weight of an edge (v_i, v_j) in M_mil is:
w(v_i, v_j) = ΔaT ΣM−1 Da + Σk βk * 𝟙[boundary k crossed]
where Da = a(v_j) - a(v_i), Sigma_M is the 9x9 military covariance matrix, and βk are boundary penalties. Critical boundary types include:
War crime boundaries (β = infinity): deliberate targeting of civilians, torture, perfidy, use of prohibited weapons. These are sacred-value boundaries that no military advantage can justify.
Geneva Convention thresholds: treatment of prisoners of war, protection of medical facilities, protection of cultural property. These carry high but finite β values.
ROE constraints: engagement authorization levels, escalation-of-force steps, positive identification requirements. These are operationally calibrated boundary penalties.
Critical covariance terms include Sigma_{1,9} (effectiveness x intelligence: military effectiveness is coupled to intelligence quality — poor intelligence degrades all targeting decisions), Sigma_{5,6} (trust x societal impact: civilian trust erodes with collateral damage, creating insurgency feedback loops), and Sigma_{7,3} (warrior identity x proportionality: soldiers whose identity includes proportionality norms internalize these constraints at lower cognitive cost).
The military decision complex is distinguished from other domain instantiations by the extremity of its d_9 conditions. The "fog of war" — Clausewitz's central insight — is not merely informational uncertainty but a structural feature of the manifold: d_9 variance in combat is higher than in any other domain, and this variance propagates through the covariance matrix to degrade every other dimension simultaneously.
28.3 Jus ad Bellum as Manifold Entry Conditions
The decision to go to war — jus ad bellum — is the highest-stakes decision in the military domain. The just war tradition, from Augustine to the UN Charter, identifies six criteria: just cause, right authority, right intention, last resort, proportionality (ad bellum), and reasonable chance of success. The geometric framework formalizes these as entry conditions on the military decision complex.
Theorem 28.1 (Jus ad Bellum as Manifold Threshold). The decision to go to war is justified if and only if the war-path geodesic on the full military decision complex satisfies entry conditions on specified dimensions: (i) d_1 (just cause): the threat level exceeds the Just Cause Threshold, below which the peace-path geodesic dominates the war-path on the full manifold; (ii) d_8 (right authority): the decision is made by a legitimate authority with institutional sanction; (iii) d_7 (right intention): the goal region of the war-path geodesic is a morally acceptable end state, not territorial expansion, resource extraction, or revenge; (iv) d_2 (last resort): all non-violent paths on the manifold have been exhausted or are demonstrably infeasible; (v) d_3 (proportionality ad bellum): the total expected cost of the war-path across all nine dimensions does not exceed the cost of the threat it addresses; and (vi) d_1 (reasonable chance of success): the war-path geodesic reaches the goal region with non-negligible probability.
Proof. The war-path and peace-path are competing geodesics on M_mil. The peace-path has lower d_1 immediate cost (no combat casualties) but may have higher long-term d_1 cost if the threat materializes (aggression unchecked, genocide unimpeded). The war-path has higher immediate d_1 cost but may have lower long-term cost if the threat is neutralized. The six jus ad bellum criteria are necessary conditions for the war-path to dominate: just cause ensures the threat is real (the d_1 differential is genuine, not fabricated); right authority ensures the decision has institutional backing (d_8 is satisfied); right intention ensures the goal region is acceptable (d_7 is not compromised); last resort ensures the peace-path is genuinely infeasible (d_2 obligations have been met); proportionality ad bellum ensures the total manifold cost is justified; and reasonable chance of success ensures the war-path actually reaches the goal region. If any condition fails, there exists a peace-path that dominates on the full manifold. []
Definition 28.3 (Just Cause Threshold). The Just Cause Threshold T_jc is the minimum d_1 threat level below which the war-path geodesic is dominated by the peace-path on the full manifold. Formally, T_jc is the d_1 value at which the total manifold cost of the war-path (including d_2 through d_9 costs of armed conflict) equals the total manifold cost of the peace-path (including the d_1 cost of the unrealized threat). Below T_jc, the peace-path is cheaper on the full manifold even if the war-path is cheaper on d_1 alone.
Application: Iraq 2003 as Jus ad Bellum Failure. The 2003 invasion of Iraq violated at least three jus ad bellum conditions simultaneously. On d_9 (intelligence), the casus belli — Iraqi weapons of mass destruction — was catastrophically wrong; the intelligence was fabricated, cherry-picked, or misinterpreted, meaning the Just Cause Threshold was not actually met because the threat was illusory. On d_8 (institutional legitimacy), the invasion lacked UN Security Council authorization; the "coalition of the willing" was a workaround for institutional rejection, not a legitimate substitute. On d_7 (right intention), the goal region was ambiguous — regime change, democracy promotion, WMD elimination, and regional transformation were variously cited, none coherently specified. The geometric framework predicts that violations of multiple entry conditions simultaneously should produce catastrophic failure of the resulting war-path, a prediction borne out by the subsequent insurgency, sectarian civil war, and regional destabilization.
Remark (Nuclear Weapons and the Just Cause Threshold). Nuclear weapons raise the Just Cause Threshold T_jc to near-infinity for wars between nuclear-armed states: the total manifold cost of nuclear exchange on d_1 (mass casualties), d_6 (civilizational destruction), d_4 (annihilation of autonomy), and d_5 (permanent trust destruction) is so extreme that no threat short of existential attack can justify crossing T_jc. This is the geometric content of nuclear deterrence: the war-path geodesic between nuclear-armed states is dominated by the peace-path on the full manifold for virtually all threat levels, because the d_6 cost of the war-path includes potential civilizational extinction.
28.4 Jus in Bello as Constrained Pathfinding
Once armed conflict has begun, jus in bello constrains how the war is fought. The geometric framework formalizes jus in bello as constrained A* pathfinding on the military decision complex: the soldier's feasible path set is restricted by boundary penalties on non-d_1 dimensions, even when unconstrained paths would be militarily optimal.
Theorem 28.2 (Jus in Bello Constraints). Within the military decision complex, combat operations are constrained A* pathfinding: the soldier's feasible path set F_constrained is a proper subset of the unconstrained feasible set F_unconstrained, restricted by boundary penalties on d_2 (IHL obligations: no targeting of civilians, no torture, no perfidy), d_3 (proportionality in bello: incidental harm must not be excessive), and d_4 (civilian autonomy: no forced displacement, humane treatment of detainees). The minimum-cost path on F_constrained has cost greater than or equal to the minimum-cost path on F_unconstrained.
