Scott: So what do you make of general intelligence?
John Tooby: [chuckles] To heck if I know!
***Exchange at the 2006 Annual Meeting of the Human Behavior and Evolution Society***
Obviously, John Tooby, one of the founders of evolutionary psychology, was being a bit cheeky. But there was also a very large grain of truth to his response. Traditionally, evolutionary psychologists have focused their research efforts on discovering dedicated information-processing mechanisms (‘modules’) that operate on specific content. Evolutionary psychologists have done an impressive job looking at these species-typical cognitive adaptations, elucidating the nature of things that are universally important to humans such as love, sex, social status, music, and art.
Traveling on a separate path, however, intelligence researchers have amassed just as much evidence that individual differences among many disparate cognitive abilities are correlated with one another. This suggests the possibility of causal forces that influence performance on most cognitively complex cognitive tests, regardless of the content. Recently intelligence researchers have proposed two possible causal forces: (a) deleterious mutations or developmental abnormalities that influence many different cognitive mechanisms or (b) cognitive mechanisms that are utilized to some extent in most or all complex cognitive tasks.
Scientists such as Matthew Keller, Geoffrey Miller and Ronald Yeo and others have done important research on (a). I have tended to focus on (b), as have other scientists such as Christopher Chabris, Andrew Conway, Jeremy Gray, Nicholas Mackintosh, and Han L.J. van der Maas. Of course, the two causal forces aren’t mutually exclusive. It is most certainly an interaction of multiple cognitive mechanisms, all of which are affected by developmental instability, that causes the general factor of intelligence (or g for short) to emerge. It is highly unlikely that g is the result of a single process.
Regardless of what comprises general intelligence, at first blush the mere existence of a general intelligence factor seems incompatible with the strong modularity view of the mind– g is domain-general rather than domain-specific, since it is associated with performance on cognitive tasks in a multitude of different contexts.
To be fair, in Leda Cosmides and John Tooby’s original formulation of the evolution of the human mind, they acknowledge the existence of such domain generality, calling such forms of reasoning improvisational intelligence. They argue that this form of reasoning is employed whenever a module doesn’t exist to solve a particular problem. They didn’t get too much into individual differences in domain-general cognition, though, and the field of evolutionary psychology in general has tended to focus on what Leda Cosmides and Tooby refer to as dedicated intelligences that exist, in some degree, in nearly every member of the human species. Evolutionary psychologists have tended to assume that for any trait important to fitness, selection pressure would reduce variance around an optimal level of the trait, with individual differences being random noise.
Recently, a number of researchers, including Lars Penke, Geoffrey Miller, and Satoshi Kanazawa, have attempted to unite evolutionary psychology with differential psychology. To further explore this tension between evolutionary psychology and the theory of general intelligence, I recently teamed up with a superb team of scientists (Colin DeYoung, Jeremy Gray, and Deidre Reis) to examine individual differences in a paradigm that has been used extensively by evolutionary psychologists to provide evidence that cognitive abilities are domain-specific rather than domain-general: the Wason four-card selection task.
On the Wason card selection task, participants are presented with four cards. Participants are told that each card has a letter on one side and a number on the other side. Their task is to decide which cards (and only those cards) need to be turned over to find out whether a given statement is true or false. Here is a common item (try it yourself!):
If you’re like most people, you probably chose A or you choose A and 3. The correct answer is A and 7. It’s important to select the 7 card in order to actively try to falsify the statement– the hallmark of good scientific reasoning. Bad scientific reasoners only search for theories (‘cards’) that confirm their preconceived ideas. In this example, if we turn over the 7 card and there is an A on the other side, we know we have violated the rule. Only about 10-20% of people choose both A and 7.
Try another version of the same task:
If you’re like most people, this version of the task was much easier to solve. Now over 75% of people solve this version of the problem. What a stark difference! But why? what is it about contextualizing the task that makes it so much easier to solve? Various proximal psychological theories have been proposed to explain this effect, but here I want to focus on evolutionary explanations.
