According to Jagiello et al., because low-fidelity cultural learning relies on individual causal reasoning and trial-and-error, it is only likely to be successful in some circumstances. For example, when the goal-directed action is causally transparent and the end goal is resolvable. When such circumstances do not hold we should expect the adoption of the ritualistic stance and high-fidelity cultural learning. Despite being in broad agreement with Jagiello et al., we suggest an addition to bifocal stance theory. In particular, there are circumstances where low-fidelity learning will robustly fail despite the causal transparency and/or end-goal resolvability of the cultural trait involved. Such circumstances feature three key characteristics: specificity, riskiness, and complexity.

Specificity concerns the range of possible action sequences for achieving a desired goal. When this range is very narrow (i.e., the specificity is high), there is little room for learning error. Correspondingly, there is limited scope for the learner to tinker with the action sequence being modelled, as deviation from the appropriate sequence will not achieve the desired end. For example, preparing an edible flour from the sporocarps of the Australian nardoo fern, Marsilea drummondii, requires following a highly specific action sequence of washing and other activities (Earl & McCleary, Reference Earl and McCleary1994; see also Boyd, Reference Boyd2017, pp. 9–19). If this sequence is not followed, the consumption of the resultant flour over prolonged periods leads to thiaminase poisoning. In such contexts, we should expect to see selection for the ritualistic stance to reduce the possibility of error. In contrast, when there is a broad range of possible action sequences capable of producing a desired end goal (i.e., the specificity is low) there is room for the learner to engage in tinkering whilst still achieving success. For instance, an egg can be cracked on a sharp edge, using a knife, or using a custom-made tool. Unlike in the high-specificity case, here the instrumental stance is likely to be sufficient for robust transmission of the cultural trait.

The riskiness of an action sequence also influences whether employing the instrumental stance is a fruitful strategy (regardless of whether the specificity of the action sequence is high or low). Riskiness concerns the costs the learner will incur should they make a mistake (rather than the risk they will make the error in the first place). When the risk of error is high (e.g., where failure of the learner to perform the exact sequence of actions produced by the model results in death or serious injury) the trial-and-error style learning characteristic of low-fidelity cultural inheritance is expensive and inefficient. Conversely, action sequences with low-risk profiles offer a potentially fruitful opportunity for innovation. Again, food processing is illustrative. Bitter yam, Dioscorea dumetorum, is a staple carbohydrate amongst some Nigerian cultures which, like nardoo, is poisonous without careful processing by soaking, washing, and drying. Unlike nardoo, however, even a very small amount of poorly or unprocessed bitter yam can prove fatal (Omefe, Ajetunmobi, Onyema, & Atoyebi, Reference Omefe, Ajetunmobi, Onyema and Atoyebi2021). Whilst both action sequences are highly specific, bitter yam processing is more risky for the learner than nardoo processing because failure to carry it out properly is very likely to cause immediate death or severe illness.

For both risky and high-specificity action sequences the costs associated with error (either in failure alone, or death or serious injury) mean we should expect to see selective pressure for the ritualistic stance. As Jagiello et al. note, casual transparency and resolvability can mitigate this possibility by increasing the likelihood that the learner will rationally recreate the action sequence of the model with high accuracy. Whilst causal understanding goes some way to overcoming the challenges posed by riskiness and specificity, we are sceptical that even perfect causal understanding is entirely up to the task when the action sequence in question is complex in nature.

Action sequences vary in their complexity: the number and variety of steps they are comprised of. A simple action sequence may have only one step which is repeated (e.g., cracking an egg). A more complex action sequence may have multiple, heterogeneous steps (e.g., processing nardoo). For simple sequences the learner has relatively little to rehearse and memorise to successfully achieve the desired goal. Even if highly specific, with anything but a very high cost of error, simple sequences can be expected to be rapidly learned with low-fidelity learning and causal understanding. For complex action sequences, however, there are many steps to rehearse, order, and memorise to achieve the desired end goal. The cognitive load is much higher. Where an action sequence is complex and involves high risk and/or high specificity, this challenge is even greater, as trial-and-error or tinkering is costly and/or ineffective. Whilst causal understanding would be beneficial, it remains much harder for the learner to achieve the desired outcome with low-fidelity learning than in the low-complexity case because of the sheer number of possible “wrong moves.”

This sort of situation (high specificity or high risk, plus complexity) would not have been unusual in the history of our lineage. Even focusing solely on foods, there are many plants and animals (pufferfish, raw kidney beans, bitter almonds, cassava, to name a few), which require specific, complex processing to make them edible by humans. In our view, such situations would generate selection pressure for the ritualistic stance, regardless of causal understanding. This would lead to the cultural evolution of various forms of cultural scaffolding designed to invoke the ritualistic stance (and thus high-fidelity learning). This sort of scaffolding might include song, narrative, ceremony, and other typically ritualistic cultural practices we see used around the social transmission of technology. A work song, for example, which invokes the order of steps of a highly specific task makes it easier for a teacher to successfully transmit the correct action sequence to the novice. This is a testable prediction that, if found to be true, would provide further support for the bifocal stance theory.