We are on board with the authors' broad enterprise: to identify, sort, and understand influences on social learning and their downstream effects on the stability of cultural traditions. Yet we have reservations about two key elements of the project: the role of the posited stances and the use of the “fidelity” concept. The two concerns are related. Reflecting on fidelity shows the stances provide only rough approximations of learning and its outcomes. This suggests the stances are not descriptions of cognitive processes, but rather, are rough-and-ready heuristic strategies for explaining downstream consequences of learning.

Consider first that acquiring ritual behaviour does not straightforwardly involve molecular (fine-grained behavioural) learning, or instrumental behaviour molar (goal-oriented) learning. Cases abound where learning goal-oriented behaviour requires attending to specific action sequences and the fine-grained details of those actions. The Levallois method of flint-knapping requires learning not just the core-trimming sequence through flake removal (molar), but also the specific kinematic profile for achieving the correct angle, velocity, and force of impact (molecular) (Roux & Bril, Reference Roux and Bril2005). Cases also abound where attentiveness to a ritual's goal is more important than its fine-grained profile: think of ritual handwashing, dressing-up for a wedding, or properly kneeling during salah. The lessons to take from these counterexamples are two-fold: the grain at which traits are learned varies with the behavioural specifics, irrespective of ritualistic or instrumental cues, and this undermines anything but a contextual, particularist evaluation of fidelity (Charbonneau & Bourrat, Reference Charbonneau and Bourrat2021).

Building on this, a second concern emerges out of the proposed links between the stances and long-term cultural stability. The authors suggest the two stances generate fidelity differences, understood as a propensity to produce faithful learning episodes (Charbonneau, Reference Charbonneau2020). Yet these propensity claims extrapolate from single episodes of learning without attending to the task structure or its effects on transmission. To see why this is a problem, consider two learners, each perfectly exemplifying one of the two stances: one ritual and one instrumental learner. Both learners observe the same model acting in order to complete a task. Some actions are causally transparent and instrumentally relevant, others opaque and conventional. Adopting the ritual stance, the first learner copies all actions faithfully. By contrast, the learner adopting the instrumental stance drops causally opaque actions, retaining only those transparently linked to task completion. To simplify matters, let us assume this leads the instrumental learner to acquire the optimal approach to the task.

After this first learning episode, the ritual learner will have acquired a trait with more fidelity than the instrumental one – the ritual learner's behaviour will be more similar to the model's than the instrumental learner. The situation changes, however, when we consider a second generation of stance-exemplary learners. In a lineage of ritual learners, the full sequence will be retained – including the instrumental and the conventional actions. In a lineage of instrumental learners, however, the full sequence, composed strictly of causally transparent action, will also be copied faithfully, as there are no further opaque or ineffective actions to be dropped. This can be projected into the future. While more content – more information – will be passed on in the ritual lineage, both lineages will be characterized by faithful transmission episodes. Though differing, both will be equally stable. What this toy example shows is that the authors' distinction between the (long-term) fidelity of the two stances extrapolates from an insufficiently characterized learning episode: an initial and uncertain encounter with a learning input composed of both causally transparent and causally opaque actions. This extrapolation misses important facts, notably, that the task structure can itself influence transmissibility, and in turn, have important effects on stability over multiple episodes (Claidière et al., Reference Claidière, Amedon, André, Kirby, Smith, Sperber and Fagot2018).

Third, this points to an important explanatory gap between accounts of learning and accounts of long-term fidelity. Bridging this gap requires attending not only to the learning of traits, but also to their expression (Morin, Reference Morin2015). If traditions are to remain stable over time, individuals must be motivated to produce behaviour similar to what they have learned.

In some situations – say, when they know the optimal trait – agents will express the same behaviour day in, day out. Otherwise, the stable expression of behaviour requires additional, often exogenous, machinery: social sanctioning, reputation loss, threats of supernatural punishment, and so on. Deviant behaviour needs to be recognizable, (potentially) punishable, and (definitely) correctable. When this social machinery is weak or absent – for instance, when agents find themselves in novel social contexts – opportunistic behaviours might creep in. Even longstanding rituals might give way to instrumental exploration.

The emerging science on the acquisition of behaviour and norms suggests that learners – especially children – learn not just how to produce an action, but also what not to do, which they can use to decide what they can get away with (Nichols, Reference Nichols2021). Work studying moral sense acquisition in children dovetails with cultural evolution, showing how children are sensitive to the in-group/out-group statuses of agents, authority-dependence of prescriptive norms, and influence this has on regulating their own behaviour and that of other agents (Ayars & Nichols, Reference Ayars and Nichols2020; Nichols, Reference Nichols2021; Schmidt, Rakoczy, & Tomasello, Reference Schmidt, Rakoczy and Tomasello2012). Explanations of the long-term stability of culture, in other words, must incorporate the wide range of information that agents acquire and weigh when deciding what behaviour to express (Buskell and Tennie, Reference Buskell and Tennieforthcoming; Morin, Reference Morin2015).

Are the “ritual” and “instrumental” stances descriptions of real cognitive patterns? Do these patterns correspond to differences in fidelity? The complex interplay between informational cues, trait history, and expression undermines easy inferences between learning and fidelity. Because of this we think that, rather than offering insight into the black box of cognition, the authors' stances are best understood as heuristic explanatory strategies; first-pass glosses that orient the investigator towards trait features influencing the learning, expression, and long-term fidelity of culture that can be fleshed-out with further investigations into the task, social context, and historical pressures. The stances are not descriptions of what goes on in the heads of learners, but useful guides for explaining the stability of traditions.