Note: The ethical judgments on this page refer exclusively to the action — never to the person who performs it or who came into existence through it. Cf. Note on Ethical Judgments.
Algorithmic embryo selection is the choice of which embryos are to be transferred and the discarding of the rest by an automated decision process. It stands at the end of a chain: an automated reproduction laboratory prepares gametes, fertilizes, cultures — and a model generates from image features a ranking by which it is decided which embryo is transferred and which is not.
The concept closes a gap that lay open for a long time. Preimplantation selection had been described, automated decision-making about persons had been described — but between the two there was no connection. It is precisely in this connection that the yield of the concept lies, and at the same time its most important result: automation adds no new moral quality to selection. It withdraws the personal performance from it. What is added is the responsibility gap, not the selection. The scandal does not begin with the software.
Ontological classification
- is a subclass of: eugenic selection, and hence a case of practical forgetfulness of the person
- is a subclass of: automated decision process (cf. AI-algorithmic arrangement)
- presupposes: algorithmic embryo assessment
- presupposes: in vitro fertilization
- produces: responsibility gap
- is carried out by: automated reproduction laboratory
- stands in connection with: preimplantation genetic diagnosis (PGD), supernumerary embryo
The twofold assignment is the actual point. Whoever treats algorithmic embryo selection merely as an AI application loses sight of the fact that it is selection. Whoever treats it merely as selection loses sight of the fact that no one is selecting any more. Both descriptions are incomplete, and the second is the more dangerous, because it creates the appearance that everything has stayed as it was.
What is actually evidenced
Honesty demands a distinction between what has been announced and what has been shown. In the only published controlled study of a fully automated laboratory system to date, the embryo assessment was still performed manually, according to the customary morphological scheme. What was automated were sperm preparation and parts of the microinjection; culture, assessment, identity verification and transfer were not. The algorithmic selection around which the public debate revolves was not in use in that study at all.
To this must be added that algorithmic embryo assessment as such has not produced proof of its efficacy. The largest randomized double-blind trial to date failed to reach the noninferiority margin against conventional morphological assessment; different commercial systems agree with one another less well than human assessors do among themselves; and repeated training of the same model on the same data produces rankings that diverge considerably from one another.
Two things follow for the assessment. First: the concept describes an announced, not an evidenced practice, and that has to be said. Second — and this matters more —: the efficacy argument has an expiry date. “It does not even work” is a strong objection today and possibly obsolete in ten years. An assessment that rests on this alone has no position left as soon as predictive power increases. The argument from personal ontology must carry independently of this — and it does.
Ethical assessment
The assessment here follows not an authority but the line of justification of the ontology of personhood itself. Ontological dignity is the inalienable, objective worth of the person and the sufficient ground of the personalistic norm: the person is to be affirmed for her own sake. This norm is violated in two ways — by instrumentalization, that is, the treatment of a person as a mere means, and by forgetfulness of the person, that is, the affirmation of a person not for her own sake but on account of her characteristics.
What stands objected to before any selection
Before anything the algorithm does, there stands a finding that does not concern it and without which the assessment would begin halfway through. Algorithmic selection constructively presupposes in vitro fertilization — it needs several simultaneously available embryos among which a choice can be made. In the ontology of personhood this is carried, by way of artificial fertilization, as an intrinsically evil act: generation is detached from the personal act of the spouses and handed over to the disposal of third parties.
The evaluation therefore reaches algorithmic selection along the path of presupposition and not by inheritance. In magisterial terms this is the statement of Donum vitae II, B, 5 on the so-called simple case: even a homologous fertilization in which no embryo is lost remains illicit, because it deprives generation of the dignity proper to it. Whoever begins only at the selection has already passed over the question of the procedure’s admissibility — and this independently of whether a human being or a machine does the choosing.
The selection itself
The embryo is a person in the First Dimension: spiritual substance in a body, with active potency toward unfolding. A ranking by image features affirms him not for his own sake but on account of a predicted probability of implantation. That is forgetfulness of the person in the precise sense: not malice, but a person dropping out of view behind a characteristic. That at the end of the ranking stands the discarding turns the forgetfulness into a disposal over existence.
What the assessment actually predicts must moreover be noted. Common systems rank by implantation or fetal heartbeat — not by live birth, not by the health of the child. The attribution is temporally limited and by no means essence-determining. Precisely in this lies the discrepancy that makes the procedure person-relevant: a non-essence-determining attribution becomes the ground of a decision about existence.
What automation adds
Not the selection, but the responsibility gap. Between physician, embryologist, clinic, manufacturer and developer, attributability of the decision is dispersed so widely that in the end no one has performed it — although it was performed.
The actual shape of this gap in regulated operation is not the absence of a responsible party but nominal oversight: a responsible party is named and signs off on what he cannot examine, for want of explainability of the system, for want of time, or for want of access to the grounds of the decision. The demand for genuine human control is then formally satisfied and substantively missed. That is the more difficult case, because it does not present itself as a breach of the rules.
