Burkart et al. make a substantial contribution to the literature on the evolution of intelligence. We agree with the implicit view of the authors that fostering connections between contiguous fields is essential in working towards a comprehensive understanding of intelligence. The shared goal includes identifying, at least, the selection pressures that shaped cognitive abilities in different species; the structure of cognitive abilities within different species; outcomes associated with intelligence; and the genetic architecture of intelligence. The target article helpfully reaches out to engage with scholars, questions, and methods emerging from several fields including comparative and differential psychology. We value highly this drawing together of disciplines. Here we raise some points arising from Burkart et al.'s work.
First, we do not find compelling the authors' argument that positive correlations among different cognitive abilities – and the resulting latent variable g – reflect domain-general intelligence. (“[E]vidence for domain-general intelligence in humans, estimated by the first factor derived in psychometric, factor-analytical approaches, is pervasive …” [sect. 1.1.2, para. 3].) By contrast, our empirically testable prediction is that positive correlations among cognitive abilities, and the resulting g factor, will be found within most animal species, whether the species exhibit domain-general intelligence. We expect this because random alterations to a complex system usually degrade its function; genetic mutations that affect multiple cognitive abilities will tend to affect them in the same direction (i.e., deleteriously). Such directional pleiotropy would cause positive correlations among cognitive abilities (even within species that do not exhibit domain-general intelligence). This conjecture is supported by the finding that lower scores on cognitive tests are linked with a greater proportion of the genome in runs of homozygosity (a measure of the extent to which recessive alleles are expressed) (Howrigan et al. Reference Howrigan, Simonson, Davies, Harris, Tenesa, Starr, Liewald, Deary, McRae, Wright, Montgomery, Hansell, Martin, Payton, Horan, Ollier, Abdellaoui, Boomsma, DeRosse, Knowles, Glahn, Djurovic, Melle, Andreassen, Christoforou, Steen, Hellard, Sundet, Reinvang, Espeseth, Lundervold, Giegling, Konte, Hartmann, Rujescu, Roussos, Giakoumaki, Burdick, Bitsios, Donohoe, Corley, Visscher, Pendleton, Malhotra, Neale, Lencz and Keller2016).
Likewise, G – a latent variable arising from factorial analysis of task scores between species – need not reflect domain-general intelligence. In the absence of domain-general intelligence, between-species differences in brain size, neural integrity, complexity, or myelination, for example, could affect different cognitive abilities in the same direction, leading to G. Therefore, neither g nor G is evidence for domain-general intelligence. Further, the causes of g and G may be unrelated; g might be caused by directional pleiotropy, but G could not be. We agree that existing within-species psychometric studies are few, small, and underpowered. The cure is larger studies.
Another important point is that, because latent variables are by definition unobservable, neither g nor G can itself be a direct target of selection – contrary to Burkart et al.'s suggestion that “G is thus the principal locus of selection in the evolution of primate intelligence” (sect. 2.2, para. 4); g or G may reflect a real trait that is visible to selection (Borsboom & Dolan Reference Borsboom and Dolan2006), but we know of no conclusive evidence on this. Identifying biological or cognitive correlates of g and G is a useful approach to this question, but correlation is not causation, and so the cause(s) of g and G remain unclear. An additional note on the topic of selection is that, contrary to the target article (and Woodley et al. Reference Woodley of Menie, Fernandes and Hopkins2015), greater heritability does not indicate stronger recent selection – in fact, all else being equal, the opposite is true (Fisher Reference Fisher1930).
A linked issue is that the nature and cause(s) of g and G, and their relation to natural selection, depend on the tasks that are used to derive the factors. For example, interspecies differences in performance on behavioural tasks may depend on the match of the tasks to the species' typical environments and physical abilities as well as to their cognitive abilities (Barrett Reference Barrett2011) – in which case, the cause(s) of G could have environmental, physical, or cognitive sources.
Also, probing G does not answer the question “why are some species better at ‘catching on’ more generally than others?”; the answer to that lies in the recurrent problems posed by different ecologies and the costs and benefits of solving them. The costs of “generalising” make relatively more domain-general brains a better deal in some settings than in others. We should be cautious in rank-ordering intelligence between species, especially in the absence of comprehensive descriptions of cognitive abilities at the within-species level. Although it is manifestly true that some species are generalisers more than others (compare, for example, koalas with racoons), it is also the case that a smart elephant makes a lousy bat.
It should be noted that even human intelligence, which has been shaped by selection, is not completely general; it is better described as under-specified. For example, although we may inhabit a 10-dimensional universe (Green & Schwarz Reference Green and Schwarz1984), we are unable to form a mental image of higher dimensional figures because our minds have evolved in a space containing relevant objects of only three or fewer dimensions.
We note that we can learn much about the evolution of intelligence from genetic analyses of cognitively well-characterised populations including parameters such as heritabilities, genetic correlations (among mental traits and biological substrates within species), and coefficients of genetic variation. Genetic studies will allow us to test relations among any observed g factors and other fitness-related traits, and to explore evolutionary questions concerning convergence and homologies of cognitive abilities, or mechanisms that contribute to them, across species.
Last, we urge upon us all, conscious perspective-taking of those in other fields. We are all “cursed with knowledge” (Pinker Reference Pinker2014, p. 11). Unpalatable phrases like “positive manifold” (e.g., sect. 1.1.1, para. 3) and “phylogenetic inflection” (sect. 1.2.2, para. 4) act as caltrops impeding the free flow of knowledge and scholarship across disciplines. Reviewers and journals can help by emphasising writing clarity. In saying this, we are not criticising the target article but celebrating and promoting the shared mission to help scholars talk to one another effectively. The focal problem, understanding the evolution of intelligence, is hard; maximising bandwidth across fields is essential.