Proof. The constrained feasible set F_constrained is obtained from F_unconstrained by adding boundary penalties βk > 0 for crossings of d_2, d_3, and d_4 boundaries. Adding penalties can only increase or maintain the cost of any path that crosses a boundary, and can only decrease or maintain the size of the feasible set (paths whose boundary-crossing cost exceeds any finite budget become infeasible). Therefore F_constrained is a subset of F_unconstrained. The minimum-cost path on a subset is at least as costly as the minimum-cost path on the full set. The difference C_constrained - C_unconstrained >= 0 is the jus in bello cost — the cost of fighting lawfully. []
Corollary 28.1 (The Ethical Cost of Lawful Combat). The jus in bello cost C_jib = C_constrained - C_unconstrained is strictly positive whenever the unconstrained optimal path crosses at least one jus in bello boundary. This cost is borne by the soldier, the military unit, and the nation as reduced military effectiveness on d_1. The acceptance of this cost is the defining commitment of lawful belligerency: a military that refuses to pay C_jib is, by definition, committing war crimes.
The practical significance is immediate. When commanders complain that IHL "ties our hands," they are correctly identifying C_jib > 0. The geometric framework validates this observation while insisting that the cost is non-negotiable: the boundary penalties on d_2, d_3, and d_4 encode sacred-value constraints (β = infinity for war crime boundaries) that cannot be traded against d_1 gains regardless of their magnitude.
28.5 Proportionality as Multi-Dimensional Cost-Benefit
The IHL proportionality test is the most contested principle in the law of armed conflict. Additional Protocol I, Article 51(5)(b) prohibits attacks "which may be expected to cause incidental loss of civilian life, injury to civilians, damage to civilian objects, or a combination thereof, which would be excessive in relation to the concrete and direct military advantage anticipated." This formulation invites scalar comparison: one number (civilian harm) weighed against another (military advantage).
Theorem 28.3 (Scalar Proportionality Failure). The IHL proportionality test, as typically applied, compares d_1 cost (civilian casualties as a subset of consequences) against d_1 benefit (military advantage as another subset of consequences). This is a one-dimensional comparison within a single dimension. The geometric framework demonstrates that this scalar test collapses at least four additional dimensions: d_2 (civilian rights to protection under IHL), d_4 (civilian autonomy — displacement, loss of freedom of movement), d_6 (societal impact — infrastructure destruction, community disruption, environmental damage), and d_7 (cultural heritage and identity — destruction of religious sites, historical monuments, cultural practices). By Scalar Irrecoverability (Chapter 15), the information destroyed by this collapse is mathematically unrecoverable: no post hoc analysis can reconstruct the multi-dimensional cost from the scalar comparison.
Proof. The scalar proportionality test evaluates f: R^9 -> R, mapping the nine-dimensional cost-benefit vector to a single ratio. By Brouwer's invariance of dimension, no continuous injection from R^9 to R exists. Therefore, distinct nine-dimensional states — attacks with identical civilian casualty counts but vastly different d_4, d_6, and d_7 costs — map to the same scalar proportionality score. The test declares them equally proportionate when they are, in the geometric sense, profoundly different. []
Proposition 28.1 (Geometric Proportionality). Geometric proportionality requires that the military action's cost on EACH non-d_1 dimension be independently proportionate to the military advantage on d_1. Formally, for each dimension k in {d_2, d_3, d_4, d_5, d_6, d_7, d_8, d_9}, the ratio |Da_k| / |Da_1^{benefit}| must not exceed a dimension-specific threshold tau_k. This multi-dimensional test preserves the information that scalar proportionality destroys.
Lemma 28.1 (Proportionality Asymmetry). The scalar proportionality test is asymmetrically biased toward the attacker. The military advantage (d_1 benefit) is evaluated from the attacker's perspective using the attacker's information (d_9). The civilian harm (d_1 cost) is evaluated using the attacker's estimate of civilian presence, which is systematically lower than the true civilian cost because the attacker has incentives to underestimate and lacks ground-truth access. In the geometric framework, this asymmetry means the scalar proportionality test is computed with biased inputs: the d_1 benefit is overestimated (motivated reasoning) and the d_1 cost is underestimated (information asymmetry). The resulting proportionality ratio is systematically biased toward permitting strikes.
Example. Consider the bombing of a hospital that serves as a military command post. The scalar proportionality test weighs X civilian casualties (d_1 cost) against the military value of destroying the command post (d_1 benefit). If the ratio is acceptable, the strike is proportionate. But the geometric test identifies additional costs invisible to the scalar test: destruction of medical infrastructure serving a civilian population of thousands (d_6), violation of the protected status of medical facilities under Geneva Convention I, Art. 19 (d_2), forced displacement of patients and medical personnel (d_4), destruction of community trust in the protection of medical facilities (d_5), and damage to the cultural norm that hospitals are inviolable (d_7). Each of these costs must be independently proportionate to the military advantage. The scalar test captures only the first; the geometric test captures all six.
28.6 The Doctrine of Double Effect as Dimensional Decomposition
The Doctrine of Double Effect (DDE), originating in Aquinas's Summa Theologiae (II-II, Q. 64, Art. 7), holds that an action with both good and bad effects is permissible if four conditions are met: (1) the action itself is not intrinsically wrong; (2) the agent intends only the good effect; (3) the bad effect is not a means to the good effect; and (4) the good effect is proportionate to the bad effect. DDE is foundational to military targeting: the killing of civilians as a side effect of a legitimate military strike may be permissible, while the deliberate targeting of civilians is always prohibited.
Theorem 28.4 (DDE as Dimensional Decomposition). The Doctrine of Double Effect decomposes a military action into its effects on each dimension separately. The "good effect" is the d_1 (military effectiveness) component of the action. The "bad effect" is the aggregate of d_1 (civilian harm), d_2 (rights violations), d_3 (proportionality costs), d_4 (autonomy restrictions), d_5 (trust damage), and d_6 (societal impact) components. DDE's condition (3) — "the bad effect is not a means to the good effect" — requires that the bad-effect dimensions are not instrumentally necessary for achieving the good-effect dimension. Formally: there exists an alternative path P' with comparable d_1 benefit (|Da_1(P') - Da_1(P)| < epsilon) and strictly lower cost on at least one bad-effect dimension (Da_k(P') < Da_k(P) for some k in {d_2, d_3, d_4, d_5, d_6}) if and only if the bad effect is NOT a means. If no such P' exists — if the bad-effect cost is inextricable from the good-effect benefit — then the bad effect IS a means, and DDE prohibits the action.