According to evolutionary psychologists such as Leda Cosmides and John Tooby, reasoning about realistic scenarios such as social exchange and precautions is supported by dedicated information processing modules that result from evolutionary selection pressures exerted by situations involved in social exchanges or physical danger. These sort of situations involve if-then reasoning concerning violation of rules of specific content. Since no such pressure has been exerted by situations that weren’t reoccurring themes in our evolutionary ancestry, the human mind doesn’t have cognitive modules for more decontextualized forms of reasoning or reasoning using arbitrary rules. In those situation, performance is expected to be much worse.
This hypothesis doesn’t leave much room for general intelligence in domain-specific forms of reasoning that were important in our evolutionary ancestry. I very much wondered, though, whether this is really true. It seemed entirely possible to me that there could be both domain-general and domain-specific contributions to any reasoning task. The prior research on this matter has been mixed, with some studies finding a stronger relation between cognitive ability (sometimes measured using SAT scores) and abstract reasoning relative to contextualized reasoning, another study finding the opposite, and another finding intelligence to be similarly associated with both types of problem.
All of these prior studies have suffered from a few major limitations: (a) a proper general intelligence (g) factor wasn’t extracted, and (b) they involved presentation of only a very few items. This doesn’t allow for assessment of the reliability of deductive reasoning. Prior studies have also presented participants with all four cards at the same time, not allowing assessment of speed of processing on each card. To overcome the limitations of these prior studies, I teamed up with Jeremy Gray and Deidre Reis, who developed an absolutely fantastic computerized version of the Wason Card Selection Task that allows for card-by-card presentation of a large number of items and the logging of both accuracy and reaction time.
All of the items involved contextualized if-then deductive reasoning put into a narrative vignette context. In particular, we administered reasoning on three types of content. Arbitrary-rule problems had arbitrary rules (e.g., “If the soda is diet, then it has to be in a purple container“). These rules are arbitrary in that they do not correspond to established rules in the individual’s experience. Precautionary reasoning problems involve rules related to avoiding potential physical danger (e.g., “If you surf in cold water, then you have to wear a wetsuit“). Finally, Social exchange problems concern the mutual exchange of good or services between individuals in specific situations. The rules for social exchange items generally involve detecting if one party might be taking a benefit without fulfilling an obligation.
Here is an example of a social exchange scenario:
We also administered a number of measures of general intelligence, including verbal reasoning, non-verbal reasoning, spatial reasoning, working memory, processing speed, and explicit associative learning ability. This paradigm allowed us to rigorously test prior theories, such as Satoshi Kanazawa’s theory that general intelligence is only correlated with performance on evolutionarily unfamiliar, but not evolutionarily familiar, problems.
Indeed, Kanazawa’s hypothesis struck us as highly unlikely, in light of the pervasiveness of general intelligence in many spheres of human functioning. Also, the logic behind Kanazawa’s hypothesis that g should be associated only with performance on evolutionarily novel problems seems to rely on the premise that individual differences in an evolved cognitive ability will be reflected in performance only on the type of problem that the ability evolved to solve. This premise overlooks the existence of exaptation, in which traits evolved for one purpose are eventually used for other purposes.
We also found his evolutionary logic debatable. Evolutionarily novel events of the kind that Kanazawa describes are rare by definition. Although rare events can have consequences for evolution if they affect sufficiently large numbers of a species, most rare events are likely to affect a small proportion of individuals, and their rarity will prevent them from exerting consistent selection pressures.
It seemed more likely to us that that mechanisms for general intelligence would have evolved in response to all situations for which a pre-existing adaptation did not produce an optimal response (also see David Geary’s important work on the evolution of general intelligence). Sure, this class of situations would include evolutionarily novel situations, but importantly it would also include evolutionarily familiar situations of sufficient complexity to interfere with the heuristic response of a dedicated cognitive module or to render its effectiveness uncertain. Thus, rare evolutionarily novel events may simply be one example of a larger class of situations, namely those that are complex and unpredictable.
