What is it about human cognition that has enabled us as a species to continue to sophisticate our tools and how effectively we go about using them to accomplish our everyday endeavours? Osiurak and Reynaud (O&R) contend that research into the cognitive origins of cumulative technological culture (CTC) – the continual improvement in human tool use over generations – has, to its serious detriment, been overwhelmingly focused on the notion that skilled tool practices are transmitted socially (through e.g., teaching and imitation). This focus on the social dimension of CTC has, they claim, lead researchers to underestimate the importance of uniquely human-reasoning capacities in explaining its cognitive genesis.
O&R appeal to a wide range of empirical data to support what they refer to as the technical-reasoning hypothesis: the contention that CTC “originates in non-social cognitive skills, namely technical-reasoning skills which enable humans to develop the technical potential necessary to constantly acquire and improve technical information” (abstract). Questions might reasonably be asked as to what O&R take technical information to be, but I want to focus here on their characterisation of technical-reasoning capacities. For them, technical reasoning is “the necessary cognitive structure that enables humans to constantly acquire and develop new [tool incorporating] techniques” (sect. 1, para. 4). They claim that it “starts by using mechanical knowledge in order to find a physical principle that is thought to be […] appropriate for solving the targeted physical problem (e.g., cutting) as well as the associated physical object properties (i.e., sharp, etc.)” (sect. 2.1, para. 4, item 3). An issue that immediately arises for the technical-reasoning hypothesis is what the nature of the thought and reasoning capacities involved in technical reasoning is supposed to be.
This commentary raises two interrelated theoretical concerns about O&R's technical-reasoning hypothesis. The first is that their notion of mechanical knowledge is underdefined and perhaps incoherent; the second is that their characterisation of technical reasoning as a kind of Bayesian learning potentially undermines their key claim that technical reasoning is uniquely human. Each of these worries pertains to a more fundamental concern: namely that O&R's account of the cognitive origins of CTC seem to rely on an overly-intellectualised characterisation of the cognitive activity involved in skilled tool practices.
For O&R, mechanistic knowledge is “a kind of magic” (sect. 2.3). It is, more specifically, a kind of non-declarative “knowledge about physical principles” (sect. 2.1, para. 4, item 2), which they identify with Stephen Pinker's (Reference Pinker2010) notion of know-how. Mechanical knowledge, for them, like “any form of knowledge, is a belief but not necessarily true” (sect. 2.3, para. 1). O&R readily allow that false beliefs – e.g., those about physical objects held by infants – can constitute knowledge providing they are “useful for inferring potential outcomes in the environment” (sect. 2.3, para 2.). Philosophers will be curious to know how it is possible to reduce knowledge to belief, given that they have been trying unsuccessfully for centuries to reduce knowledge to a kind of true belief. Setting the issue of whether knowledge is factive to one side, we should here recognise that philosophers of mind and cognitive scientists regularly posit procedural knowledge that does not reduce to knowledge of facts precisely because it is difficult to conceive of the know-how involved in paradigmatic instances of skilled sensorimotor action such as fluid tool use (e.g., competent stone knapping) as straightforwardly a matter of grasping certain facts about how to act (Fridland Reference Fridland2017). A possible challenge for O&R is to offer compelling reasons as to why the mechanistic knowledge involved in skilled tool use is best construed not as a dispositional competence but as a kind of conceptual, propositional knowledge, namely “a physical principle that is thought to be […] appropriate for solving the targeted physical problem” (sect 2.1, para. 4, item 3). As proponents of embodied cognition emphasise, that a system's behaviour is reliably attuned to certain physical principles does not entail that it knows or represents said principles (Colombetti Reference Colombetti2017).
Another potential problem emerges when we consider how, qua the technical-reasoning hypothesis, O&R take mechanical knowledge to be produced and honed. They claim that an agent's proficiency in using a tool increases through practice as she gleans progressively more accurate mechanical knowledge. Over time, technical reasoning leads to inaccurate beliefs being “invalidated through a kind of Bayesian learning” (sect. 2.3, para. 2). This form of Bayesian learning is presumably not deliberate or explicit, but O&R offer no details whatsoever about how technical reasoning is supposed to implement or conform to Bayesian norms. The notion of sub-personal cognitive activity somehow implementing Bayesian inference is increasingly popular in fields such as computational neuroscience (see Clark Reference Clark2013), but researchers usually claim that the human brain approximates, rather than genuinely implementing Bayesian inference. Some researchers question whether we can really describe cognitive activity approximating Bayes as an inferential procedure at all (rather than one that unfolds as though it followed certain norms see, e.g., Bowers & Davis Reference Bowers and Davis2012). But, there is a more pertinent point here for the technical-reasoning hypothesis.
Technical reasoning is characterised by O&R as an intelligent enterprise in virtue of its amounting to a form of Bayesian learning. Recent research into the free-energy principle, however, claims that the behaviours of non-human animals in constructing their environmental niche (and thereby avoiding states with high surprisal), will, on average and over time, approximate a kind of Bayesian belief optimisation (Friston Reference Friston2018). Indeed, even non-biological systems might reasonably be construed as approximating a form of Bayesian inference (Kirchhoff & Robertson Reference Kirchhoff and Robertson2018). It is therefore difficult to see how technical-reasoning capacities are, in virtue of their cohering to Bayesian norms at least, uniquely human. But, this means that technical reasoning might not be able to explain our uniquely human tool making/using capacities. A challenge going forward, then, is to better detail how the Bayesian credentials of technical reasoning facilitate a uniquely human capacity.
In sum, O&R should be praised for their robust defence of the claim that CTC is too focused on social cognition, but they face clear theoretical challenges in characterising technical reasoning.