Proof. DDE's condition (3) distinguishes instrumental from incidental harm. In the geometric framework, instrumental harm means that the path from current state to goal state necessarily traverses the harmful vertices — there is no detour. Incidental harm means that the path traverses harmful vertices but an alternative path with comparable d_1 benefit avoids them. The existence of such an alternative path P' is a topological property of the military decision complex: it holds when the harmful vertices are not on every geodesic from the current state to the goal region. If every geodesic traverses the harmful vertices, the harm is instrumental (a necessary means), and DDE prohibits the action. If at least one geodesic avoids them, the harm is incidental, and DDE permits the action subject to proportionality (condition 4). []
Application: Strategic Bombing versus Precision Strikes. The Allied strategic bombing campaign of World War II — the firebombing of Dresden, Tokyo, Hamburg — treated d_6 (civilian infrastructure destruction) as instrumental to d_1 (military effectiveness): the theory of strategic bombing held that destroying civilian morale and industrial capacity was the means to military victory. Under DDE, this is impermissible: the bad effect (mass civilian death and infrastructure destruction) is the means to the good effect (reduced enemy war-fighting capacity). Precision strikes, by contrast, target specific military objectives (a bridge, a command post, a weapons depot) with civilian harm as incidental — an alternative path with comparable d_1 benefit and lower d_6 cost exists (a more precise weapon, a different angle of attack, a different time of day). DDE permits precision strikes subject to proportionality but prohibits strategic bombing that treats civilian destruction as the mechanism of military effect.
Remark (The Trolley Problem Formalized). In the classic trolley problem, the fat man's body stops the trolley — the bad effect (killing the man) IS the means to the good effect (stopping the trolley). In the geometric framework, there is no alternative path P' that achieves the same d_1 benefit (trolley stopped) without the d_2 cost (killing the man): the harmful vertex is on every geodesic to the goal. DDE prohibits pushing the fat man. In the switch variant, diverting the trolley kills a bystander incidentally — the death is not the mechanism that saves the five. An alternative path exists in principle (stop the trolley by other means), but if no such path is feasible, DDE permits diversion subject to proportionality.
28.7 Civilian-Combatant Distinction as Boundary Classification
The principle of distinction — the requirement that parties to a conflict distinguish between combatants and civilians, and direct attacks only against combatants — is codified in Additional Protocol I, Article 48, and is recognized as customary international law binding on all states. In the geometric framework, distinction is a boundary classification problem on the military decision complex.
Definition 28.4 (Distinction as Boundary Classification). Each vertex in the military decision complex is classified as combatant (targetable) or civilian (protected). The boundary penalty for targeting a vertex classified as civilian is beta_civilian = infinity. This is a sacred-value boundary: no military advantage, however great, justifies the deliberate targeting of a civilian. The penalty is not finite and large — it is infinite, reflecting the jus cogens status of the prohibition on deliberate civilian targeting under the Rome Statute, Article 8(2)(b)(i).
Theorem 28.5 (Distinction Degradation in Asymmetric Warfare). In asymmetric warfare — insurgency, counterterrorism, urban combat — the combatant-civilian boundary becomes probabilistic. Vertices cannot be cleanly classified as combatant or civilian: the insurgent is a farmer by day and a fighter by night; the civilian carries a phone that may or may not be a detonator; the building is a home that may or may not contain a weapons cache. The effective boundary penalty becomes:
beta_effective = beta_civilian * P(civilian | observation)
where P(civilian | observation) is the probability that the target is civilian given available intelligence. As d_9 (intelligence quality) degrades, P(civilian | observation) approaches the base rate P(civilian) in the population — which, in most conflict zones, is greater than 0.9. The distinction boundary becomes computationally intractable: nearly every engagement carries near-infinite expected boundary cost.
Proof. Under perfect intelligence (d_9 maximum), P(civilian | observation) = 0 for combatants and P(civilian | observation) = 1 for civilians. The boundary classification is deterministic, and the boundary penalty is applied correctly. Under degraded intelligence, P(civilian | observation) is a continuous function of observation quality on [0, 1]. The expected boundary cost for targeting vertex v is E[β] = beta_civilian * P(civilian | observation at v). Since beta_civilian = infinity, the expected cost is infinite whenever P(civilian | observation) > 0. In practice, forces treat beta_civilian as very large but finite, implicitly accepting a non-zero probability of civilian targeting. The d_9 degradation characteristic of asymmetric warfare drives P(civilian | observation) upward for every engagement, systematically increasing the expected boundary cost and the probability of civilian casualties. []
Application: Human Shields as Boundary Exploitation. The use of human shields — placing civilian vertices adjacent to combatant vertices — is a deliberate exploitation of the distinction boundary. The adversary forces the geometric structure of M_mil into a configuration where combatant and civilian vertices are topologically inseparable: every path to the combatant vertex traverses civilian vertices. This makes the distinction boundary computationally intractable in the most literal sense: there is no path to the legitimate military objective that does not cross the sacred-value boundary. The attacker faces a dilemma that is geometric, not merely moral: the manifold has been engineered to make lawful engagement impossible. IHL addresses this by placing responsibility on the party using human shields (Additional Protocol I, Art. 51(7)) while maintaining that the attacking party's obligation to distinguish is not extinguished.
Siege Warfare and Collective Punishment
Siege warfare — the encirclement and starvation of a population to compel surrender — is the limiting case of distinction degradation. In a siege, the distinction boundary is not merely ambiguous but deliberately ignored: the entire besieged population, combatant and civilian alike, is subjected to d_4 (autonomy: denied freedom of movement), d_6 (societal impact: destruction of infrastructure, food systems, medical care), and d_1 (consequences: starvation, disease, death) costs. The geometric framework classifies siege that targets civilian survival as collective punishment — a war crime under Geneva Convention IV, Article 33 — because it sets beta_civilian = 0 for an entire population, treating every vertex as targetable regardless of classification.
The framework predicts a testable correlation: civilian casualty rates should increase monotonically with d_9 degradation. In conflicts with good intelligence (clear front lines, uniformed combatants, reliable surveillance), civilian casualties should be lower. In conflicts with poor intelligence (urban insurgency, unmarked combatants, degraded communications), civilian casualties should be higher. The empirical evidence — comparing casualty ratios across conventional warfare (World War II European theater), counterinsurgency (Vietnam, Iraq), and urban warfare (Gaza, Mosul) — is consistent with this prediction.
