A child watches one person try to help and fail, then another cause harm by accident. An adult sees the same events and asks what each person intended. An older adult might bring decades of experience to the judgment, yet still weigh the immediate outcome heavily in a particular task. These differences are not a story of morality appearing once in childhood and then fading. They are questions about how people learn, interpret and choose at different points in life.
Patricia Lockwood, Wouter van den Bos and Jean-Claude Dreher’s 2025 review, Moral Learning and Decision-Making Across the Lifespan, organizes decision-neuroscience research into four components: moral emotions, learning about others’ moral character, deciding for others, and moral influence. This guide explains the framework and its evidence limits. It treats age patterns as research findings tied to tasks and samples, never as a ranking of people’s moral worth.
What changes across a life?
Moral life involves more than knowing a rule. People feel concern and guilt, infer intentions, learn whom to trust, choose whether to help and respond to what others do. These capacities interact with attention, memory, social experience and opportunity. A lifespan account asks which processes are available at each age and how their relative influence changes.
The review describes self–other distinction as foundational in early years. A child needs to distinguish their own state from another person’s to understand whose need or feeling is at stake. Later judgment increasingly draws on beliefs and intentions, sometimes called theory of mind. Learning from outcomes also remains important. The review’s figure proposes shifts in the relative weight of model-based planning and mental-state reasoning from childhood through adulthood and older age.
These are not fixed stages that every person passes on a birthday. Studies use different tasks, cultures and measures. A young child can sometimes consider intention when a question and story make it understandable; an adult can still react strongly to a harmful outcome even when the harm was accidental. Context shapes what a participant can show in an experiment.
First component: moral emotions
Guilt, empathy, envy and responses to fairness can affect moral learning and choice. They are not interchangeable. Feeling distress when another person suffers differs from judging an unequal outcome, and feeling guilty after causing harm differs from anticipating blame. The review traces how such emotions are studied from infancy through older adulthood.
In infancy, researchers cannot simply ask, “Did you feel guilty?” They use looking time, preference, imitation and other behavioral indicators. Those measures can reveal discrimination among social events, but interpreting an infant’s inner moral experience requires caution. A longer look may indicate surprise rather than mature moral condemnation. The developmental significance lies in the emerging ability to register another person’s action and outcome, not in claiming an infant has an adult ethical theory.
In childhood, growing self–other distinction and social learning can support more differentiated responses. A child may learn that they caused another person’s distress, then respond to that person’s need. Yet a child can also imitate a rule to obtain approval. Researchers need to distinguish concern, compliance and fear of punishment instead of treating one act of sharing as proof of a stable moral trait.
Adults and older adults continue to experience moral emotions. The review does not present a simple peak-and-decline curve for empathy or guilt. Different emotions, relationships and tasks may show different age patterns. A caregiver’s sustained concern, for example, is difficult to capture with a brief monetary choice in a laboratory.
Second component: learning who can be trusted
Moral-character judgment asks what a person’s actions reveal about whether they are helpful, harmful or reliable. A single event can be ambiguous: did the person intend harm, act under pressure or misunderstand the situation? Over repeated interactions, a learner can update a belief. The review discusses how studies move from infant looking-time paradigms to repeated social decisions and computational models with older participants.
Intention and outcome need to be separated. Cushman and colleagues studied the development of intent-based judgment in children and distinguished judgments of moral wrongness from deserved punishment. A child may recognize that an accident was not intended while still attending to the damage it caused. The pattern is more informative than saying younger children care only about outcomes and adults care only about intentions.
Siegel and colleagues’ experiments on learning about good and bad agents found that beliefs about bad people could be more volatile or uncertain in their tasks. That may leave room for updating after a later good action. But a laboratory sequence of choices is not a license to advise someone to keep trusting a person who harms them. In real relationships, safety and boundaries matter alongside the cognitive mechanics of belief updating.
In adulthood, people may integrate inferred intent, observed behavior and potential future harm. In older age, task demands can alter how much intention information a person uses. One study of healthy older adults found continued intent-based judgment, challenging a blanket claim that ageing necessarily returns people to outcome-only reasoning. The proper conclusion is conditional: age-related changes in attention or mental-state inference are hypotheses to test, not a moral verdict about older people.
