Stibbard-Hawkes provides an important warning against naively deducing from an absence of (positive) evidence E for cognitive trait C, the total absence of C in a given population. He is right in criticizing any proposal that deductively ties absence of C to absence of E: Logically, even if we could reliably establish that E⇒C, that would not entail that ¬E⇒¬C, and he provides compelling evidence against deducing lack of cognitive modernity for human populations that do not produce enduring evidence of it.
I am very sympathetic to the article's main empirical contribution, and I understand it as an important cautionary note. However, I think that its most general conclusions should be nuanced, in particular in their evolutionary and cross-species comparative prisms: One thing is to assume universality of capacities within a species, and a very different one to do it across species. As a matter of fact, it seems to me that the adoption of such an overarching prior as the one proposed in the article (“until proven otherwise, all members of at least our genus had comparable capacities” [target article, sect. 16, para. 8]) may amount to a scientific dead end.
To begin with, Popperian conjecture and refutation can only be effective if the conjecture can be refuted with empirical observations, but this will hardly be the case with the hypothesis proposed; no artifact finding in an archaeological setting will show that species S did not possess C. Thus, I believe that it is safer and epistemologically sounder to start with the opposite hypothesis (¬C).
Besides, Bayesian epistemology can also help vindicating the possibility of substantive inductive inferences from settings where evidence is absent (Howson & Urbach, Reference Howson and Urbach1989; Oaksford & Hahn, Reference Oaksford and Hahn2004; Stephens, Reference Stephens2011) – a very common situation in archaeology (Thomas & Darvill, Reference Thomas and Darvill2022; Wallach, Reference Wallach2019):
1. In order to infer any cognitive capacity C from evidence E, there has to be a causal link between E and C (say, we theorize that it is C that enables E), and therefore the probability of finding E given C must be larger than that of finding E in the absence of C: Pr(E|C) > Pr(E|¬C). Thus, in Bayesian jargon, finding E in a given archaeological setting confirms (Popperian corroborates) the plausibility of attribution of C to S. But in logical consequence it also follows that Pr(¬E|¬C) > Pr(¬E|C), and thus ¬E – given certain conditions (see below) – could be taken as confirming (positively updating our credence that) ¬C.
2. The conditions for an inference from absent evidence depend on a number of factors. One is the causal link between C and E (see points 3 and 4 below). But as the work of Stibbard-Hawkes rightly points out there are also factors like taphonomy and particularities of cultural practices that may cause ¬E. To these we should also add our own research (whether we looked deep enough, etc.). If factoring in taphonomy, and so on, we consider that Pr(¬E|¬C) > Pr(¬E|C), and our research efforts were sufficiently accurate but still we find out ¬E, then we may be justified in positively updating our belief that ¬C. Even more so if our research on S uncovered analogous sets of evidences E' (say, made with the same materials of E, but lacking the characteristics that would make them E), so that we may be justified in inferring that had S produced E, we would have discovered it.
3. We are dealing with cognitive and linguistic archaeology, and the purported evidence is not direct evidence of C, but evidence E that could signal C, modulo a range of assumptions about the link between E and C. It goes without saying, the devil is in the details, and the nature of C, E, and of their link ought to be formulated in formally precise (and therefore consequential) propositions, modeling the strong generative procedures to go from C to the set of possible E. For instance, a notion such as “symbolism” is too vague (as many species ostensibly display some capacity for “symbolic thought” – see, e.g., Gallistel, Reference Gallistel, Scarborough and Sternberg1998, Reference Gallistel2011), likewise for any unqualified communicative view of language (as communication is present widely in the animal kingdom [Hauser, Reference Hauser1996; Hauser, Chomsky, & Fitch, Reference Hauser, Chomsky and Fitch2002], but also among fungi and plants [e.g., Boyno & Demir, Reference Boyno and Demir2022]).
4. In this respect, the analysis of the cognitive-computational power allegedly required for particular behaviors (e.g., the generation of recursive linguistic structures, knotting, or complex-patterned art), abstractly analyzed as pertaining to different levels of the Chomsky hierarchy may be a route to explore (cf. Camps & Uriagereka, Reference Camps, Uriagereka, Rosselló and Martín2006). If – and this is a big “if” – we could reliably ascribe particular behaviors/productions to a well-defined cognitive capacity (say, the capacity to compute context-sensitive grammars [producing Type-1 formal languages]), then evidence E of a particular behavior by species S could be taken as confirming the possession of C by S. Consequently, and given the necessary conditions, ¬E could also be justifiably taken as (probabilistically) confirming ¬C.
To conclude, a certain homogeneity of traits is a core assumption for intra-species groupings; we can expect C to be “universal” across the individuals conforming a species (e.g., trichromatic vision, or the capacity for context-sensitive symbolic computations in humans). But in the absence of the relevant evidence, this cannot be extended to inter-species or inter-taxa groupings. In cross-species comparison I think that it is safer to depart from the hypothesis that, say, Homo antecessor or zebra finches lack C (especially when C is a rather unique trait), and then look for evidence of C (cf. i.a. Fitch & Hauser, Reference Fitch and Hauser2004; Beckers, Bolhuis, Okanoya, & Berwick, Reference Beckers, Bolhuis, Okanoya and Berwick2012) than just be satisfied by assuming that they have C and stop further research. Assuming as Stibbard-Hawkes proposes a substantive positive prior could not, by essence, be negatively updated. That is, if we depart from the hypothesis that extinct species S did possess cognitive property C, then there can be no positive evidence E that will disproof (negatively update) our hypothesis.
In sum: We should lower our priors that ¬C, but not invert them to C.