28.8 Autonomous Weapons and Dimensional Collapse
Autonomous weapons systems (AWS) — weapons that can select and engage targets without human intervention — represent the military instantiation of the dimensional collapse phenomenon analyzed in Chapter 23 (Flash Crash) and Chapter 26 (AI Ethics). This section addresses AWS from the military ethics perspective, focusing on IHL compliance and just war theory; Chapter 26 addresses them from the AI ethics perspective, focusing on alignment and control.
Theorem 28.6 (AWS Dimensional Collapse). An autonomous weapon system optimizing on d_1 (target elimination effectiveness) with algorithmic d_9 (sensor-based target identification) operates on a collapsed manifold of at most two dimensions. The system cannot maintain the distinction boundary (Definition 28.4) because boundary classification requires evaluation on d_2 (legal status of the target under IHL — is this person a combatant, a civilian, a medical worker, a surrendering soldier?), d_3 (proportionality of this specific engagement in this specific context — is the expected civilian harm excessive relative to the military advantage of this particular target?), d_7 (ethical judgment — does this engagement comport with the warrior ethos, with the spirit and not merely the letter of the law?), and contextual d_9 (human intelligence — information about local conditions, cultural context, and recent events that sensor data cannot capture). Current AI systems cannot compute these dimensions.
Proof. An AWS receives sensor data (visual, infrared, radar) and classifies targets using pattern recognition algorithms trained on labeled datasets. The classification operates on d_1 (is this a militarily valuable target?) and d_9 (how confident is the identification?). The system's edge weights are: w_AWS(v_i, v_j) = (Da_1)^2 / sigma_11 + (Da_9)^2 / sigma_99. Boundary penalties on d_2, d_3, d_4, d_7 are set to zero — not because the designers intended to ignore them, but because the system has no computational representation of these dimensions. The feasible path set under this collapsed metric strictly contains the feasible path set under the full IHL-compliant metric: the AWS can traverse paths (engage targets) that a full-manifold agent (a human soldier applying IHL) would never traverse because they cross d_2, d_3, or d_7 boundaries. This is the lethal analog of the Flash Crash (Chapter 23): just as algorithmic trading systems caused the Flash Crash by collapsing to d_1 (monetary return), AWS collapse to d_1 (target elimination) in a domain where the consequences of dimensional collapse are measured in human lives rather than dollars. []
Proposition 28.2 (Meaningful Human Control Requirement). Compliance with IHL requires that autonomous weapons systems maintain human oversight with minimum dimensionality k >= 4. The four minimum dimensions are: d_1 (target military value), d_2 (target legal status under IHL), d_3 (proportionality of the specific engagement), and d_9 (contextual intelligence beyond sensor data). Below this threshold, the system cannot reliably distinguish lawful from unlawful engagements, and every firing decision carries non-trivial probability of war crime boundary violation.
Remark (The Accountability Gap). AWS introduce a novel problem for IHL: the accountability gap. Under the Rome Statute, criminal responsibility attaches to individuals who order, commit, or fail to prevent war crimes. When an AWS commits a targeting error that constitutes a war crime, who is responsible? The commander who authorized deployment? The engineer who designed the algorithm? The procurement officer who selected the system? In the geometric framework, the accountability gap is a consequence of dimensional collapse: the AWS's decision was made on a two-dimensional manifold (d_1, d_9) where the war crime boundary (a d_2 boundary) does not exist. No human decision-maker traversed the boundary — the boundary was absent from the decision space entirely. The accountability gap is not a bug in the law; it is a structural consequence of delegating lethal decisions to systems that operate on collapsed manifolds.
Current technology assessment. Image classifiers — the core targeting technology in contemporary AWS — operate on d_1 (is this a tank, a truck, a person carrying a weapon?) and algorithmic d_9 (confidence score from the classifier). They cannot evaluate d_2: the legal status of a target under IHL depends on facts that no image classifier can determine — is this person a combatant who has surrendered? Is this building a protected medical facility whose protection has been forfeited by military use? Is this person a child soldier whose targeting raises distinct legal questions? They cannot evaluate d_3: proportionality requires contextual judgment about the military advantage of this specific engagement relative to the expected civilian harm in this specific situation — a judgment that depends on the overall military campaign, the availability of alternative means, and the density of the civilian population, none of which are accessible to a sensor-based classifier. The gap between what AWS can compute (d_1, d_9) and what IHL requires (d_1, d_2, d_3, d_7, d_9) is not an engineering problem to be solved by better algorithms; it is a dimensional deficit that no sensor-based system can close.
28.9 Rules of Engagement as Heuristic Constraints
Rules of engagement are pre-programmed decision rules issued to soldiers and military units to govern the use of force. In the geometric framework, ROE are heuristic constraints — pre-computed boundary penalties and dimensional weights that structure the soldier's pathfinding heuristic h(n), enabling rapid decision-making in high-stress, low-d_9 conditions where full manifold deliberation is cognitively impossible.
Definition 28.5 (ROE as Heuristic Constraints). Rules of engagement are a heuristic function h_ROE(n) that approximates the full manifold cost h*(n) by encoding boundary penalties and dimensional priorities as discrete rules. Examples: "Do not fire unless fired upon" encodes a high d_3 boundary penalty (proportionality: the first shot carries asymmetric cost). "Positive identification required before engagement" encodes a d_9 threshold (epistemic: do not engage until target identification confidence exceeds a minimum). "Do not target buildings marked with a red cross" encodes a d_2 boundary penalty (legal: medical facilities are protected). Each rule is a dimensional constraint translated into an operationally executable instruction.
Proposition 28.3 (ROE as Epsilon-Admissible Heuristics). ROE sacrifice optimality for computational tractability. The soldier in combat cannot perform full nine-dimensional A* pathfinding in real time; the cognitive and temporal demands exceed human capacity under stress. ROE pre-compute an approximate heuristic h_ROE(n) that is epsilon-admissible: |h_ROE(n) - h*(n)| <= epsilon for most military states n. The parameter epsilon is the gap between the ROE-guided path and the fully deliberated manifold geodesic — the cost of heuristic approximation.
ROE failure modes are characterized by the nature of epsilon:
ROE too restrictive: epsilon is large and biased upward on d_1 — the heuristic overestimates the cost of engagement, preventing soldiers from accomplishing legitimate military objectives. The soldier cannot fire even when firing is lawful and necessary. This was a persistent complaint in early ISAF operations in Afghanistan, where restrictive ROE were perceived as prioritizing d_3 (proportionality) over d_1 (force protection) to a degree that endangered soldiers.