Third component: decisions that affect others
Deciding for others means choosing an action whose benefits or costs fall partly on someone else. This includes sharing resources, helping at a personal cost or refraining from harm. A person’s choice may depend on who the other person is, how the outcome is represented and what the chooser expects to happen next. The same person may behave differently toward a friend, stranger or group member.
The review uses reinforcement-learning language to describe how consequences can change later choices. In a model-free account, a person updates the expected value of an action after receiving feedback. In a model-based account, they simulate possible future outcomes using a representation of the situation. The terms describe computational approaches to behavior; they are not labels for “thoughtless” and “moral.” Both can support helpful or harmful choices depending on values, information and context.
Cutler and colleagues compared younger and older adults in a task where choices could earn rewards for the participant or another person. Older adults showed a decline in self-relevant learning rates while prosocial learning was preserved in the studied sample. This is a useful counterexample to the assumption that older people inevitably become less able to learn for others. It is not proof that all older adults are more altruistic or that the result transfers from a points task to every real-world decision.
A separate large cross-national study by Cutler and colleagues found greater prosocial behavior among older adults alongside stronger in-group preference. The pairing matters. A rise in helping within one circle does not automatically imply equal help for outsiders. Researchers should report who receives the benefit, not merely count benevolent acts.
Economic games offer control but simplify morality. A person deciding how to divide tokens knows the rules and stakes in a way they often do not in daily life. A real decision may involve uncertain effects, competing duties, continuing relationships and institutional constraints. A laboratory result can isolate a mechanism; it cannot be read as the full moral character of an age group.
Fourth component: influence and contagion
People learn from the choices and norms around them. Children observe models; adolescents may be especially attentive to peers; adults respond to social evidence and group expectations. The review calls this moral influence and contagion. It can spread help as well as harm, and the pathway is more complicated than simple copying.
For a child, seeing someone share may make sharing more likely, but the effect depends on whether the action was visible, rewarded and understood. An adolescent may adjust a judgment when peers disagree, yet peer influence can support principled action as well as conformity. An adult may defer to a group because of information, social pressure or a desire to avoid conflict. Experiments must distinguish these motives when possible.
Online environments make moral influence more visible and more difficult to interpret. A message shared many times can look like a strong social norm even when a small or automated group produced it. The review flags the need to study moral influence across ages in an information-rich era. That is a research agenda, not proof that one platform changes moral values in a particular direction.
Influence also raises an ethical question for educators and designers. If a system displays how “most people” acted, it may steer users before they have understood the issue. It can be useful to model cooperation; it can also exploit conformity. A responsible design should make the source of social information clear and leave room for disagreement.
Intention, outcome and the temptation to make a simple age curve
The review’s lifespan figure emphasizes that sensitivity to intentions and outcomes can shift with age. In some tasks, younger children and older adults may attend more to outcomes than adults in between. That is not the same as saying their moral reasoning is identical. A child is still developing language and mental-state concepts; an older adult brings accumulated experience but may face different task demands.
Question wording matters. Asking whether someone was “bad,” whether the act was “wrong” or how much punishment is deserved can produce different answers. A harmful accident may be judged wrong in outcome while the actor is judged less blameworthy. A researcher who collapses these judgments may manufacture an apparent contradiction.
The 2023 study on intent-based moral judgment in old age found evidence that healthy older adults still used intention information. Its authors discussed cognitive resources beyond the measures they tested as possible influences. This creates a useful boundary around any broad summary: the literature contains variation, and simple developmental decline is not established by one comparison.
Similarly, a study of public beliefs by Mastroianni and Gilbert found widespread perceptions that morality is declining over time. That paper concerns beliefs about social change, not the actual lifespan mechanisms in the Lockwood review. It is a reminder that an intuitive “people are less moral now” story needs evidence before it becomes an explanation for age differences.
What a computational model can and cannot do
A model can estimate how strongly participants update after good or bad outcomes, how they trade a reward for themselves against a cost to another person or whether a planned consequence seems to influence choice. Those parameters make theories testable. If two age groups choose differently, a model can ask whether the difference is better explained by learning rate, valuation, attention or another candidate mechanism.
