Burkart et al. offer an impressive integration of intelligence research across humans and nonhuman species. Their commendable synthesis will serve as a valuable, centralized resource. Despite these strengths, the target article falls short of offering compelling explanations for the evolution of intelligence.
We observe three major issues with the target article. First, it poses multiple questions about intelligence but does not consistently differentiate between them – despite their likely different answers. For example, the question of whether interspecific variation in psychometric intelligence (G) exists is fundamentally distinct from whether G taps the same construct as within-human variation in intelligence (g). Independent of these questions are why G exists and why g exists – two independent questions that may have different answers.
Conflating these questions can lead to misguided conclusions about the evolution of intelligence. The article establishes the existence of both G and g. However, it does not logically follow that they therefore (1) tap the same construct or (2) share the same evolutionary origins. First, the authors offer little defense of the implicit position that g and G tap the same construct. Second, it is plausible that some species exhibit superior performance on intelligence batteries as a consequence of cross-species differences in the information-processing demands of survival- and reproduction-related problems. Individual differences in intelligence among humans may have entirely different origins. Prokosch et al. (Reference Prokosch, Yeo and Miller2005) proposed that g captures individual differences in “developmental stability at the level of brain development and cognitive functioning” (p. 203). For several reasons, this alternative evolutionary model deserves consideration alongside the target article. First, the term “evolved” refers not to the products of just selection, but also of genetic drift, gene flow, and mutation. The target article neglects these non-selective forces and how they could produce g. By contrast, Prokosch et al. considered a more comprehensive set of evolutionary forces and posited that g reflects the outcome of a balance between selection and genetic mutation. The target article offers no consideration of the mechanistic basis of variation in intelligence. Second, Prokosch and colleagues generated clear, novel predictions based on their model. It is not immediately clear what new predictions the target article's “cultural intelligence” (CI) approach yields. The crucial idea is not that we favor Prokosch et al.'s model, but rather that their work exhibits hallmarks of sound evolutionary science that the CI approach, in its current form, lacks. These include a consideration of selective and non-selective forces, as well as the generation of specific, falsifiable predictions. At present, it is not clear what evidence could disconfirm the CI model. We suggest that the CI approach could benefit from more clearly articulating its empirical predictions, with an emphasis on identifying where it and alternative models advance divergent predictions.
A second, related issue is that the target article attempts to use inappropriate criteria to discriminate between the CI and alternative evolutionary models. Here, we provide four examples of this. First, Figure 1b in the target article presents a pattern of cognitive performance expected from domain-general mechanisms in homogeneous developmental conditions. However, this pattern is identical to that expected when selective forces favor domain-specific mechanisms but non-selective forces (e.g., mutation) impair the performance of these mechanisms. Second, the target article acknowledges that intelligence tests are culture-biased. If we recognize this, then we – the creators of these tests – should certainly acknowledge that they could be species-biased. Intelligence batteries tap cognitive performance on different tasks. If the computational demands of these tasks align more closely with the computational demands of the adaptive problems faced by some species, then we should expect interspecific variation in performance on these tasks – G. As such, the existence of G is not “particularly difficult to reconcile” (sect. 2.5, para. 5) with domain-specific mechanisms. We agree with the authors that reconciliation between the massive modularity hypothesis and domain-general views of intelligence is needed, but the mere existence of G is insufficient for adjudicating between them.
Third, the target article interprets the absence of “empirical evidence … of specialized adaptive behavioral functions to specific modular neural units” (sec. 1.2.1, para. 2) as evidence against domain-specific mechanisms. This reflects a deep misunderstanding of domain-specificity. A domain-specific mechanism is one that has specialized computational functions, not one that has a delimited neural area.