ROE too permissive: boundary penalties are reduced or eliminated on d_2, d_3, and d_4 — the heuristic underestimates the cost of boundary crossings, enabling paths that constitute war crimes. The soldier can fire when firing is unlawful. This is the failure mode of Vietnam-era "free fire zones," where ROE designated geographical areas in which beta_civilian was effectively set to zero, removing the distinction boundary entirely.
ROE poorly calibrated: the heuristic is inadmissible — it sometimes overestimates and sometimes underestimates h*(n), producing paths that oscillate between excessive restraint and excessive force. This is the failure mode of ROE that are written for one operational environment and applied in another.
Application: My Lai as ROE Boundary Removal. On March 16, 1968, soldiers of Charlie Company, 1st Battalion, 20th Infantry Regiment massacred between 347 and 504 unarmed Vietnamese civilians in the hamlet of My Lai. The geometric analysis is precise: the operational context (repeated ambushes, booby traps, indistinguishable combatants and civilians) had degraded d_9 to near-zero. The implicit ROE for the operation — search and destroy in a designated "free fire zone" — set beta_civilian to zero in that area. With the distinction boundary removed and d_9 degraded, the soldiers' heuristic h_ROE(n) contained no penalty for civilian targeting. The path through the massacre was, on the collapsed manifold, the lowest-cost path on d_1 (eliminate all potential threats). On the full manifold, it crossed the war crime boundary (β = infinity) hundreds of times. My Lai is the predictable consequence of ROE that collapse the military decision complex to d_1 alone.
28.10 Moral Injury in Combat as Manifold Damage
Moral injury — the psychological, biological, and spiritual damage sustained when a person perpetrates, fails to prevent, or witnesses acts that transgress their deeply held moral beliefs — is epidemic among combat veterans. The geometric framework, building on the moral injury theory developed in Chapter 21, provides a precise characterization of combat moral injury as cumulative damage to the soldier's heuristic function.
Theorem 28.7 (Combat Moral Injury). Military moral injury occurs when the soldier is forced to traverse paths that cross their own moral boundaries — paths that their heuristic h(n) flags as high-cost but that orders, operational pressure, or circumstance require them to take. The damage is to the heuristic function itself: repeated forced boundary crossings degrade h(n)'s calibration. Specifically, let B_soldier be the set of the soldier's internalized moral boundaries (a subset of the full boundary set B_mil). A forced boundary crossing on boundary b in B_soldier inflicts damage proportional to the soldier's boundary penalty beta_b^{soldier} — the strength of the moral conviction violated. Cumulative damage D = Σt beta_{b(t)}^{soldier} * 𝟙[boundary b(t) crossed at time t] degrades h(n) by increasing its variance and reducing its reliability.
Proof. The soldier's heuristic h(n) is calibrated through training, socialization, and moral development to reflect internalized boundary penalties. When the soldier is forced to cross boundary b — ordered to fire on a target they believe to be civilian, compelled to stand by while allies commit abuses, required to execute a strike they judge disproportionate — the forced crossing creates a conflict between h(n)'s output (high cost: do not traverse) and the executed action (traverse anyway). This conflict damages h(n)'s calibration: the soldier's internal cost function no longer reliably guides action, because external commands override internal evaluation. Over time, repeated overrides produce one of two pathological adaptations: (a) h(n) collapses — the soldier stops evaluating boundary costs entirely, becoming morally numb (dimensional collapse to d_1: just follow orders), or (b) h(n) becomes hypersensitive — the soldier assigns extreme cost to all actions, producing paralysis, hypervigilance, and guilt. Both are manifestations of heuristic damage, not pathological weakness. []
Remark (Moral Injury versus PTSD). The geometric framework makes the distinction between moral injury and PTSD precise. PTSD is d_1 trauma: the soldier's threat-detection system has been damaged by exposure to life-threatening events, producing hyperarousal, avoidance, and re-experiencing of the threat. Moral injury is d_7 damage: the soldier's moral identity — their sense of themselves as a good person, a lawful combatant, a protector — has been shattered by what they did, what they failed to prevent, or what they witnessed. The two can co-occur but are mechanistically distinct. PTSD responds to exposure therapy (recalibrating the d_1 threat detector); moral injury requires narrative reconstruction (repairing the d_7 identity function). This distinction, predicted by the geometric framework, is now supported by clinical evidence (Litz et al., 2009; Shay, 2014).
Corollary 28.2 (The Moral Injury Paradox of Lawful Combat). Soldiers who fight lawfully — who maintain boundary penalties on d_2, d_3, and d_4 even under extreme pressure — are more vulnerable to moral injury than soldiers who fight unlawfully. The lawful soldier's heuristic h(n) has high boundary penalties (strong moral convictions), making every near-boundary decision costly. The unlawful soldier's h(n) has been collapsed (boundary penalties zeroed out), reducing the subjective cost of boundary crossings. This is the moral injury paradox: the better the soldier, the greater the damage. The geometric framework predicts that moral injury rates should be highest among soldiers with the strongest pre-deployment moral convictions — a prediction supported by Currier et al. (2015).
Application: Moral Injury in Drone Operators. Drone operators experience moral injury at rates comparable to deployed combat soldiers despite being physically safe thousands of miles from the battlefield. The geometric framework explains this apparent paradox. The drone operator's heuristic h(n) includes d_3 (proportionality: the engagement should involve reciprocal risk — a "fair fight") and d_7 (warrior identity: the soldier risks their life alongside the enemy, earning the moral authority to use lethal force). Drone strikes violate both: the target cannot fight back (d_3 asymmetry — the engagement is maximally disproportionate in terms of risk), and the operator is physically safe (d_7 warrior identity mismatch — the operator kills without personal risk, violating the warrior ethos of shared danger). The moral injury occurs even though the d_1 risk to the operator is zero, because the damage is on d_3 and d_7, not d_1. The scalar reduction of military experience to d_1 (physical danger) cannot explain drone operator moral injury; the multi-dimensional framework can.
28.11 Jus post Bellum and Manifold Repair
The just war tradition has historically focused on jus ad bellum (when to fight) and jus in bello (how to fight), with relatively little attention to jus post bellum — the ethics of post-conflict justice, reconstruction, and reconciliation. The geometric framework treats jus post bellum as manifold repair: the systematic restoration of the military decision complex's dimensional structure after the damage inflicted by armed conflict.
