Military triage algorithm discrimination review

Introduction

Military triage algorithm discrimination review concerns the legal examination of automated or AI-assisted systems used to prioritise wounded, injured, or medically vulnerable military personnel for treatment, evacuation, surgery, intensive care, or other medical resources. Such systems may process information including vital signs, injury severity, medical history, predicted survival, available resources, and operational circumstances.

Algorithmic triage can potentially improve speed and consistency in situations where medical resources are limited. However, it can also reproduce historical biases, rely on scientifically questionable variables, or systematically disadvantage particular categories of personnel. Because triage decisions can affect life and bodily integrity, discriminatory algorithmic decision-making raises serious constitutional, medical-ethical, and human-rights concerns.

There is currently no universally recognised legal doctrine specifically governing military triage algorithms. The applicable framework must therefore be developed from medical ethics, equality law, constitutional principles, military law, international humanitarian law, medical negligence principles, and emerging AI-governance standards.

Legal basis for military triage algorithms

Military medical services may legitimately use triage systems to allocate scarce resources during combat, disasters, mass-casualty incidents, or other emergencies. However, the existence of military necessity does not eliminate the requirement for rational and medically defensible criteria.

The first requirement is therefore lawful authority and legitimate medical purpose. An algorithm should be designed to prioritise patients according to medically relevant factors rather than characteristics unrelated to clinical need.

For example, injury severity and likelihood of benefiting from immediate treatment may be legitimate triage considerations. Race, religion, social status, or unrelated personal characteristics would generally raise serious discrimination concerns.

Equality and non-discrimination

Article 14 of the Indian Constitution provides the foundational equality principle. Government decision-making cannot be arbitrary, and classifications should have a rational relationship with legitimate objectives.

A military triage algorithm could create indirect discrimination even without explicitly using a protected characteristic. For example, an algorithm might rely on a variable that functions as a proxy for socioeconomic background, ethnicity, disability, or another protected characteristic.

The legal question is therefore not simply:

“Does the algorithm contain a discriminatory variable?”

It is also:

“Does the algorithm produce unjustified discriminatory outcomes?”

This distinction is essential because machine-learning systems can reproduce bias through seemingly neutral variables.

Medical ethics and triage

Triage is fundamentally a medical decision-making process. The central ethical principles include:

• medical need;
• urgency;
• likelihood of benefit;
• fairness;
• non-discrimination;
• preservation of life;
• responsible resource allocation; and
• professional medical judgment.

An algorithm should therefore assist medical professionals rather than automatically replace them in high-stakes circumstances.

Where an algorithm recommends deprioritising a wounded soldier, an appropriately qualified medical professional should be able to review the recommendation, particularly where the case involves unusual circumstances.

International humanitarian law

Military medical triage must comply with international humanitarian law.

The Geneva Conventions, particularly the rules concerning wounded and sick members of armed forces, establish important principles of humane medical treatment.

The principle of non-discriminatory medical care is particularly relevant. Under Geneva Convention I, Article 12, wounded and sick members of armed forces must be respected and protected and treated humanely.

Additional Protocol I also contains rules concerning the protection and care of the wounded and sick.

Accordingly, an algorithm should not deprioritise a wounded person simply because of characteristics unrelated to legitimate medical triage criteria.

Disability discrimination

Algorithms can produce particularly problematic results for people with disabilities or chronic medical conditions.

Suppose an algorithm predicts poorer survival because a patient has a pre-existing disability. If that disability becomes a reason for deprioritisation without adequate medical justification, the system could effectively discriminate against disabled personnel.

The Convention on the Rights of Persons with Disabilities (CRPD) provides important international principles concerning equality, non-discrimination, and respect for the inherent dignity of persons with disabilities.

Military medical systems should therefore carefully distinguish between:

• medically relevant prognosis;
• disability itself; and
• assumptions about the value or quality of a person's life.

A disability should not automatically be treated as evidence that treatment is less worthwhile.

Historical case law on medical discrimination

Although there is limited case law specifically concerning military AI triage, established discrimination cases provide useful principles.

In Griggs v. Duke Power Co. (1971), the U.S. Supreme Court recognised that apparently neutral criteria can produce unlawful discriminatory effects where they disproportionately disadvantage protected groups and lack sufficient justification.

Although Griggs concerned employment rather than medicine, the underlying principle is relevant to algorithmic governance: a system can be discriminatory even where discrimination is not explicit.

Similarly, Alexander v. Choate (1985) addressed disability discrimination and recognised the importance of equal access to medical benefits. The case illustrates why apparently neutral healthcare policies may require scrutiny when they disproportionately affect persons with disabilities.

Algorithmic transparency

A major challenge is the “black box” problem.

If a military triage algorithm recommends that one patient receive immediate treatment while another waits, medical personnel should be able to understand the principal factors influencing the recommendation.

