Vaesen speculates that the human capacity to learn novel tool use from observing goal-directed movements performed by others (Csibra & Gergely Reference Csibra and Gergely2007) is a hallmark of our uniqueness, and that it is based on “higher” sociocognitive skills. It has been proposed that such skills were supported by the ability to (1) decode kinematic information into causal relationships between a behavioural sequence and its result (Gergely Reference Gergely, Carruthers, Laurence and Stich2007); (2) interpret others' behaviors as rational (assuming that the most efficient observed action means are adopted to achieve a particular goal; Gergely & Csibra Reference Gergely and Csibra2003); and (3) accumulate a priori knowledge from past observations about agents' intentions and behaviours in order to predict future events (Chambon et al. Reference Chambon, Domenech, Pacherie, Koechlin, Baraduc and Farrer2011).

We agree with the author that the sophistication of such sociocognitive skills goes far beyond those of any other animals'. Yet, we believe that this sophistication could also be the result of simpler systems allocated to the detection of low-level, local sources of information, such as the manipulative properties of objects called “affordances.”

Affordances define relational properties that emerge from matching the perceived physical features of objects and the agent's biomechanical architecture, goals, plans, values, beliefs, and past experiences. We propose that affordances allow agents to delineate the number of candidate motor acts that could be performed on tools. We postulate that affordances constrain the number of possible solutions by generating biomechanical prior expectations in line with the bodily architecture of agents. These priors would bias individuals to act towards objects aiming at biomechanical optimization (Rosenbaum et al. Reference Rosenbaum, van Heugten and Caldwell1996; Weiss et al. Reference Weiss, Wark and Rosenbaum2007).

As the author rightly points out, compared with other animals', the many degrees of freedom characterizing human effectors and their striking motor control considerably enhance our ability to detect new affordances and new potential objects uses. All this contributes to increase the variety of the behavioural repertoire. Nonetheless, we are sceptical about the idea that the primary advantage such architectural properties bring for tool use acquisition is fine-grained social learning. Indeed, in many situations, detecting tools affordances allows learners to avoid such a high-level but costly strategy. Instead, this biomechanical uniqueness could increase the probability of individual innovation, particularly in situations where novel tools are physically unstructured and multi-purpose. For example, Acheulean stone tools are poorly structured and roughly symmetrical objects with a cutting edge. They do not offer affordances salient enough to constrain the number of candidate motor acts that could be performed on them. Sterelny (Reference Sterelny2003b) points out that the exact functions and uses of Acheulean stone tools, although they were the dominant element of human technology for more than a million years, remain a matter of debate. It is more plausible that our ancestors – who were predisposed to behavioural innovation thanks to their high biomechanical flexibility – progressively discovered not one or two, but a multitude of tasks that Acheulean stone tools could roughly carry out.

We argue that the evolution of the human technological environment favoured the utility of simpler systems such as affordances detection. This eases the negotiation of the highly demanding cognitive problems of tool use learning (Clark Reference Clark1997; Dennett Reference Dennett1995; Sterelny Reference Sterelny, Carruthers, Laurence and Stich2003a; Reference Sterelny2003b). Indeed, tools we interact with daily are designed for specific purposes. Affordances that are available through their complex physical attributes offer the chance for naive users to extract their functions at low cost (Dennett Reference Dennett and Woodfield1982; Reference Dennett1995; Gregory Reference Gregory1981; Norman Reference Norman1988). In our engineered environments, affordances play a crucial role in the acquisition of tool skills through individual trial and error as well as social learning. More specifically, we argue that perceiving affordances directly biases the understanding of tool behaviours performed by others, and consequently the extraction of related functional knowledge. The biomechanical priors that emerge from the perception of tools affordances constrain the number of candidate motor acts an individual could initiate. Similarly, they also tune the observer's prior expectations about which motor behaviors are most likely to be performed by others, enhancing the predictability and learnability of novel tool use. Learning about a novel tool from observing a demonstrator using it in a biomechanically “rational” way would be less costly than learning from a demonstrator that violates our expectations. That is, the convergence of the demonstrator's and observer's biomechanical expectations facilitates an efficient learning strategy, based on kinematics, rationality principle, or prior knowledge.

Taken together, these observations question the exact role of high-level, fine-grained social learning in the acquisition of new tool skills. Relevant to this is work addressing animal behavioural “traditions” – behavioural patterns that are relatively stable in groups and are at least partly maintained by some forms of social learning. These could result from constraints that limit the number of possible alternative behaviours, more than from the robustness of high-level social transmission mechanisms (Claidière & Sperber Reference Claidière and Sperber2010; Tennie et al. Reference Tennie, Hedwig, Call and Tomasello2008). Here, we posit that the crucial role affordances play in the acquisition of tool use strongly suggests that fine-grained social learning strategies, such as true imitation of observed action goals and means, are sometimes less important than previously assumed. In fact, affordances, together with ecological constraints and other products of epistemic engineering, could enhance the effectiveness of more frugal forms of socially directed learning (Acerbi et al. Reference Acerbi, Tennie and Nunn2011; Franz & Matthews Reference Franz and Matthews2010) such as emulation learning (i.e., the observer copies action goals performed by a demonstrator without considering action means) or even stimulus enhancement (i.e., when an individual directs its behaviour towards an object or a part of an object with which it saw another individual interact).