Proposition 28.4 (Jus post Bellum as Manifold Repair). Post-conflict justice requires the restoration of multiple dimensions of the military decision complex that were degraded by armed conflict. The primary dimensions requiring repair are: d_5 (trust between former adversaries, between military and civilian populations, and between the state and its citizens), d_6 (societal infrastructure — physical, economic, and social systems destroyed or degraded by conflict), d_8 (institutional legitimacy — the rule of law, democratic governance, and international standing), and d_4 (civilian autonomy — the restoration of freedom of movement, property rights, and civil liberties curtailed during conflict). Post-conflict stability is a function of the number of dimensions successfully repaired, not merely the magnitude of repair on any single dimension.
Application: The Nuremberg and Tokyo Tribunals as d_8 Repair. The post-World War II war crimes tribunals at Nuremberg (1945-1946) and Tokyo (1946-1948) were primarily d_8 (institutional legitimacy) repair operations. By establishing that individuals — including heads of state and senior military commanders — could be held criminally responsible for war crimes, crimes against humanity, and crimes of aggression, the tribunals restored the institutional legitimacy of international law as a constraint on state behavior. The tribunals also performed d_2 repair (reaffirming the legal obligations violated during the conflict) and d_7 repair (establishing that "following orders" does not extinguish individual moral responsibility). The geometric framework notes that the tribunals performed minimal d_5 repair (trust between former adversaries was not a primary objective) and no d_6 repair (economic reconstruction was addressed separately, through the Marshall Plan).
Application: Truth and Reconciliation Commissions as d_5 Repair. The South African Truth and Reconciliation Commission (TRC, 1996-2003) and the Rwandan Gacaca courts (2002-2012) prioritized d_5 (trust) repair through narrative acknowledgment: perpetrators testified publicly about their actions, victims told their stories, and the community witnessed the exchange. In the geometric framework, the TRC repaired d_5 by restoring the informational symmetry between perpetrators and victims — the perpetrator's private knowledge of what happened (high d_9 for the perpetrator, low d_9 for the victim and community) was made public, equalizing d_9 and enabling trust reconstruction. The TRC's amnesty provision traded d_8 repair (criminal accountability) for d_5 repair (truth-telling and reconciliation) — a dimensional trade-off that the geometric framework makes explicit.
Application: The Marshall Plan as d_6 Repair. The Marshall Plan (1948-1952) invested $13.3 billion (approximately $170 billion in 2024 dollars) in the economic reconstruction of Western Europe. In the geometric framework, this was primarily d_6 (societal infrastructure) repair, with secondary effects on d_5 (trust: American investment signaled commitment to European recovery) and d_8 (institutional legitimacy: the Plan was administered through the Organisation for European Economic Co-operation, building institutional capacity). The geometric framework predicts that d_6 repair alone is insufficient for post-conflict stability — a prediction borne out by comparing the Marshall Plan's success (paired with NATO's d_8 institutional framework) with the failure of purely economic reconstruction in Iraq (2003-2011), where d_6 investment was substantial but d_5 (trust), d_8 (institutional legitimacy), and d_4 (civilian autonomy) repair were inadequate.
Remark (Victor's Justice and Asymmetric Repair). The geometric framework identifies "victor's justice" as asymmetric manifold repair: the winning side repairs d_8 (institutional legitimacy) by prosecuting the losing side's leaders while exempting its own. This asymmetry undermines d_5 (trust) repair — the losing side perceives the process as illegitimate (d_8 damage to the tribunal itself), and d_3 (fairness) is compromised by the selective application of accountability. The geometric framework predicts that symmetric repair processes (like the TRC, where both sides testify) produce more durable d_5 restoration than asymmetric ones (like occupation-era tribunals), even if symmetric processes sacrifice some d_8 accountability. The International Criminal Court (Rome Statute, 1998) attempts to address this asymmetry by establishing jurisdiction over all parties to a conflict, though its enforcement remains selective in practice.
The framework generates a prediction: post-conflict stability should correlate with the number of dimensions repaired, controlling for the magnitude of repair on any single dimension. A post-conflict program that repairs d_5, d_6, d_8, and d_4 modestly should outperform one that repairs d_6 massively but neglects d_5, d_8, and d_4. The evidence from post-conflict reconstruction — the relative success of Germany and Japan (multi-dimensional repair) versus the relative failure of Iraq and Libya (primarily d_6 repair, with d_5 and d_8 neglect) — is consistent with this prediction.
28.12 Worked Examples
Example 28.1 (Kabul Drone Strike, August 29, 2021 — d_9 Catastrophe). Following the ISIS-K bombing at Abbey Gate that killed 13 US service members and approximately 170 Afghan civilians, US Central Command launched a "defensive" drone strike against a white Toyota Corolla in Kabul. The strike killed 10 civilians, including 7 children aged 2 to 12. The target, Zemari Ahmadi, was an aid worker for Nutrition and Education International who had been loading water canisters — not explosives — into his car. The strike achieved zero military effectiveness: no ISIS-K operatives were killed, and the stated objective of preventing an imminent attack was based on entirely false intelligence. d_1 = 0.
The dimensional analysis reveals catastrophic failure across every axis. d_2 (legal obligations): IHL requires positive identification of a military target before engagement, and the target was a civilian — a complete d_2 violation. d_3 (proportionality): even if the target had been an ISIS-K operative, a Hellfire missile strike in a residential courtyard with children present fails any proportionality test; the d_3 cost was effectively infinite given that children were killed as "collateral." d_4 (civilian autonomy): ten civilians, including children, had their most fundamental autonomy — the right to live — negated. d_5 (trust): trust between Afghans and US forces, already strained during the chaotic withdrawal, was catastrophically destroyed; trust between the military and the American public was further damaged when initial reports described the strike as "righteous" before the truth emerged weeks later.
d_6 (societal impact): the strike became a symbol of twenty years of civilian casualties in the war on terror, crystallizing the human cost of remote warfare into a single event that circled the globe. d_7 (warrior identity): the military's ethos of protecting civilians — the foundational principle that distinguishes a professional military from an armed mob — was violated in the most visible possible way. d_8 (institutional legitimacy): no military personnel were disciplined despite killing ten civilians, including seven children. The institutional failure to hold anyone accountable compounded the d_8 damage, signaling that the accountability mechanisms themselves were hollow.
d_9 (intelligence) was the catastrophic failure that drove all others. The entire kill chain was built on d_9 errors: pattern-of-life analysis misinterpreted water canisters as explosives; confirmation bias drove analysts to interpret ambiguous evidence as confirming the pre-existing narrative that the vehicle was an ISIS-K VBIED; the "imminent threat" framing created time pressure that compressed the d_9 evaluation window from hours to minutes. Geodesic analysis: the strike was an A* search with catastrophically miscalibrated d_9 — the heuristic h(n) assigned near-zero cost to the strike path because the analysts were certain (d_9 = high confidence) when they should have been deeply uncertain (d_9 = low confidence). On the correct manifold (d_9 = highly uncertain), the geodesic is "do not strike" — the d_2, d_3, and d_4 costs of an incorrect strike massively exceed the d_1 benefit of a correct one under genuine uncertainty. The Kabul strike is a canonical demonstration that d_9 miscalibration can make every other dimension irrelevant: when you are confidently wrong about who you are killing, proportionality, distinction, and necessity become meaningless checks on a meaningless calculation.
