An expectation is smaller than a moral verdict
An infant who watches a character protect someone may later look differently at how that character shares food. That can reveal an expectation carrying between situations. It does not, by itself, establish a moral verdict: looking, reaching, preferring and judging an obligation are different responses, and each needs its own experiment.
The distinction is easy to miss because adult language bundles several judgments together. We can expect someone to share fairly, want to approach them, feel sorry for them, or believe they have a duty to act. Those are separate conclusions. An infant’s stare or reach cannot establish all of them at once.
From protecting someone to dividing strawberries
A 2026 study in Communications Psychology tested infants aged 12 to 24 months in two experiments. Animated characters took roles involving aggression, victimization, protection, or avoidance. A later event involved distributing four strawberries between two new recipients, either equally or unequally.
This change of situation is central to the question. A character’s earlier role in conflict becomes possible evidence about its later behavior in sharing. Researchers are asking whether infants carry information about an agent beyond the original event.
The first experiment included 160 infants and the second 96. Infants looked longer when protectors shared unequally in both experiments. Victims showed that pattern in the first experiment. The initial pattern suggesting that aggressors were expected to share unequally did not repeat in the second. Bystanders generated no clear expectation.
Those differences are part of the result. “Babies identify good and bad people” removes the distinctions between roles, measures, and replication within the study. A more precise reading is that some expectations about a character appeared to carry between particular situations, with varying consistency.
Looking longer can indicate surprise
Babies cannot explain an unexpected outcome in a questionnaire. Looking-time experiments compare how long they attend to different events. In violation-of-expectation designs, longer attention may indicate that something conflicts with an expectation formed earlier.
The word “may” refers to the interpretation, not a refusal to use the measure. Researchers need the experimental comparisons and controls to rule out simpler explanations. One striking stare by itself tells us very little.
Think of a vending machine delivering a shoe instead of a snack. You might keep looking because the outcome is unexpected. That does not mean you prefer shoes, approve of the machine, or regard the shoe as morally wrong. Attention, surprise, and evaluation are different things.
The 2026 experiments counterbalanced test order and included a random-movement control. These design choices help assess competing accounts. They do not convert the stopwatch into a transcript of the infant’s thoughts.
Reaching asks a different question
Choice studies give an infant an opportunity to reach toward one of the characters after watching an interaction. This adds an active response, but reaching still requires interpretation. Approach can reflect preference without demonstrating a concept of obligation or goodness.
Earlier research on protective intervention reported a preference among six-month-olds for a character who intervened in aggression. Follow-up work distinguished responding to the protective outcome from responding to the intervener’s intentions, with sensitivity to intentions clearer among older infants.
For an adult comparison, someone might deliberately move you away from a falling object, while another person accidentally knocks you clear while chasing a dropped phone. The outcome is similar. What it suggests about the person differs. It is therefore useful to ask which part of an event a task allows the infant to use.
Victims and bystanders complicate the labels
A victim has had something happen to them; they have not necessarily demonstrated helpful behavior. Feeling concern for a distressed person does not automatically provide evidence that they will share fairly later.
That makes victim responses especially interesting. They force a distinction between wanting to comfort someone and predicting what kind of person they are. The 2026 study’s victim pattern was not equally established across both experiments, so it should not be treated as a settled general rule.
A bystander is ambiguous for a different reason. They may not notice the conflict, may be unable to intervene, or may have a reason to stay out. The visible lack of intervention cannot establish the motive.
Similarly, a statistically unclear infant response does not prove moral neutrality. It can reflect uncertainty, differences between participants, or limits of the task. “No clear expectation detected” preserves the finding without inventing a stronger psychological conclusion.
The famous helper preference needs its own caution
In the classic hill task, one character helps another climb while a different character pushes it down. Early findings suggested that infants preferred the helper. A large ManyBabies replication, published in Developmental Science, tested 1,018 infants across 37 laboratories, with 567 included in the analysis.
About 49 percent chose the helper in the social condition, which was not reliably different from chance. The result challenges treating that particular preference as a universally dependable finding.
It does not directly overturn the 2026 character-inference experiments. The infants were younger, and the task and outcome measure differed. Combining everything into one claim about “infant morality” hides exactly the differences needed to interpret the evidence.
The practical lesson is to keep the question attached to the task. Looking at unequal sharing after a protection event and choosing a helper after a hill-climbing display do not measure the same response.
Early does not necessarily mean unlearned
Dependent infants have reasons to learn who responds, obstructs, protects, or ignores. That makes social prediction potentially useful. It does not establish an evolved moral module or separate maturation from experience.
Everyday interaction offers information before an infant can articulate rules: someone cries and another person approaches; an object is taken or returned; a conflict starts and somebody changes the outcome. Learning can begin before speaking.
The evidence is most useful when understood as social understanding under construction. Keeping track of an agent, connecting events, forming an expectation, and choosing whom to approach are different pieces. Some tasks reveal them more consistently than others. A baby may be learning what to expect from another person without already deciding what that person morally deserves.