But a parameter is an interpretation of behavior under a model, not a direct reading of the mind. Different models may fit similar choices. If a task has too few trials or its rewards are misunderstood, a fitted learning rate can be unstable. The review calls for robust and reliable parameters when comparing ages and points to the need for stronger designs across the lifespan.
A model also requires a moral reference frame. Is the choice for self, for another person, for both, or for a larger group? Is the “other” known, trusted or socially distant? A parameter estimated in a self-reward task should not automatically be called a measure of empathy. The conceptual question must lead the computation.
Finally, brain correlates are not moral scores. A pattern of neural activity can help test a mechanism, but no scan tells us whether a person is good. The review combines neuroscience with behavior to generate hypotheses; it does not provide a device for ranking citizens.
Implications for teaching and intergenerational life
A teacher can use stories that separately identify what a character intended, what happened and whom it affected. Asking children to explain each part may help reveal how they understand the case. That is an educational application inspired by developmental research, not a validated curriculum outcome from the review.
Intergenerational groups can also avoid assuming that one age has all the moral insight. Younger participants may notice a social norm an older participant takes for granted. Older participants may bring experience with long-term consequences. Good discussion can make these different forms of evidence explicit. It should still permit participants to challenge a norm, including an unfair norm shared by the group.
For designers of AI-mediated decisions, the lesson is to keep human stakes visible. A system that optimizes a reward score may neglect intention, uncertainty or who bears the harm. Testing across age groups can reveal accessibility and interpretation differences. It does not imply that an AI can replace moral responsibility with a fitted computational parameter.
What the evidence does not show
The 2025 review does not establish that moral worth rises or falls with age. It does not show that a single brain region causes altruism, that an economic game captures everyday character or that a child’s response in one task predicts their adult morality. Its four components are a research framework rather than an exhaustive list of ethical capacities.
Age differences in cross-sectional studies can also reflect cohort, culture, education and task familiarity. A group of older adults studied today grew up under different conditions from a group of children studied today. To describe change within a life, longitudinal evidence is especially valuable. Researchers need broader populations and real-world outcomes alongside controlled experiments.
Nor does the review settle normative ethics. Neuroscience can describe how people learn and choose; it cannot alone decide what they ought to do. A mechanism may help explain a judgment without justifying it. That boundary keeps a scientific guide from silently becoming a moral command.
Return to the accidental harm
The child, adult and older adult in the opening scene may attend to different details, ask different questions and draw on different experiences. Each may also change their judgment when new information appears. A good study makes the intention, outcome and social context visible enough to learn which process shaped the response.
The lifespan perspective is valuable because it refuses a single snapshot of morality. It asks how emotions, character learning, decisions for others and social influence develop and interact over time. Its best use is to make moral research more precise, while leaving every person’s dignity outside the reach of an age stereotype.
Research references
- Lockwood, P. L., van den Bos, W., & Dreher, J.-C. (2025). Moral learning and decision-making across the lifespan. Annual Review of Psychology, 76, 475–500. View source
- Cushman, F., Sheketoff, R., Wharton, S., & Carey, S. (2013). The development of intent-based moral judgment. Cognition, 127(1), 6–21. View source
- Siegel, J. Z., Mathys, C., Rutledge, R. B., & Crockett, M. J. (2018). Beliefs about bad people are volatile. Nature Human Behaviour. View source
- Cutler, J., et al. (2021). Ageing is associated with disrupted reinforcement learning whilst learning to help others is preserved. Nature Communications, 12, 4440. View source
- Cutler, J., et al. (2021). Older adults across the globe exhibit increased prosocial behavior but also greater in-group preferences. Nature Aging, 1, 880–888. View source
- Margoni, F., Cho, I., & Gutchess, A. (2023). Intent-based moral judgment in old age. Journals of Gerontology: Series B, 78(7), 1136–1141. View source
- Mastroianni, A. M., & Gilbert, D. T. (2023). The illusion of moral decline. Nature, 618, 782–789. View source