Example 28.2 (My Lai Massacre, March 16, 1968 — Systematic Dimensional Collapse). Charlie Company, 1st Battalion, 20th Infantry Regiment entered the village of My Lai (Sơn Mỹ) in Quảng Ngãi Province and over four hours killed between 347 and 504 unarmed Vietnamese civilians — women, children, the elderly. No enemy combatants were found in the village. Many victims were sexually assaulted before being killed. Homes were burned, livestock slaughtered, wells poisoned. Of the soldiers involved, only Lieutenant William Calley was convicted, serving three and a half years of house arrest for twenty-two counts of murder.
The dimensional analysis shows total collapse. d_1 (military effectiveness): zero — no Viet Cong were found or engaged; all victims were unarmed civilians. d_2 (legal obligations): every killing violated the Hague and Geneva Conventions; the soldiers were not in combat and faced no resistance. d_3 (proportionality): the violation was infinite — unarmed civilians, including infants, posed zero military threat, making any force infinitely disproportionate. d_5 (trust): the military covered up the massacre for over a year until investigative journalist Seymour Hersh broke the story in November 1969. Warrant Officer Hugh Thompson, the helicopter pilot who intervened to stop the killing by positioning his aircraft between soldiers and fleeing civilians and ordering his crew to fire on American troops if they continued, was initially shunned and threatened by fellow soldiers and superiors rather than commended.
d_7 (warrior identity and moral injury): this is where the geometric framework's explanatory power is most visible. Charlie Company had suffered weeks of casualties from mines, booby traps, and sniper fire without ever engaging a visible enemy. Their heuristic function h(n) had been systematically degraded by repeated forced boundary crossings — destroying villages under orders, operating in designated "free fire zones," taking casualties from an invisible adversary while surrounded by villagers who might or might not be sheltering that adversary. The massacre is the predictable consequence of sustained moral injury as described in Theorem 28.7: repeated forced boundary crossings eroded the boundary penalties on d_2 and d_3 to near-zero, until the civilian-combatant distinction collapsed entirely. My Lai was not an aberration; it was the terminal state of a heuristic function under sustained degradation.
d_8 (institutional legitimacy): the subsequent cover-up, the conviction of only one officer, and the leniency of his sentence devastated public trust in the military as an institution and became a catalyst for the anti-war movement. d_9 (intelligence): the company had been told My Lai harbored the 48th Local Force Battalion of the Viet Cong; no meaningful intelligence verification was performed. ROE analysis: Charlie Company operated in a designated "free fire zone" — rules of engagement that explicitly set the civilian protection weight (beta_civilian) to near-zero, exemplifying Proposition 28.3's failure mode 2 (ROE too permissive). The free-fire zone designation was a deliberate institutional removal of distinction boundaries. My Lai is the canonical case demonstrating that when an institution removes boundary penalties, catastrophic paths become not merely possible but probable.
Example 28.3 (South Africa Truth and Reconciliation Commission, 1996-2003 — Jus post Bellum as Manifold Repair). After the end of apartheid and the election of Nelson Mandela in 1994, South Africa faced the question that confronts every post-conflict society: how to address the crimes of the previous regime. The country chose a "third way" between Nuremberg-style prosecutions (which risked destabilizing the fragile new democracy) and blanket amnesty (which would leave victims unacknowledged and perpetrators unaccountable). The Truth and Reconciliation Commission, chaired by Archbishop Desmond Tutu, offered conditional amnesty to perpetrators who made full public disclosure of politically motivated crimes committed between 1960 and 1994.
d_5 (trust repair): the TRC's core mechanism was truth-telling as trust reconstruction. Perpetrators described their actions in public hearings broadcast on national television and radio; victims told their stories; and the nation heard both. Over 21,000 victim statements were taken. Of the 7,112 amnesty applications filed, only 1,512 were granted — applicants had to demonstrate that their crimes were politically motivated and that they had made full disclosure. This was d_5 repair through narrative: rebuilding trust not by punishment but by acknowledgment, not by retribution but by the perpetrator's willingness to face the community with the truth. d_6 (societal impact): by avoiding mass prosecutions that might have triggered a white exodus, economic collapse, or civil conflict, the TRC preserved social infrastructure while acknowledging the crimes of apartheid — a deliberate d_6 preservation strategy that traded retributive justice for social stability.
d_7 (identity repair): both victims and perpetrators were given the opportunity to reconstruct their narratives. Victims became agents with publicly acknowledged suffering rather than invisible casualties of a system that denied their humanity. Perpetrators were offered the possibility of reintegration through the act of acknowledgment rather than permanent exile from the social fabric. Ubuntu philosophy — "I am because we are" — provided the d_7 framework: identity is constituted through relationship, and relationship is restored through truth. d_8 (institutional legitimacy): the TRC established the new South African state's commitment to rule of law and human rights without the destabilizing effects of mass prosecutions, and the new constitution's legitimacy was reinforced by the process. d_1 (consequences and reparations): the TRC recommended reparations of 21,000 rand per year for six years per victim; the government ultimately implemented a one-time payment of 30,000 rand — far less than recommended. This d_1 failure partially undermined the d_5 and d_7 repairs, leaving many victims feeling that acknowledgment without material justice was insufficient.