Why the distinction between types of objection is decisive
Objections to a practice fall into two kinds, and the distinction is not classificatory but practical. An act-based objection is directed against the action itself, independently of circumstances, diligence and consequences; it is curable by no procedure and can only be contested or acknowledged. A consequence-based objection is directed against effects, risks or attendant circumstances and lapses when these lapse; it is curable by procedure.
Biased training data, lacking explainability, diffusion of responsibility, inadequate information of the patient — these are one and all consequence-based objections. The discarding by characteristics is not: it obtains independently of how reliably, transparently and supervisedly it takes place.
On this hangs the practically most consequential thesis of this page. A governance framework — oversight, validation, transparency, assignment of responsibility, data protection, information of the patient, equity of access — answers without exception consequence-based objections. Act-based objections remain untouched by it, not because it is badly made, but because it operates on a different level. Where an act-based objection obtains and only a governance framework is offered, the question of the admissibility of the practice is not answered but passed over. Precisely this displacement marks the present debate: it has shifted almost entirely onto automation and has largely set aside the question of the admissibility of the underlying procedure.
The strongest objection to this assessment
The most serious counter-objection concerns the word “eugenic”. Historical eugenics was state coercion against living human beings; embryo selection is a private decision between already produced embryos, taken by parents who want a child. No one is coerced, nothing is taken from anyone: the embryo not transferred is not made worse off than he would be without the procedure — without in vitro fertilization he would not exist at all. Whoever speaks of eugenics here, so the objection runs, uses a word of historical force for a matter that lacks the structure of coercion, and thereby spares himself the justification.
The objection strikes a true point, and it strikes it precisely: the analogy of coercion does not hold. But the assessment maintained here does not rest on it either. It rests not on a harm that would be inflicted on an individual, but on the manner of affirmation: a person is judged by a characteristic that decides about her existence. Whether anyone is coerced is irrelevant to this. To this is added what escapes the individual decision — the sum of free decisions produces an order in which certain modes of existence count as an avoidable error, and this order voices a judgment about already living human beings who exist in that way.
What remains open should be named as open: this argument presupposes that the embryo is a person. Against someone who denies this, it does not carry — he sees in the ranking the choice among clumps of cells and not the judging of persons. The justification of this presupposition does not lie on this page; it lies in the concept of the First Dimension. Algorithmic embryo selection adds nothing to the question of status and does not decide it. It only makes visible how much hangs on it.
See also
- Algorithmic Embryo Assessment — the presupposed attribution
- Automated Reproduction Laboratory
- Nominal Oversight, Responsibility Gap
- Act-Based Objection, Consequence-Based Objection, Governance Framework
- Eugenic Selection, Preimplantation Genetic Diagnosis (PGD)
- In Vitro Fertilization, Supernumerary Embryo, Embryo Transfer
- Embryo, First Dimension
- AI-Algorithmic Arrangement
- Forgetfulness of the Person, Instrumentalization, Personalistic Norm, Ontological Dignity
Sources: Generated by querying the ontology of personhood. Research as of 29 July 2026.
Further sources:
- Chávez-Badiola, A., Cohen, J. et al. (2026): Automated oocyte retrieval, denudation, sperm preparation, and ICSI in the IVF laboratory: a proof-of-concept study and report of the first live births. Human Reproduction 41(2): 214–230.
- Illingworth, P. J., Venetis, C., Gardner, D. K., Nelson, S. M. et al. (2024): Deep learning versus manual morphology-based embryo selection in IVF: a randomized, double-blind noninferiority trial. Nature Medicine 30(11): 3114–3120.
- Zaninovic, N., Sierra, J. T., Malmsten, J. E., Rosenwaks, Z. (2024): Embryo ranking agreement between embryologists and artificial intelligence algorithms. F&S Science 5(1): 50–57.
- Thirumalaraju, P., Kanakasabapathy, M. K. et al. (2026): Stability and reliability of artificial intelligence models in embryo selection for in vitro fertilization. Fertility and Sterility 125(2): 277–286.
- Koplin, J. J., Johnston, M., Webb, A. N. S., Whittaker, A., Mills, C. (2025): Ethics of artificial intelligence in embryo assessment: mapping the terrain. Human Reproduction 40(2): 179–185.
- Congregation for the Doctrine of the Faith (1987): Donum vitae. Instruction on Respect for Human Life in Its Origin, II,B,4–5.
- Congregation for the Doctrine of the Faith (2008): Dignitas personae. Instruction on Certain Bioethical Questions, nn. 14, 17, 22.
- Dicasteries for the Doctrine of the Faith and for Culture and Education (2025): Antiqua et nova. Note on the Relationship Between Artificial Intelligence and Human Intelligence, n. 74.