What is it about human cognition that has enabled us as a species to continue to sophisticate our tools and how effectively we go about using them to accomplish our everyday endeavours? Osiurak and Reynaud (O&R) contend that research into the cognitive origins of cumulative technological culture (CTC) – the continual improvement in human tool use over generations – has, to its serious detriment, been overwhelmingly focused on the notion that skilled tool practices are transmitted socially (through e.g., teaching and imitation). This focus on the social dimension of CTC has, they claim, lead researchers to underestimate the importance of uniquely human-reasoning capacities in explaining its cognitive genesis.
O&R appeal to a wide range of empirical data to support what they refer to as the technical-reasoning hypothesis: the contention that CTC “originates in non-social cognitive skills, namely technical-reasoning skills which enable humans to develop the technical potential necessary to constantly acquire and improve technical information” (abstract). Questions might reasonably be asked as to what O&R take technical information to be, but I want to focus here on their characterisation of technical-reasoning capacities. For them, technical reasoning is “the necessary cognitive structure that enables humans to constantly acquire and develop new [tool incorporating] techniques” (sect. 1, para. 4). They claim that it “starts by using mechanical knowledge in order to find a physical principle that is thought to be […] appropriate for solving the targeted physical problem (e.g., cutting) as well as the associated physical object properties (i.e., sharp, etc.)” (sect. 2.1, para. 4, item 3). An issue that immediately arises for the technical-reasoning hypothesis is what the nature of the thought and reasoning capacities involved in technical reasoning is supposed to be.
This commentary raises two interrelated theoretical concerns about O&R's technical-reasoning hypothesis. The first is that their notion of mechanical knowledge is underdefined and perhaps incoherent; the second is that their characterisation of technical reasoning as a kind of Bayesian learning potentially undermines their key claim that technical reasoning is uniquely human. Each of these worries pertains to a more fundamental concern: namely that O&R's account of the cognitive origins of CTC seem to rely on an overly-intellectualised characterisation of the cognitive activity involved in skilled tool practices.
For O&R, mechanistic knowledge is “a kind of magic” (sect. 2.3). It is, more specifically, a kind of non-declarative “knowledge about physical principles” (sect. 2.1, para. 4, item 2), which they identify with Stephen Pinker's (Reference Pinker2010) notion of know-how. Mechanical knowledge, for them, like “any form of knowledge, is a belief but not necessarily true” (sect. 2.3, para. 1). O&R readily allow that false beliefs – e.g., those about physical objects held by infants – can constitute knowledge providing they are “useful for inferring potential outcomes in the environment” (sect. 2.3, para 2.). Philosophers will be curious to know how it is possible to reduce knowledge to belief, given that they have been trying unsuccessfully for centuries to reduce knowledge to a kind of true belief. Setting the issue of whether knowledge is factive to one side, we should here recognise that philosophers of mind and cognitive scientists regularly posit procedural knowledge that does not reduce to knowledge of facts precisely because it is difficult to conceive of the know-how involved in paradigmatic instances of skilled sensorimotor action such as fluid tool use (e.g., competent stone knapping) as straightforwardly a matter of grasping certain facts about how to act (Fridland Reference Fridland2017). A possible challenge for O&R is to offer compelling reasons as to why the mechanistic knowledge involved in skilled tool use is best construed not as a dispositional competence but as a kind of conceptual, propositional knowledge, namely “a physical principle that is thought to be […] appropriate for solving the targeted physical problem” (sect 2.1, para. 4, item 3). As proponents of embodied cognition emphasise, that a system's behaviour is reliably attuned to certain physical principles does not entail that it knows or represents said principles (Colombetti Reference Colombetti2017).
Another potential problem emerges when we consider how, qua the technical-reasoning hypothesis, O&R take mechanical knowledge to be produced and honed. They claim that an agent's proficiency in using a tool increases through practice as she gleans progressively more accurate mechanical knowledge. Over time, technical reasoning leads to inaccurate beliefs being “invalidated through a kind of Bayesian learning” (sect. 2.3, para. 2). This form of Bayesian learning is presumably not deliberate or explicit, but O&R offer no details whatsoever about how technical reasoning is supposed to implement or conform to Bayesian norms. The notion of sub-personal cognitive activity somehow implementing Bayesian inference is increasingly popular in fields such as computational neuroscience (see Clark Reference Clark2013), but researchers usually claim that the human brain approximates, rather than genuinely implementing Bayesian inference. Some researchers question whether we can really describe cognitive activity approximating Bayes as an inferential procedure at all (rather than one that unfolds as though it followed certain norms see, e.g., Bowers & Davis Reference Bowers and Davis2012). But, there is a more pertinent point here for the technical-reasoning hypothesis.
Technical reasoning is characterised by O&R as an intelligent enterprise in virtue of its amounting to a form of Bayesian learning. Recent research into the free-energy principle, however, claims that the behaviours of non-human animals in constructing their environmental niche (and thereby avoiding states with high surprisal), will, on average and over time, approximate a kind of Bayesian belief optimisation (Friston Reference Friston2018). Indeed, even non-biological systems might reasonably be construed as approximating a form of Bayesian inference (Kirchhoff & Robertson Reference Kirchhoff and Robertson2018). It is therefore difficult to see how technical-reasoning capacities are, in virtue of their cohering to Bayesian norms at least, uniquely human. But, this means that technical reasoning might not be able to explain our uniquely human tool making/using capacities. A challenge going forward, then, is to better detail how the Bayesian credentials of technical reasoning facilitate a uniquely human capacity.
In sum, O&R should be praised for their robust defence of the claim that CTC is too focused on social cognition, but they face clear theoretical challenges in characterising technical reasoning.
Financial support
This work was supported by the Australian Research Council Discovery Project “Minds in Skilled Performance” (DP170102987).
Conflict of interest
None.