Fourth, the article ascribes an inability to learn to “primary modules” (sec. 1.2.3, para. 2), which it synonymizes with domain-specific mechanisms. Consequently, the authors use learning as an evidentiary criterion against domain-specific mechanisms. This misconception has been addressed in two recent publications in the flagship journal of the American Psychological Association (e.g., Confer et al. Reference Confer, Easton, Fleischman, Goetz, Lewis, Perilloux and Buss2010; Lewis et al. [Reference Lewis, Al-Shawaf, Conroy-Beam, Asao and Buss2017]).
These problems point toward our third major issue: the target article badly mischaracterizes contemporary evolutionary psychological thinking. The domain-specific mechanisms proposed by evolutionary psychologists process inputs from the environment, execute computational procedures on these inputs, and produce outputs – including social learning (see Henrich & Gil-White Reference Henrich and Gil-White2001; Lewis et al. [Reference Lewis, Al-Shawaf, Conroy-Beam, Asao and Buss2017]). Accordingly, portraying social learning and domain-specific mechanisms as competing alternatives is highly misleading. Domain-specific adaptations can ontogenetically canalize social learning (e.g., see Henrich & Gil-White Reference Henrich and Gil-White2001; see also Karmiloff-Smith's “domain relevant” approach [Reference Karmiloff-Smith2015, p. 91]). Crucially, this view squares with the literature presented in the target article without forcing the unnecessary and outdated dichotomy between innate versus learned.
We have critiqued several aspects of this article, but we believe it has the potential to advance research on the evolution of intelligence. In particular, the article implicitly points toward cost-benefit analysis as a valuable tool. Applying this tool to cross-species differences in the computational complexity of survival- and reproduction-related problems could be fruitful for understanding G. For example, whether a species faces a heterogeneous or homogeneous environment and whether the adaptive problems it faces are characterized by social contingencies (e.g., the psychology of conspecifics) may influence the information-processing complexity of the species' adaptive problems. Comparative analysis of the information-processing complexity of these problems, in conjunction with cost-benefit analyses of the cognitive architecture needed to solve them, has the potential to yield new and testable hypotheses about the evolution of G.
Burkart et al. offer an impressive integration of intelligence research across humans and nonhuman species. Their commendable synthesis will serve as a valuable, centralized resource. Despite these strengths, the target article falls short of offering compelling explanations for the evolution of intelligence.
We observe three major issues with the target article. First, it poses multiple questions about intelligence but does not consistently differentiate between them – despite their likely different answers. For example, the question of whether interspecific variation in psychometric intelligence (G) exists is fundamentally distinct from whether G taps the same construct as within-human variation in intelligence (g). Independent of these questions are why G exists and why g exists – two independent questions that may have different answers.
Conflating these questions can lead to misguided conclusions about the evolution of intelligence. The article establishes the existence of both G and g. However, it does not logically follow that they therefore (1) tap the same construct or (2) share the same evolutionary origins. First, the authors offer little defense of the implicit position that g and G tap the same construct. Second, it is plausible that some species exhibit superior performance on intelligence batteries as a consequence of cross-species differences in the information-processing demands of survival- and reproduction-related problems. Individual differences in intelligence among humans may have entirely different origins. Prokosch et al. (Reference Prokosch, Yeo and Miller2005) proposed that g captures individual differences in “developmental stability at the level of brain development and cognitive functioning” (p. 203). For several reasons, this alternative evolutionary model deserves consideration alongside the target article. First, the term “evolved” refers not to the products of just selection, but also of genetic drift, gene flow, and mutation. The target article neglects these non-selective forces and how they could produce g. By contrast, Prokosch et al. considered a more comprehensive set of evolutionary forces and posited that g reflects the outcome of a balance between selection and genetic mutation. The target article offers no consideration of the mechanistic basis of variation in intelligence. Second, Prokosch and colleagues generated clear, novel predictions based on their model. It is not immediately clear what new predictions the target article's “cultural intelligence” (CI) approach yields. The crucial idea is not that we favor Prokosch et al.'s model, but rather that their work exhibits hallmarks of sound evolutionary science that the CI approach, in its current form, lacks. These include a consideration of selective and non-selective forces, as well as the generation of specific, falsifiable predictions. At present, it is not clear what evidence could disconfirm the CI model. We suggest that the CI approach could benefit from more clearly articulating its empirical predictions, with an emphasis on identifying where it and alternative models advance divergent predictions.
