According to bifocal stance theory (BST), how faithfully someone imitates depends on their goals. We copy actions faithfully to affiliate with others or to highlight our membership in a group (the “ritual stance”), but selectively copy only what is necessary to achieve instrumental goals (the “instrumental stance”). We agree that social learning can serve both affiliative and instrumental ends. However, we disagree that high-fidelity copying is necessarily triggered by non-instrumental goals. Humans can perform a variety of computations to learn from others, from faithfully copying others' actions to inferring the values and beliefs that caused them. Collectively, these computations trade off the cost of performing the computation against the flexibility and compositionality of its outputs. Understanding social learning through the lens of this trade-off can guide theorizing about when high-fidelity imitation and mentalizing may be deployed toward the same goal, and provides a mechanism by which causal insight can be bootstrapped from faithfully transmitted cultural practices.

A general principle of intelligent behavior is to use simple methods whenever possible and more complex strategies when necessary. An emerging framework has framed the arbitration between simple and complex strategies as a resource-rational trade-off (Lieder & Griffiths, Reference Lieder and Griffiths2020). Much like a thrifty shopper or an efficient long-distance runner, adaptive organisms should not only maximize rewards, but also account for the cognitive costs of different strategies. While resource-rational adaptations have been widely studied in the context of individual decision making (Kool, Gershman, & Cushman, Reference Kool, Gershman and Cushman2018; Shenhav et al., Reference Shenhav, Musslick, Lieder, Kool, Griffiths, Cohen and Botvinick2017), we propose that a similar trade-off exists in social learning (Wu, Vélez, & Cushman, Reference Wu, Vélez, Cushman, Dezza, Schulz and Wu2022).

To illustrate this trade-off, suppose you are watching your friend bake baguettes. As she pops the loaves into the oven, she pours boiling water into a skillet on the bottom rack. There are several ways that you could learn from this observation. First, you could directly imitate this action the next time you bake baguettes. This may quickly improve your technique, at the cost of flexibility: You may continue copying this action even when it is unnecessary or maladaptive. Alternatively, you could try to infer why she performed that action. For example, you could infer she used the boiling water to create steam, because steam gives bread a crunchy crust. Inferring the goals and beliefs driving your friend's actions is more costly than simply copying her, but it affords increased flexibility. The next time you bake bread yourself, you could use this insight to find alternative solutions to the same goal (e.g., by spraying water on the loaves) and to skip it when it is not needed (e.g., when baking soft, chewy breads).

What distinguishes these possibilities is not the observer's goal, but whether the benefits outweigh the computational costs. This trade-off helps identify cases where high-fidelity imitation is not only possible, but even preferable to mentalizing in instrumental contexts. If you are baking bread for the first time, or operating a complex and expensive MRI machine, you will likely maximize your rewards (and avoid catastrophic costs) by strictly following procedure.

Just as high-fidelity imitation can sometimes be beneficial to instrumental action, this computational trade-off can also guide theorizing about contexts where strategic innovation may be deployed in ritual. For example, medieval charms often required certain words to be invoked verbatim, but allowed ingredient substitutions (Luft, Reference Luft2020). One charm for curing rabid dogs involved buttering a slice of barley bread (“or if you cannot get that [type of bread], take another”) and writing ritual words on it before feeding it to the dog (Leach, Reference Leach2022). It is possible these deviations from ritual were guided by intuitive theories about which aspects were causally relevant – perhaps the charm depends on the words, but not on the type of bread on which they are written. Indeed, recent work suggests that modern adults' judgments about magic, such as the difficulty of a charm, are governed by intuitive theories of how the real world works (Lewry, Curtis, Vasilyeva, Xu, & Griffiths, Reference Lewry, Curtis, Vasilyeva, Xu and Griffiths2021; McCoy & Ullman, Reference McCoy and Ullman2019). While we agree that rituals serve an important affiliative function, these examples raise the possibility that rituals have their own causal logic and may allow a greater degree of behavioral flexibility than accounted for in BST.

So far, we have identified cases where observers may use high-fidelity imitation or mentalizing in the service of the same goal. This flexibility also provides a mechanism by which causal insights can be bootstrapped from faithfully transmitted cultural practices, thus blurring the lines between ritual and instrumental actions. Returning to the baking example above, you may assume that your friend's technique is the result of rational planning – that is, that she understands why each step in the recipe works and has arrived at her technique through deliberate utility maximization. But this is often not the case. The chemical reactions involved in bread-baking are sufficiently opaque that even a seasoned baker may faithfully copy a technique out of habit or conformity to cultural norms, without understanding why it works. If an observer were to then impute beliefs and rational planning to the demonstrator where there were none, they would be constructing a fiction – a “rationalization” of the demonstrator's behavior (Cushman, Reference Cushman2020).

This fiction can be quite useful. Technologies are often adopted and refined long before we discover why they work (Henrich, Reference Henrich2015). For example, the bark of the cinchona tree was used to treat malaria for centuries before its active ingredient, quinine, was first isolated and its pharmacological mechanism understood (Achan et al., Reference Achan, Talisuna, Erhart, Yeka, Tibenderana, Baliraine and D'Alessandro2011). Rationalization provides a means of representational exchange across different forms of social learning, enabling observers to extract generalizable, causal insights from cultural practices. This exchange may enable observers to innovate by design, by re-examining and refining long-held practices using their current internal models of the world.

In sum, beyond faithful copying, humans have access to a variety of cognitive capacities that enable us to learn from others. These capacities can be flexibly deployed and can support one another through representational exchange. Viewing social learning through the lens of computational trade-offs paints a more dynamic, agentic picture of how humans build culture.