Comparative dimensional analysis reveals how different post-conflict societies chose different dimensional priorities: Nuremberg emphasized d_8 (establishing legal precedent and institutional accountability through prosecution); South Africa emphasized d_5 (trust repair through truth-telling); Rwanda's gacaca courts (2002-2012) pursued d_5 repair at community scale for 1.2 million cases that would have overwhelmed any conventional court system. Each approach repaired some dimensions while necessarily sacrificing others. The geometric framework predicts (Proposition 28.4) that post-conflict stability correlates with the number of dimensions repaired, not the depth of repair on any single dimension. South Africa's relative stability compared to post-intervention societies that received only d_1 (economic) or d_8 (institutional) reconstruction supports this prediction: the TRC's multi-dimensional approach — repairing d_5, d_6, d_7, and d_8 simultaneously, even imperfectly — produced more durable peace than single-dimensional interventions that repaired one dimension deeply while neglecting the others entirely.
28.13 Connection to the Framework
The Geometric Military Ethics program connects to the parent framework in six directions:
Chapter 21 (Moral Injury) established the theory of moral injury as cumulative heuristic damage. This chapter applies it to the military domain, where forced boundary crossings under orders are the primary mechanism of injury, and where the distinction between PTSD (d_1 trauma) and moral injury (d_7 damage) has immediate clinical and policy implications.
Chapter 22 (Jurisprudence) formalized law as boundary structure on the moral manifold. This chapter applies that formalization to International Humanitarian Law — the Geneva Conventions, the Additional Protocols, the Rome Statute — showing that IHL boundaries are domain-specific instantiations of the general boundary theory, with war crime boundaries as sacred-value boundaries (β = infinity).
Chapter 23 (Finance) identified dimensional collapse as the mechanism of the Flash Crash. This chapter identifies the same mechanism in autonomous weapons systems: AWS that optimize on d_1 alone can traverse paths that full-manifold agents would never take, with consequences measured in lives rather than dollars. The Flash Crash is the financial analog of the AWS targeting failure; both are consequences of algorithmic dimensional collapse.
Chapter 26 (AI Ethics) will address autonomous weapons from the perspective of AI alignment and control. This chapter addresses them from the perspective of IHL compliance and just war theory, establishing the military-ethical requirements that any AWS must satisfy — requirements that are, as shown in Theorem 28.6, beyond the current capabilities of AI systems.
Chapter 14 (Collective Agency) analyzed how individual agency is structured by institutional context. Military command structures are the most extreme case: the chain of command creates a hierarchy of agency in which individual soldiers' heuristic functions h(n) are partially overridden by command directives. Moral injury is the cost of this override when the command directive and the individual's h(n) diverge on boundary crossings.
Military ethics is the highest-stakes instantiation of the geometric framework. Decisions are irreversible (lethal force cannot be undone), consequences are maximal (measured in human lives), d_9 conditions are worst (the fog of war degrades every dimension), and the cost of dimensional collapse is catastrophic (war crimes, civilian massacres, post-conflict instability lasting generations). If the geometric framework can illuminate decision-making in this domain — where the pressure to collapse to d_1 is greatest and the cost of collapse is highest — it can illuminate decision-making anywhere.
Falsifiable Predictions
The framework generates five predictions that distinguish it from existing military ethics frameworks:
Prediction 1 (Civilian Casualty Correlation): Civilian casualty rates in a given conflict should correlate with d_9 degradation (intelligence quality), controlling for weapon type and conflict intensity. Conflicts with higher d_9 variance (urban insurgency, unmarked combatants) should have higher civilian casualty ratios than conflicts with lower d_9 variance (conventional warfare, uniformed combatants). Falsified if: civilian casualty ratios are independent of d_9 proxies.
Prediction 2 (Moral Injury Prevalence): Moral injury rates should be highest among soldiers with the strongest pre-deployment moral convictions (highest internalized boundary penalties), controlling for combat exposure. Falsified if: moral injury correlates only with combat intensity and not with pre-deployment moral conviction strength.
Prediction 3 (Post-Conflict Stability): Post-conflict stability (measured by years to renewed violence, democratic consolidation, or economic recovery) should correlate with the number of dimensions repaired (d_4, d_5, d_6, d_8), not merely the magnitude of repair on any single dimension. Falsified if: single-dimension repair (e.g., economic reconstruction alone) is equally effective as multi-dimensional repair.
Prediction 4 (AWS Targeting Errors): Autonomous weapons systems should produce targeting errors that are systematically different from human targeting errors — specifically, AWS errors should cluster on d_2 (legal status misclassification) and d_3 (proportionality failures) because these dimensions are absent from the AWS decision space, while human errors should cluster on d_9 (intelligence failures). Falsified if: AWS and human error distributions are indistinguishable.
28.14 Summary
This chapter has shown that the geometric ethics framework, when applied to military decision-making and the law of armed conflict, yields:
1. A formal construction of the military decision complex M_mil as a domain-specific instantiation of the moral manifold, with the nine dimensions calibrated for military contexts and the fog of war (d_9 degradation) identified as the distinguishing structural feature.
2. Jus ad bellum as manifold entry conditions: the decision to go to war is justified only when six dimensional conditions are simultaneously satisfied, and violation of multiple conditions (as in Iraq 2003) predicts catastrophic war-path failure.
3. Jus in bello as constrained A* pathfinding: the ethical cost of lawful combat (C_jib >= 0) is the price of maintaining boundary penalties on d_2, d_3, and d_4 even when unconstrained paths would be militarily cheaper.
4. Scalar proportionality failure: the IHL proportionality test collapses at least four dimensions (d_2, d_4, d_6, d_7) into a single civilian-harm scalar, destroying information irrecoverably. Geometric proportionality requires independent proportionality on each dimension.
5. The Doctrine of Double Effect as dimensional decomposition: DDE's "not a means" condition is formalized as the existence of an alternative path with comparable d_1 benefit and lower non-d_1 cost.
6. Autonomous weapons as dimensional collapse: AWS operating on d_1 and d_9 alone cannot satisfy IHL requirements that demand evaluation on d_2, d_3, d_7, and contextual d_9 — the military analog of the Flash Crash.
7. Moral injury as heuristic damage: combat moral injury occurs when forced boundary crossings degrade the soldier's heuristic function, explaining both moral numbing (heuristic collapse) and hypervigilance (heuristic hypersensitivity), and distinguishing moral injury (d_7) from PTSD (d_1).
8. Jus post bellum as manifold repair: post-conflict stability correlates with the number of dimensions repaired (d_4, d_5, d_6, d_8), not merely the magnitude of repair on any single dimension, explaining why multi-dimensional programs (Marshall Plan + NATO) outperform single-dimensional ones (Iraq reconstruction).
The cage is not a set of rules. It is an architecture — layers of mathematical structure, from quantum dynamics to democratic governance, compiled into running code.
The geometry is not decorative. It is load-bearing.