A second, related issue is that the target article attempts to use inappropriate criteria to discriminate between the CI and alternative evolutionary models. Here, we provide four examples of this. First, Figure 1b in the target article presents a pattern of cognitive performance expected from domain-general mechanisms in homogeneous developmental conditions. However, this pattern is identical to that expected when selective forces favor domain-specific mechanisms but non-selective forces (e.g., mutation) impair the performance of these mechanisms. Second, the target article acknowledges that intelligence tests are culture-biased. If we recognize this, then we – the creators of these tests – should certainly acknowledge that they could be species-biased. Intelligence batteries tap cognitive performance on different tasks. If the computational demands of these tasks align more closely with the computational demands of the adaptive problems faced by some species, then we should expect interspecific variation in performance on these tasks – G. As such, the existence of G is not “particularly difficult to reconcile” (sect. 2.5, para. 5) with domain-specific mechanisms. We agree with the authors that reconciliation between the massive modularity hypothesis and domain-general views of intelligence is needed, but the mere existence of G is insufficient for adjudicating between them.
Third, the target article interprets the absence of “empirical evidence … of specialized adaptive behavioral functions to specific modular neural units” (sec. 1.2.1, para. 2) as evidence against domain-specific mechanisms. This reflects a deep misunderstanding of domain-specificity. A domain-specific mechanism is one that has specialized computational functions, not one that has a delimited neural area.
Fourth, the article ascribes an inability to learn to “primary modules” (sec. 1.2.3, para. 2), which it synonymizes with domain-specific mechanisms. Consequently, the authors use learning as an evidentiary criterion against domain-specific mechanisms. This misconception has been addressed in two recent publications in the flagship journal of the American Psychological Association (e.g., Confer et al. Reference Confer, Easton, Fleischman, Goetz, Lewis, Perilloux and Buss2010; Lewis et al. [Reference Lewis, Al-Shawaf, Conroy-Beam, Asao and Buss2017]).
These problems point toward our third major issue: the target article badly mischaracterizes contemporary evolutionary psychological thinking. The domain-specific mechanisms proposed by evolutionary psychologists process inputs from the environment, execute computational procedures on these inputs, and produce outputs – including social learning (see Henrich & Gil-White Reference Henrich and Gil-White2001; Lewis et al. [Reference Lewis, Al-Shawaf, Conroy-Beam, Asao and Buss2017]). Accordingly, portraying social learning and domain-specific mechanisms as competing alternatives is highly misleading. Domain-specific adaptations can ontogenetically canalize social learning (e.g., see Henrich & Gil-White Reference Henrich and Gil-White2001; see also Karmiloff-Smith's “domain relevant” approach [Reference Karmiloff-Smith2015, p. 91]). Crucially, this view squares with the literature presented in the target article without forcing the unnecessary and outdated dichotomy between innate versus learned.
We have critiqued several aspects of this article, but we believe it has the potential to advance research on the evolution of intelligence. In particular, the article implicitly points toward cost-benefit analysis as a valuable tool. Applying this tool to cross-species differences in the computational complexity of survival- and reproduction-related problems could be fruitful for understanding G. For example, whether a species faces a heterogeneous or homogeneous environment and whether the adaptive problems it faces are characterized by social contingencies (e.g., the psychology of conspecifics) may influence the information-processing complexity of the species' adaptive problems. Comparative analysis of the information-processing complexity of these problems, in conjunction with cost-benefit analyses of the cognitive architecture needed to solve them, has the potential to yield new and testable hypotheses about the evolution of G.