Genetic predictions and social choices: accuracy is not the only question
A discussion of the social-genomics debate in What We Inherit, based on the February 2026 Ars review, without clinical or current regulatory instructions.
AI-assisted, source-linked analysis. Product testing is only claimed when explicitly documented. How we work.
A genetic prediction can influence a choice even when its predictive power is limited. That is a question about the social use of information, not proof that a commercial test works or that a particular clinical decision is justified. This historical discussion follows Diana Gitig’s February 21, 2026 Ars Technica review of What We Inherit; it does not claim to be a direct review of the full book or a guide to medical testing.
The disagreement described by the review
The review presents a collaboration between bioethicist Daphne O. Martschenko and sociologist Sam Trejo. As reported, Martschenko emphasizes the history of genetic claims reinforcing inequality and questions whether more genetic information is needed to address known social problems. Trejo emphasizes the potential value of research and the difficulty of anticipating its benefits.
The review also discusses polygenic scores as probabilistic rather than deterministic predictions and describes concerns about unequal evidence and access. It does not turn those scores into individual destinies. This article offers no numerical estimate of clinical benefit, recommendation for embryo selection or account of current law.
The two positions supply a useful tension. One asks what information might reveal; the other asks what institutions and markets might do with it. Answering the first question does not automatically answer the second.
Why a weak signal can still have consequences
Information has consequences through the decisions attached to it. A number can be uncertain as a prediction while being treated as decisive by a buyer, institution or family. The possibility does not establish that every user behaves that way, but it explains why predictive accuracy alone is an incomplete ethical test.
Consider a hypothetical label attached to a future outcome. If people interpret it as a fixed identity rather than an uncertain estimate, their behavior may change even when the label adds little useful knowledge. This is an illustration of a mechanism, not a report that a particular population or company has already produced a measured social effect.
The relevant questions are therefore paired: how well does the estimate perform, and what action is being justified by it? A modest estimate used as one carefully bounded research variable is not the same use as a label presented as a person's destiny.
Better prediction would not remove every value judgment
Even an accurate estimate would leave questions about which outcomes people value, which trade-offs they accept and who has access to a service. Those are not answered merely by improving a statistical model.
A useful discussion can distinguish the validity of a measurement from the legitimacy of a decision rule. The first concerns whether the tool measures or predicts what it claims. The second concerns what someone proposes to do with that information and whose interests are affected. Confusing them allows a technical result to appear to settle a social disagreement it has not addressed.
This distinction also limits the opposite inference. Showing a harmful use would not by itself establish that every related research question is scientifically worthless. An ethical critique should identify the use and its consequences rather than rely on a blanket conclusion about all genetic inquiry.
A way to compare the arguments fairly
For a classroom or reading group, write each argument as a claim followed by the evidence it would need. Keep empirical predictions apart from normative commitments. If someone expects a technology to reduce inequality, ask what would demonstrate that outcome. If someone expects it to entrench inequality, ask what observations would support that concern and at what scale.
Next, identify where the disagreement actually lies. People can agree about uncertainty in a prediction but disagree about whether its possible benefits justify further development. They can also agree that research has value while disagreeing about a particular commercial application. A discussion is more productive when it does not force every position into simply being for or against science.
The exercise should preserve uncertainty. Do not invent evidence about consumer behavior or assume a future class of genetically optimized people already exists. A warning about a possible social mechanism remains a warning until its empirical claim is established.
What this discussion cannot decide
The review is not a clinical validation study, a statement of current professional guidance or a current regulatory map. Its historical examples cannot establish what a particular test is authorized to claim in a jurisdiction at the time a reader encounters it.
Nor can this discussion determine an individual’s health, identity or educational potential. Its useful contribution is to separate predictive evidence, the interpretation of a score and the choices institutions attach to it. Keeping those questions distinct makes the debate more demanding than either trusting a number because it looks scientific or dismissing every question that uses genetic data.