A system that cannot be meaningfully audited creates accountability problems. The military should maintain documentation concerning:

• training data;
• clinical variables;
• model design;
• validation methodology;
• error rates;
• demographic performance;
• update procedures;
• override mechanisms; and
• responsibility for final decisions.

Commercial secrecy should not prevent appropriate medical, legal, or governmental oversight.

Human oversight

Human review is especially important in emergency medicine because algorithms can make mistakes.

A triage algorithm may fail when:

• the patient has an unusual injury;
• available medical data are incomplete;
• battlefield conditions differ from the training environment;
• a new disease or injury pattern emerges;
• the algorithm encounters a demographic group underrepresented in its training data.

A human clinician should therefore retain authority to override an algorithm where medical judgment indicates that the automated recommendation is inappropriate.

The person responsible for the final decision should remain identifiable.

Data quality and algorithmic bias

The quality of an algorithm depends heavily on its training data.

If historical military medical records contain unequal treatment or underrepresentation of particular groups, an algorithm trained on those records can reproduce those patterns.

A proper review should therefore examine:

1. representativeness of the training dataset;
2. missing-data patterns;
3. outcome definitions;
4. proxy variables;
5. subgroup error rates;
6. calibration;
7. false-positive and false-negative rates; and
8. performance under battlefield conditions.

Testing overall accuracy alone is insufficient. An algorithm can be highly accurate overall while performing poorly for a particular population.

Predictive survival and resource allocation

One of the most controversial issues is whether an algorithm should prioritise patients according to predicted survival.

Predictive models can potentially help allocate scarce medical resources, but they create difficult ethical questions.

A low predicted survival probability should not automatically mean that a person receives no treatment. Prognostic information should be interpreted alongside urgency, treatment availability, reversibility, and professional medical judgment.

The legal framework should prevent algorithms from silently converting statistical predictions into determinations about whose life is more valuable.

AI-assisted evacuation decisions

Military triage algorithms may also influence evacuation decisions. For example, a system could rank casualties according to predicted benefit from evacuation.

Such systems should be governed by clear medical criteria and should be regularly reviewed. Operational considerations may legitimately affect evacuation priorities, but they should not become a disguised mechanism for discriminatory treatment.

International humanitarian law remains applicable even when AI systems are involved.

Accountability for algorithmic errors

Determining responsibility is another central issue.

Potentially responsible actors may include:

• software developers;
• medical institutions;
• military procurement authorities;
• commanders;
• military doctors;
• system operators; and
• government departments.

Responsibility should not disappear merely because the decision was generated by an algorithm.

If a military hospital deploys an inadequately validated triage system, questions of institutional negligence may arise.

In India, Jacob Mathew v. State of Punjab (2005) provides an important general framework concerning medical negligence and professional standards. While the case predates modern AI systems, its principles concerning professional medical responsibility remain relevant.

Judicial review and administrative accountability

Where government-operated military medical systems produce discriminatory outcomes, judicial review may provide an accountability mechanism, subject to legitimate restrictions concerning military operations and national security.

Courts can potentially examine whether:

• the system had lawful authority;
• the criteria were rational;
• discriminatory effects were unjustified;
• medical standards were followed;
• adequate safeguards existed; and
• the decision-making process was arbitrary.

The constitutional principles of equality and life and personal liberty can become particularly important where medical decisions have life-threatening consequences.

Governance framework

A responsible military triage algorithm should include:

Pre-deployment validation: Independent clinical and technical testing before operational use.

Bias testing: Evaluation across relevant demographic and medical groups.

Human override: Qualified clinicians must be able to reject an algorithmic recommendation.

Auditability: Decisions should generate records showing the relevant factors.

Continuous monitoring: Performance should be reviewed after deployment.

Incident reporting: Serious errors and unexpected discriminatory outcomes should trigger investigation.

Data protection: Medical data should be securely stored and accessed only by authorised personnel.

Legal review: Major systems should undergo periodic examination for compliance with applicable law.

Conclusion

Military triage algorithms can improve the speed and consistency of medical decision-making during mass-casualty situations, but their use creates substantial discrimination and accountability risks. A system that appears neutral can nevertheless produce unequal outcomes through biased training data, inappropriate proxies, or poorly designed clinical variables.

The most relevant legal principles emerge from Geneva Convention I, Article 14 of the Indian Constitution, K.S. Puttaswamy v. Union of India, Griggs v. Duke Power Co., Alexander v. Choate, and Jacob Mathew v. State of Punjab.

The central rule should be that military triage algorithms may assist medical professionals but should not become unreviewable authorities over life-and-death medical priorities. Their criteria must be medically justified, scientifically validated, non-discriminatory, auditable, and subject to meaningful human oversight.

Ultimately, military necessity may justify rapid automated assistance, but it cannot justify discriminatory medical treatment. Human dignity, equal medical protection, clinical judgment, and accountability must remain at the centre of military algorithmic triage governance.