Indeed, Linda Gottfredson has demonstrated the pervasiveness of g in the complexity of everyday life. Also, social group size increases and rapidly increasing cultural complexity are likely to have rendered pre-existing heuristic adaptations increasingly fallible in human ancestors, thus increasing the selection pressure on domain-general mechanisms that could logically analyze the causal structure of situations that were too complex to be to adequately processed by modular heuristics. Research has indeed shown (a) a positive correlation between expansion of the neocortex with social group size across primate species, and (b) a correlation across these species between brain size and domain general learning ability. (I must point out that it could be sensibly argued that this still shows the domain-specificity of general intelligence: it is indeed specific to a given domain (complex and unpredictable situations), but the domain is just very broad. While I acknowledge this conceptual possibility, I personally like to make a distinction between processes that affect a very large part of the system and those that are much more narrow to specific content).
Evolutionary psychologists have sometimes argued that a class of situations must be relatively narrow to exert a consistent selection pressure, but this claim is insufficiently justified. Any regularity in the environment can exert selection pressure if it poses a challenge or opportunity to the organism, and whether this will prompt adaptation simply reflects the liklihood that genetic variation might lead to variation in the ability to meet the challenge or seize the opportunity. In the case of complex, unpredictable situations, regardless of their superficial dissimilarity, selection for increased ability to analyze causal structure is highly likely. Existing adaptations may facilitate performance on evolutionarily familiar problems, but general intelligence should provide additional facilitation.
At the end of the day, information-processing is typically accomplished through a combination of domain-general and domain-specific mechanisms. Therefore, we predicted that although there would be group-level differences in performance between evolutionarily novel and evolutionarily familiar forms of reasoning, performance on all types of problems would be correlated with each other and also with g, reflecting the additional effect of domain general processes, over and above any species-typical biases conferred by evolved modules.
So now all this theoretical speculation is out of the way, what did we actually find?
First, we replicated some major findings in the prior literature. All of our measures of general intelligence positively loaded on a g factor, suggesting that they were all tapping into a single explicit cognitive ability construct. Secondly, at the group level of analysis, the precautionary and social exchange reasoning items were solved by a much larger proportion of the participants than were the arbitrary reasoning items. In other words, reasoning that was place in an evolutionarily familiar context was found much easier to solve than when arbitrary rules were applied.
The findings at the individual differences level were also telling. The reliability of accuracy on all 70 reasoning items, regardless of content, was extremely high (α = .88). Therefore, collapsing across trial type, there is considerable variance that is common among all the reasoning problems. This prompted us to create a general factor of both speed and accuracy of deductive reasoning, which we used to assess relations with g.
As it turned out, accuracy of deductive reasoning was correlated significantly with g. Each form of reasoning (arbitrary, precautionary, and social exchange), and the the deductive reasoning accuracy factor, was significantly correlated with g. Interestingly though, g was not related to speed of deductive reasoning whatsoever. Presumably, explicit reasoning plays more of a role in logical accuracy, and cognitive modules play more of a role in facilitating the speed of contextualized reasoning than contributing to the reporting of logical accuracy.
Here are those results:
These results suggest that reasoning on evolutionarily familiar content shows reliable and consistent individual differences, and accuracy is associated with general intelligence. These results support our hypothesis that domain general cognitive abilities should facilitate the solution of any explict cognitively complex problem, regardless of whether it is additionally facilitated by evolved modular heuristics. These results directly contradict the idea that general intelligence should only be related to performance on “evolutionarily novel” problems.
In fact, our results suggest that such a stark contrast between “evolutionarily novel” and “evolutionarily familiar” problems is probably misguided when considering individual differences, because we found that g is significantly associated with evolutionarily familiar forms of reasoning involving precautions and social exchanges. In sum, it appears that domain general cognitive mechanisms underlying g are actively involved in any form of explicit reasoning of sufficient complexity, even if more specific psychological mechanisms are also brought to bear on the task at hand.
These results also have important implications for theories of multiple intelligences and the great rationality debate. You can read the full, paper here, which is currently in-press in the journal Intelligence, for further discussion of these important issues.
It seems that general intelligence is very much compatible with evolutionary psychology. I look forward to more research bringing these two fields closer together, as both fields have a lot to learn from the other.
© 2011 by Scott Barry Kaufman
Follow me on Twitter or Facebook.
Note: Portions of this post were taken from the following paper:
Kaufman, S.B., DeYoung, C.G., Reis, D.L., & Gray, J.R. (2011). General intelligence predicts reasoning ability even for evolutionarily familiar content. Intelligence. [pdf]