You have
Access
Stibbard-Hawkes provides an important warning against naively deducing from an absence of (positive) evidence E for cognitive trait C, the total absence of C in a given population. He is right in criticizing any proposal that deductively ties absence of C to absence of E: Logically, even if we could reliably establish that E⇒C, that would not entail that ¬E⇒¬C, and he provides compelling evidence against deducing lack of cognitive modernity for human populations that do not produce enduring evidence of it.
I am very sympathetic to the article's main empirical contribution, and I understand it as an important cautionary note. However, I think that its most general conclusions should be nuanced, in particular in their evolutionary and cross-species comparative prisms: One thing is to assume universality of capacities within a species, and a very different one to do it across species. As a matter of fact, it seems to me that the adoption of such an overarching prior as the one proposed in the article (“until proven otherwise, all members of at least our genus had comparable capacities” [target article, sect. 16, para. 8]) may amount to a scientific dead end.
To begin with, Popperian conjecture and refutation can only be effective if the conjecture can be refuted with empirical observations, but this will hardly be the case with the hypothesis proposed; no artifact finding in an archaeological setting will show that species S did not possess C. Thus, I believe that it is safer and epistemologically sounder to start with the opposite hypothesis (¬C).
Besides, Bayesian epistemology can also help vindicating the possibility of substantive inductive inferences from settings where evidence is absent (Howson & Urbach, Reference Howson and Urbach1989; Oaksford & Hahn, Reference Oaksford and Hahn2004; Stephens, Reference Stephens2011) – a very common situation in archaeology (Thomas & Darvill, Reference Thomas and Darvill2022; Wallach, Reference Wallach2019):
1. In order to infer any cognitive capacity C from evidence E, there has to be a causal link between E and C (say, we theorize that it is C that enables E), and therefore the probability of finding E given C must be larger than that of finding E in the absence of C: Pr(E|C) > Pr(E|¬C). Thus, in Bayesian jargon, finding E in a given archaeological setting confirms (Popperian corroborates) the plausibility of attribution of C to S. But in logical consequence it also follows that Pr(¬E|¬C) > Pr(¬E|C), and thus ¬E – given certain conditions (see below) – could be taken as confirming (positively updating our credence that) ¬C.
2. The conditions for an inference from absent evidence depend on a number of factors. One is the causal link between C and E (see points 3 and 4 below). But as the work of Stibbard-Hawkes rightly points out there are also factors like taphonomy and particularities of cultural practices that may cause ¬E. To these we should also add our own research (whether we looked deep enough, etc.). If factoring in taphonomy, and so on, we consider that Pr(¬E|¬C) > Pr(¬E|C), and our research efforts were sufficiently accurate but still we find out ¬E, then we may be justified in positively updating our belief that ¬C. Even more so if our research on S uncovered analogous sets of evidences E' (say, made with the same materials of E, but lacking the characteristics that would make them E), so that we may be justified in inferring that had S produced E, we would have discovered it.
3. We are dealing with cognitive and linguistic archaeology, and the purported evidence is not direct evidence of C, but evidence E that could signal C, modulo a range of assumptions about the link between E and C. It goes without saying, the devil is in the details, and the nature of C, E, and of their link ought to be formulated in formally precise (and therefore consequential) propositions, modeling the strong generative procedures to go from C to the set of possible E. For instance, a notion such as “symbolism” is too vague (as many species ostensibly display some capacity for “symbolic thought” – see, e.g., Gallistel, Reference Gallistel, Scarborough and Sternberg1998, Reference Gallistel2011), likewise for any unqualified communicative view of language (as communication is present widely in the animal kingdom [Hauser, Reference Hauser1996; Hauser, Chomsky, & Fitch, Reference Hauser, Chomsky and Fitch2002], but also among fungi and plants [e.g., Boyno & Demir, Reference Boyno and Demir2022]).
4. In this respect, the analysis of the cognitive-computational power allegedly required for particular behaviors (e.g., the generation of recursive linguistic structures, knotting, or complex-patterned art), abstractly analyzed as pertaining to different levels of the Chomsky hierarchy may be a route to explore (cf. Camps & Uriagereka, Reference Camps, Uriagereka, Rosselló and Martín2006). If – and this is a big “if” – we could reliably ascribe particular behaviors/productions to a well-defined cognitive capacity (say, the capacity to compute context-sensitive grammars [producing Type-1 formal languages]), then evidence E of a particular behavior by species S could be taken as confirming the possession of C by S. Consequently, and given the necessary conditions, ¬E could also be justifiably taken as (probabilistically) confirming ¬C.
To conclude, a certain homogeneity of traits is a core assumption for intra-species groupings; we can expect C to be “universal” across the individuals conforming a species (e.g., trichromatic vision, or the capacity for context-sensitive symbolic computations in humans). But in the absence of the relevant evidence, this cannot be extended to inter-species or inter-taxa groupings. In cross-species comparison I think that it is safer to depart from the hypothesis that, say, Homo antecessor or zebra finches lack C (especially when C is a rather unique trait), and then look for evidence of C (cf. i.a. Fitch & Hauser, Reference Fitch and Hauser2004; Beckers, Bolhuis, Okanoya, & Berwick, Reference Beckers, Bolhuis, Okanoya and Berwick2012) than just be satisfied by assuming that they have C and stop further research. Assuming as Stibbard-Hawkes proposes a substantive positive prior could not, by essence, be negatively updated. That is, if we depart from the hypothesis that extinct species S did possess cognitive property C, then there can be no positive evidence E that will disproof (negatively update) our hypothesis.
In sum: We should lower our priors that ¬C, but not invert them to C.
Financial support
This work was supported by the ANR (grant numbers ANR-21-CE27-0005, ANR-22-CE28-0024) and the InSHS of the CNRS (grant code PLRS).
Competing interests
None.