Singh’s article focuses on “super attractors” – complex cultural inventions that appear in nearly all cultures. He argues that they result in part from subjective selection pressures. That is, people choose to do things that they judge as useful (either implicitly or explicitly). Because people across cultures share high-level goals, such as caring for offspring or acquiring food, and because of similarities across environments, they tend to select similar actions to achieve them.

In this commentary, we offer an extension to Singh’s model, informed by research and theories on infant cognitive development, as well as an idea first put forth by computational cognitive science, the blessing of abstraction (Goodman, Ullman, & Tenenbaum, Reference Goodman, Ullman and Tenenbaum2011). We suggest that considering domain-specific core knowledge systems – for objects, agents, numbers, space, and social relations – may help explain which cultural variants become super-attractors.

Core knowledge is theorized to be the set of innate (or very easily learned), abstract cognitive representations that humans possess. It is theorized to be domain-specific and ancient: a large body of work suggests that infants have knowledge about objects, number, space, agency (Carey, Reference Carey2009; Spelke, Reference Spelke2022), and potentially social relationships, social interactions, and communication (Kaufmann & Clément, Reference Kaufmann and Clément2014; Tatone, Reference Tatone2017; Thomas, Reference Thomas2025; Thomsen & Carey, Reference Thomsen and Carey2013; Samani & Thomas, under review).

Aspects of core knowledge that apply to Singh’s proposal. Core knowledge is theorized to be abstract. Adopting a term first proposed in computational cognitive science, Spelke has argued that core knowledge can be thought of as the primordial blessing of abstraction, allowing infants to make sense of entities the first time they encounter them, while also retaining flexibility to adapt to diverse cultures and environments (Spelke, Reference Spelke2022). While core knowledge is theorized to be both skeletal and a learning mechanism that enables infants to constrain what they attend to, it is also theorized to persist throughout life. Thus, there may be universal priors that lead to overlapping intuitions of adults, making some solutions more likely to be considered and easier to learn. This argument builds off many previous similar arguments about how innate mechanisms may shape cultural attractors (Sperber & Hirschfeld, Reference Sperber and Hirschfeld2004).

Second, these innate priors may also create shared knowledge, thereby shaping our intuitions about other people’s minds. This could mean that when people are choosing between actions, they may consider not only what is the best way to achieve a goal, but also whether others will agree. Intuitions about what falls in the domain of shared knowledge will, of course, also be shaped by culture. However, innate priors and our ability to reason about others’ knowledge may shape which practices become super attractors. In other words, while Singh’s article focuses on people across cultures judging actions similarly, people may also reason similarly about others’ interpretations of such actions.

This reasoning about shared knowledge may explain why many super attractors – like those Singh emphasizes – are public. Super attractors are not simply the outcome of private beliefs about utility; they also need to be learned. Thus, innate priors may shape which practices are intuitive to understand and therefore easy to learn, and which practices people intuit are effective; however, they may also shape intuitions about what other people will find intuitive and effective. When a shaman claims to intercede with spirits, she does so believing not just in her own efficacy, but also that others will share her intuitions about agency and causality, etc., both of which are rooted in core knowledge. Or, a shaman may believe that she can punish spirits, but also that others will share her intuition, which is why shamanism, religion, and institutions that facilitate punishment are public rather than private. Or take another example: while someone may believe that a particular social structure is the best way for their society to be structured, if they know that only a handful of others agree with them, they may not choose to devote their lives to a political revolution. Likewise, people may accept and justify practices that work against their own interests if they believe that a large proportion of the population endorses those practices (e.g., meritocracy beliefs in the United States; e.g., Benabou & Tirole, Reference Benabou and Tirole2006; Kogelmann, Reference Kogelmann2024).

Consider the case of thinking about others’ minds – naive psychology. Naive psychology has been formalized in computational models, in which people assume others act rationally, maximizing benefits and minimizing costs (Jara-Ettinger et al., Reference Jara-Ettinger, Gweon, Schulz and Tenenbaum2016; Liu et al., Reference Liu, Ullman, Tenenbaum and Spelke2017). This way of thinking about others’ actions aligns even with the expectations of young infants, suggesting it is part of core knowledge (Liu et al., Reference Liu, Ullman, Tenenbaum and Spelke2017; Liu, Brooks, & Spelke, Reference Liu, Brooks and Spelke2019). People may use these intuitions not just to understand other people’s behavior but also to plan their own actions, knowing that other people make sense of actions in a similar way (Ho, Saxe, & Cushman, Reference Ho, Saxe and Cushman2022). That is, if I call attention to how hard I worked on something, then I will intuit that others will intuit that I care about it. This can be expanded to include intuitions about other domains as well – people may have an understanding of how others think about agency, physics, etc.

All this being said, one question that applies to both our proposal and Singh’s is why we should expect this type of reasoning, which may be cognitively resource-intensive, rather than the sort of “model-free” learning he discusses (Boyd & Richerson, Reference Boyd and Richerson1988; Lake et al., Reference Lake, Ullman, Tenenbaum and Gershman2017). Singh proposes that individual evaluation, not blind copying, is central to cultural evolution. However, take the example of shamans, who might make different claims, and those whose intuitions resonate with others could gain prominence; the next generation of shamans could simply imitate the “celebrity” shaman without any intuitions about why they are doing so (Chudek et al., Reference Chudek, Heller, Birch and Henrich2012). In this model, there need not be any knowledge, either explicit or implicit, about either the “best course of action” or what others may think about them. One answer is that, just as core knowledge helps infants constrain the vast amount of information to which they have access and their hypothesis space, universal priors and the intuitions about shared knowledge that may arise from them constrain which actions are considered when people engage in model-based learning. Our proposal fits naturally into this framework: core knowledge systems provide a foundation for the structured “models” that make model-based reasoning computationally tractable and worthwhile in some cases. Studying infant cognition can help shed light on these starting states, while developmental cross-cultural studies can help us understand how cultural input shapes the processes by which these super attractors come to be (e.g., Astuti, Solomon, & Carey, Reference Astuti, Solomon and Carey2004).