Burkart et al. make a substantial contribution to the literature on the evolution of intelligence. We agree with the implicit view of the authors that fostering connections between contiguous fields is essential in working towards a comprehensive understanding of intelligence. The shared goal includes identifying, at least, the selection pressures that shaped cognitive abilities in different species; the structure of cognitive abilities within different species; outcomes associated with intelligence; and the genetic architecture of intelligence. The target article helpfully reaches out to engage with scholars, questions, and methods emerging from several fields including comparative and differential psychology. We value highly this drawing together of disciplines. Here we raise some points arising from Burkart et al.'s work.
First, we do not find compelling the authors' argument that positive correlations among different cognitive abilities – and the resulting latent variable g – reflect domain-general intelligence. (“[E]vidence for domain-general intelligence in humans, estimated by the first factor derived in psychometric, factor-analytical approaches, is pervasive …” [sect. 1.1.2, para. 3].) By contrast, our empirically testable prediction is that positive correlations among cognitive abilities, and the resulting g factor, will be found within most animal species, whether the species exhibit domain-general intelligence. We expect this because random alterations to a complex system usually degrade its function; genetic mutations that affect multiple cognitive abilities will tend to affect them in the same direction (i.e., deleteriously). Such directional pleiotropy would cause positive correlations among cognitive abilities (even within species that do not exhibit domain-general intelligence). This conjecture is supported by the finding that lower scores on cognitive tests are linked with a greater proportion of the genome in runs of homozygosity (a measure of the extent to which recessive alleles are expressed) (Howrigan et al. Reference Howrigan, Simonson, Davies, Harris, Tenesa, Starr, Liewald, Deary, McRae, Wright, Montgomery, Hansell, Martin, Payton, Horan, Ollier, Abdellaoui, Boomsma, DeRosse, Knowles, Glahn, Djurovic, Melle, Andreassen, Christoforou, Steen, Hellard, Sundet, Reinvang, Espeseth, Lundervold, Giegling, Konte, Hartmann, Rujescu, Roussos, Giakoumaki, Burdick, Bitsios, Donohoe, Corley, Visscher, Pendleton, Malhotra, Neale, Lencz and Keller2016).
Likewise, G – a latent variable arising from factorial analysis of task scores between species – need not reflect domain-general intelligence. In the absence of domain-general intelligence, between-species differences in brain size, neural integrity, complexity, or myelination, for example, could affect different cognitive abilities in the same direction, leading to G. Therefore, neither g nor G is evidence for domain-general intelligence. Further, the causes of g and G may be unrelated; g might be caused by directional pleiotropy, but G could not be. We agree that existing within-species psychometric studies are few, small, and underpowered. The cure is larger studies.
Another important point is that, because latent variables are by definition unobservable, neither g nor G can itself be a direct target of selection – contrary to Burkart et al.'s suggestion that “G is thus the principal locus of selection in the evolution of primate intelligence” (sect. 2.2, para. 4); g or G may reflect a real trait that is visible to selection (Borsboom & Dolan Reference Borsboom and Dolan2006), but we know of no conclusive evidence on this. Identifying biological or cognitive correlates of g and G is a useful approach to this question, but correlation is not causation, and so the cause(s) of g and G remain unclear. An additional note on the topic of selection is that, contrary to the target article (and Woodley et al. Reference Woodley of Menie, Fernandes and Hopkins2015), greater heritability does not indicate stronger recent selection – in fact, all else being equal, the opposite is true (Fisher Reference Fisher1930).
A linked issue is that the nature and cause(s) of g and G, and their relation to natural selection, depend on the tasks that are used to derive the factors. For example, interspecies differences in performance on behavioural tasks may depend on the match of the tasks to the species' typical environments and physical abilities as well as to their cognitive abilities (Barrett Reference Barrett2011) – in which case, the cause(s) of G could have environmental, physical, or cognitive sources.
Also, probing G does not answer the question “why are some species better at ‘catching on’ more generally than others?”; the answer to that lies in the recurrent problems posed by different ecologies and the costs and benefits of solving them. The costs of “generalising” make relatively more domain-general brains a better deal in some settings than in others. We should be cautious in rank-ordering intelligence between species, especially in the absence of comprehensive descriptions of cognitive abilities at the within-species level. Although it is manifestly true that some species are generalisers more than others (compare, for example, koalas with racoons), it is also the case that a smart elephant makes a lousy bat.
It should be noted that even human intelligence, which has been shaped by selection, is not completely general; it is better described as under-specified. For example, although we may inhabit a 10-dimensional universe (Green & Schwarz Reference Green and Schwarz1984), we are unable to form a mental image of higher dimensional figures because our minds have evolved in a space containing relevant objects of only three or fewer dimensions.
We note that we can learn much about the evolution of intelligence from genetic analyses of cognitively well-characterised populations including parameters such as heritabilities, genetic correlations (among mental traits and biological substrates within species), and coefficients of genetic variation. Genetic studies will allow us to test relations among any observed g factors and other fitness-related traits, and to explore evolutionary questions concerning convergence and homologies of cognitive abilities, or mechanisms that contribute to them, across species.
Last, we urge upon us all, conscious perspective-taking of those in other fields. We are all “cursed with knowledge” (Pinker Reference Pinker2014, p. 11). Unpalatable phrases like “positive manifold” (e.g., sect. 1.1.1, para. 3) and “phylogenetic inflection” (sect. 1.2.2, para. 4) act as caltrops impeding the free flow of knowledge and scholarship across disciplines. Reviewers and journals can help by emphasising writing clarity. In saying this, we are not criticising the target article but celebrating and promoting the shared mission to help scholars talk to one another effectively. The focal problem, understanding the evolution of intelligence, is hard; maximising bandwidth across fields is essential.