According to Chater and Christiansen (C&C), the uniqueness and complexity of human culture hinge on social tinkering, our ability to jointly create shared conventions that solve coordination problems. They proposed six interacting stages to explain how individual and social tinkering unfold, spanning non-social tinkering, social transmission, joint convention-making, language, the emergence of laws and institutions, and the development of science and technology. C&C focuses primarily on the later stages, detailing how social roles, norms, and institutions emerge from the local demands of coordination and cooperation, with language playing a critical role. However, the cognitive and computational mechanisms that drive the foundational stages – individual-directed innovation and social transmission – are left largely unexamined. We argue that their account of non-social tinkering (Stage 1) and social transmission (Stage 2) can be strengthened by integrating recent advances in cognitive science. To this end, we detail the mechanisms underlying directed innovation and highlight how strategic social transmission creates a synergy that accelerates collective innovation.

C&C rightly cast individual, non-social tinkering as directed rather than blind variation – a “sighted watchmaker” choosing among recombinations of existing solutions to solve problems. This view aligns with an emerging consensus (Morgan & Feldman, Reference Morgan and Feldman2024) but leaves open the question of how individuals mentally plan and direct their innovation attempts. We propose that the first step is goal-oriented planning. Faced with complex environments, humans typically begin by establishing concrete goals that define successful outcomes and guide behavior (Davidson et al., Reference Davidson, Todd, Togelius, Gureckis and Lake2025; Molinaro & Collins, Reference Molinaro and Collins2023). With sufficient environmental knowledge, individuals can decompose goals into plans and mentally simulate them to avoid costly trial and error (van Opheusden et al., Reference van Opheusden, Kuperwajs, Galbiati, Bnaya, Li and Ma2023). However, because planning is constrained by limited cognitive resources and environmental uncertainty (Callaway et al., Reference Callaway, van Opheusden, Gul, Das, Krueger, Griffiths and Lieder2022; Keramati et al., Reference Keramati, Smittenaar, Dolan and Dayan2016), individuals often rely on structured exploration to reduce uncertainty and map the environment. For example, when creating a stone tool, individuals might deliberately test unfamiliar materials to learn about their properties (Gershman, Reference Gershman2018; Schulz et al., Reference Schulz, Bhui, Love, Brier, Todd and Gershman2019) or seek states that maximize future possibilities, such as acquiring versatile tools before a specific need arises (Brändle et al., Reference Brändle, Stocks, Tenenbaum, Gershman and Schulz2023). Through such exploratory heuristics, people gather information, update internal models, and refine their actions toward solutions.

This process of iterative exploration is possible because environments are not random; they contain structural regularities that learners can extract and exploit. In doing so, individuals do not just learn isolated associations between actions and outcomes; they infer the underlying structure that produces them (Momennejad, Reference Momennejad2020). By extracting these structures, humans build robust internal models (semantic knowledge) that organize objects and abstract ideas based on their underlying properties, functions, and mutual relationships (Kumar, Reference Kumar2021). Our recent work shows that leveraging such structural knowledge greatly improves innovation efficiency by pruning nonsensical actions before effort is wasted testing them. When an individual discovers a successful mechanical or functional property during exploration, they abstract it into existing semantic networks and apply it to unseen situations. This provides a mechanistic account of how humans transform localized discoveries into general principles that can be reused and shared. Conversely, when such structural regularity is removed from the environment, innovation falls back to random exploration (Yaman et al., Reference Yaman, Tian and Lindström2026). Together, goal-oriented planning, directed exploration, and the capacity to leverage environmental structure give cognitive substance to C&C’s “sighted watchmaker,” specifying what directed means at the level of mechanism.

For Stage 2, C&C outline a “rich range of cognitive mechanisms that enable transmission,” from imitation to active demonstration. As with Stage 1, these mechanisms are left underspecified. We highlight three central, complementary aspects of social transmission: selectivity (who and when to copy), abstraction (what is transmitted), and synergy (how social information interacts with individual cognition). First, human social learning is fundamentally strategic rather than blind. Learners deploy sophisticated social learning mechanisms, dynamically determining when to copy and whom to copy based on different social cues, such as following the majority or recognized authority (Kendal et al., Reference Kendal, Boogert, Rendell, Laland, Webster and Jones2018; Schultner, Molleman, & Lindström, Reference Schultner, Molleman and Lindström2025a; Schultner, Molleman, & Lindström, Reference Schultner, Molleman and Lindström2025b). Likewise, innovators also actively and selectively share information (Mesoudi et al., Reference Mesoudi, Jiménez, Jensen and Chang2025). They use distinct strategies when engaging in teaching or demonstration, frequently tailoring their pedagogical approach to the needs of the receiver (Harootonian et al., Reference Harootonian, Griffiths, Niv and Ho2025; Sumers et al., Reference Sumers, Ho, Hawkins and Griffiths2023). Second, social transmission not only allows for direct replication of observed actions but also for the transmission of abstract mental states such as goals. Learners are sensitive to social influence when assessing value, strategically using the behavior of others to infer which goals are worth pursuing (Aarts, Gollwitzer, & Hassin, Reference Aarts, Gollwitzer and Hassin2004; Charpentier, Iigaya, & O’Doherty, Reference Charpentier, Iigaya and O’Doherty2020). For example, if an observer watches an innovator put significant effort into sharpening a blade, the observer does not merely copy the physical movements; they also infer the underlying utility of the goal itself. Consequently, the learners socially inherit the goal, applying the abstract principle of tool preparation to direct their own behavior even in entirely novel situations. Third, social transmission and individual cognition operate synergistically. Although individuals might face different environments, shared structural regularities allow observation to transmit more than just the solution to a problem; a learner might also infer what others are trying to achieve and how they think the world works (Jara-Ettinger, Reference Jara-Ettinger2019). Therefore, social observation acts as refined “training data” that dynamically updates the learner’s own understanding of the environment (Yaman et al., Reference Yaman, Tian and Lindström2026), thereby facilitating generalization to novel problems (Witt et al., Reference Witt, Toyokawa, Lala, Gaissmaier and Wu2024). These aspects recast Stage 2 as inferential rather than passive (Gweon, Reference Gweon2021), demonstrating that social transmission not only spreads discoveries but feeds back into the cognitive mechanisms that produced them.

In sum, we build on C&C’s framework by specifying the cognitive mechanisms underlying directed innovation and social transmission, which pave the way for broad social tinkering. However, how these micro-level processes scale up to the macro-level institutional complexity of C&C’s later stages remains open. Integrating agent-based modeling with large-scale online experiments (Schultner et al., Reference Schultner, Pärnamets, Yarmolenko and Lindström2026; Yaman et al., Reference Yaman, Tian and Lindström2026) could link individual behavioral signatures to the emergent population dynamics they generate, providing a unified platform for studying how C&C’s stages interact and scale.