The ethics and limits of AI in homeopathy

Artificial intelligence now sits inside most clinical software, and homeopathic repertorisation packages are no exception. The practical question is narrow: when you turn on an AI feature, what belongs inside a homeopathic workflow, and what should you refuse?

Where AI helps and where it fails inside a homeopathic workflow

Three patterns recur. AI is genuinely useful where the task is information-retrieval over a corpus the model has actually indexed and where the retrieval surface preserves the source citation. AI fails where the task requires judgement that depends on the patient being in the room — affect, narrative, the practitioner's own clinical pattern-library built over a career. The most dangerous failure mode is not refusal but confident output that looks like retrieval and is in fact fabrication. The working rule is therefore not "use AI" or "do not use AI" but use AI only where the cost of a fabricated citation is contained by a downstream verification step the practitioner actually performs. A wider treatment of each capability sits under AI in homeopathy and AI case analysis.

AI tooling inside a living clinical tradition

Homeopathy is a 200-year practitioner tradition with its own observational base and clinical norms. The canonical corpus — Hahnemann's Organon, Kent's Repertory and Lectures, Hering's Guiding Symptoms, Boenninghausen's Therapeutic Pocket Book, Boger's Synoptic Key, Vithoulkas's Science of Homeopathy, and Sankaran's sensation-method volumes — records case-based findings built up across generations of clinical practice. An AI tool is only as reliable as the corpus it has actually indexed. A model trained on an incomplete or out-of-date version of these primary texts will produce confident-sounding output that diverges from the sources practitioners rely on.

The practical consequence is that AI tooling has to be evaluated against the primary texts it claims to cover. A model blind to Boenninghausen's Therapeutic Pocket Book cannot reliably surface Boenninghausen rubrics. A model whose materia medica corpus stops at a 2019 snapshot will miss remedy refinements that entered the literature since. Neither gap is obvious from the output surface, which is why source-checking against the primary texts is not optional.

Four ethical pressure points

Consent

Patients should be told when AI is involved in their case work. Consent extends to the tools used to reach a recommendation, not only to the recommendation itself. A one-sentence disclosure during intake — "I use a repertorisation tool with an AI component; the prescription is mine" — satisfies the consent point without disrupting the consultation.

Fabrication and source-checking

Large language models fabricate citations. A practitioner who quotes a remedy attribution from a model output without checking the source has just put an unsourced therapeutic claim into the record under their own clinical authority. The fix is mechanical, not philosophical: every model-suggested attribution is verified against the primary text — Hahnemann's Organon, Kent's Repertory, the listed proving, or Hering's Guiding Symptoms — before it enters a case record or a patient explanation.

Deskilling

Repertorisation is a clinical skill, and skills atrophy when delegated. If the model does the rubric selection, the practitioner does not learn the materia medica; over a five-year arc the practitioner's case-taking degrades even if the model's output stays constant. The risk is well documented in the medical literature on AI-assisted radiology and applies with the same shape in homeopathic case work. The guardrail is to position AI as a second reader after your own repertorisation, never as a first reader that anchors the prescription.

Marketing-grade overclaim

Vendor marketing around AI in homeopathy has used language that is a publish-blocker: AI that "diagnoses", AI that "predicts the simillimum", AI that "replaces the practitioner". Such language overstates what any current model can deliver within the tradition's own standards of individualised case-taking. Refuse to use software whose marketing copy makes claims its feature set does not support.

Five guardrails

1. Source-check every citation. A model-supplied citation is a candidate, not a fact, until the source is opened and read in the primary text.
2. Disclose AI use to the patient. Short, factual, recorded in the case file alongside the prescription.
3. Know the tradition first. Case notes grounded in the canon reflect the practitioner's clinical judgement, not the model's output. The model surfaces candidates; the practitioner evaluates them.
4. Position AI as a junior associate, not an attending. The practitioner signs every prescription; the model does not. Any vendor that suggests otherwise has crossed the line.
5. Re-audit the model every quarter. Models change without notice — corpora are refreshed, retrieval layers are swapped, output styles drift. A guardrail that held in February may not hold in May.

Verdict

AI in homeopathy is acceptable as a research assistant, a structured-intake support, and a second reader on repertorisation. It is not acceptable as a prescriber or as a substitute for materia medica study. Tools that log their sources, refuse to issue prescriptions, and respect the tradition's own standards of individualised case-taking belong in a practice; tools that do the opposite do not. Practitioners who want to audit a model-backed repertory under exactly these guardrails can run cases through Similia and judge the output against their own case files.

References

Hahnemann, S. Organon of the Medical Art, sixth edition. Kent, J. T. Repertory of the Homoeopathic Materia Medica; Lectures on Homeopathic Materia Medica. Hering, C. Guiding Symptoms of Our Materia Medica. von Boenninghausen, C. Therapeutic Pocket Book. Boger, C. M. Synoptic Key of the Materia Medica. Vithoulkas, G. The Science of Homeopathy. Sankaran, R. The Substance of Homeopathy and sensation-method volumes. World Health Organization (2021) Ethics and governance of artificial intelligence for health: WHO guidance, Geneva. https://www.who.int/publications/i/item/9789240029200 General Medical Council, United Kingdom (2024) Good medical practice and accompanying guidance on the use of artificial intelligence in patient care. https://www.gmc-uk.org/professional-standards/professional-standards-for-doctors/good-medical-practice Geis, J. R. et al. (2019) "Ethics of artificial intelligence in radiology: summary of the joint European and North American multisociety statement," Radiology 293:2. https://pubs.rsna.org/doi/10.1148/radiol.2019191586