Project 1: Fractionating the Rational Brain

I recently returned from the grocery store accompanied by my young daughter. We bought kiwi fruit for the first time. I told her that kiwi are perishable fruit, and that perishable fruit are placed in the refrigerator. She promptly placed the kiwi in the refrigerator. She had never seen kiwi before. I did not explicitly tell her that kiwi are to be placed in the refrigerator. How did she know to put them in the refrigerator? Also, by the time things were put away, she had formed the belief that all kiwi have a brown, furry skin. …. but we only bought six kiwi. From seeing these six, she generalized to all kiwi. Yet I am not surprised by her actions. Her behaviour is not a mystery. It is just an example of the reasoning brain at work.

Reasoning is the activity of evaluating arguments. All arguments involve the claim that one or more propositions (the premises) provide some grounds for accepting another proposition (the conclusion). Philosophers have sorted arguments into two broad categories -- induction and deduction -- based on the nature of the relationship between premises and conclusion. Valid deductive arguments involve the claim that their premises provide absolute grounds for accepting the conclusion. For example: (A) All kiwi are perishable; All perishable fruit belong in the refrigerator \ Kiwi belong in the refrigerator. Validity is a function of the logical structure as opposed to sentence content. This argument is equally valid irrespective of whether it is about kiwi or chairs.

Arguments where the premises provide only limited grounds for accepting the conclusion are broadly called inductive arguments. For example: (B) These are kiwi fruit; They have brown, furry skin \ All kiwi fruit have brown, furry skin. (C) These are kiwi fruit; They are wrapped in plastic \ All kiwi fruit are wrapped in plastic. Neither is a valid argument. However, most of us would be prepared to accept the conclusion in (B) as plausible or reasonable but we would not accept the conclusion of (C) as plausible or reasonable. But interestingly, both arguments have an identical logical structure; they differ only in content. So, unlike deduction (e.g. A), induction (e.g. B & C) is a function of the content of the sentence and our knowledge of the world. It is usually a matter of knowing which properties generalize in the required manner and which don't.

Deductive Reasoning

Two theories of deductive reasoning dominate the cognitive literature. Sentential (mental logic) theories of reasoning claim that deductive reasoning is a rule governed syntactic process where internal representations preserve structural properties of linguistic strings in which the premises are stated. This linguistic hypothesis predicts that language processing mechanisms underwrite human reasoning processes. Mental model theories claim that deductive reasoning is a process requiring spatial manipulation and search where internal representations preserve the structural properties of the world (e.g. spatial relations) that the sentences are about. This spatial hypothesis suggests that the neural structures for visuo-spatial processing contribute the basic representational building-blocks used for logical reasoning.

In a series of studies we have made five significant contributions to the cognitive neuroscience of reasoning.

1) Reasoning involves multiple pathways (Goel, Buchel, Frith, & Dolan, 2000; Goel & Dolan, 2003; Goel, Makale, & Grafman, in press). Reasoning about familiar material or content (e.g. All apples are red; All red things are sweet; All apples are sweet) engages a left hemisphere frontal-temporal system. By contrast, in a formally identical reasoning task, with unfamiliar (or unspecified) content (e.g. All A are B; All B are C; All A are C), a bilateral frontal-parietal system is recruited. An important factor in the determination of which mechanism is engaged is the presence or absence of content terms in the argument.

2) There is greater involvement of prefrontal cortex in reasoning about familiar material than unfamiliar material.

3) The left hemisphere is necessary and sometimes sufficient for reasoning, while the right hemisphere is sometimes necessary, but not sufficient (Goel et al., 2000; Goel & Dolan, 2003; Goel, Gold, Kapur, & Houle, 1997, 1998).

4) When a logical argument results in a belief-logic conflict, the right prefrontal cortex plays a critical role in detecting and/or resolving the incongruency (Goel et al., 2000; Goel & Dolan, 2003).

5) When logical reasoning is overcome by belief-bias, there is engagement of ventral medial prefrontal cortex, a region implicated in affective processing. This latter involvement suggests that belief-bias effects in reasoning may be mediated through an influence of emotional processes on reasoning.

Our findings shed new light on the mechanisms involved in human reasoning and provide support for a dual mechanism theory, along lines, not anticipated by the dominant cognitive theories (Johnson-Laird, 1994; Rips, 1994). The distinction that our results point to is between reasoning with familiar, conceptually coherent material vs. unfamiliar, nonconceptual or incoherent material, The former engages a left frontal-temporal system (language and long-term memory) while the latter engages a bilateral parietal (visuo-spatial) system. We believe that the frontal-temporal system is more “basic,” and effortlessly engaged. It has temporal priority. By contrast, the parietal system is effortfully engaged when the frontal-temporal route is blocked due to a lack of familiar content, or when a conflict is detected between the logical response and belief-bias. This is very consistent with the dual mechanism account developed by Newell & Simon (1972) for the domain of problem solving. On this formulation our frontal-temporal system corresponds to the “heuristic” system while the parietal system corresponds to the “universal” system. Reasoning about familiar situations automatically utilizes situation-specific heuristics, which are based on background knowledge and experience. Where no such heuristics are available (as in reasoning about unfamiliar situations), universal (formal) methods must be used to solve the problem. In the case of syllogistic reasoning this may well involve a visuo-spatial system.

Our results go beyond addressing cognitive theories of reasoning and provide new insight into the role of the prefrontal cortex in human reasoning. In particular, the involvement of the prefrontal cortex in logical reasoning is selective and asymmetric. Its engagement is greater in reasoning about familiar, content-rich situations than unfamiliar, content-sparse situations. The left prefrontal cortex is necessary and often sufficient for reasoning. The right prefrontal cortex is sometimes necessary, but not sufficient for reasoning. It is engaged in the absence of conceptual content and in the face of conflicting or conceptually incoherent content (as in the belief-logic conflicts discussed above). Finally, the VMPFC is engaged by non-logical, belief-biased responses.
They also raise a number of questions. We are currently pursuing the following four issues: First, the above results have been generated with syllogisms, which are quite difficult. Will they generalize to basic low-level inferences such as modus ponens and modus tollens? Second, given the involvement of visuo-spatial processing systems in much of reasoning, and the postulated differences between males and females in processing spatial information, one might expect neural-level differences in reasoning between the sexes. Third, the issue of task difficulty has not been explored. As reasoning trials become more difficult, are additional neural resources recruited, or are the same structures activated more intensely? Fourth, what is the effect of learning/training on the neural mechanisms underlying reasoning?

Inductive Reasoning

Cognitive theories typically view induction as a form of hypothesis generation and testing, where the crucial issue is one of searching a large data base and determining which pieces of information are relevant and how they are to be mapped onto the present situation. In several articles we have shown a dissociation in the functional anatomy of inductive & deductive reasoning and fractionated the neural correlates of hypothesis generation and evaluation in inductive reasoning.

In an early PET study (Goel, Gold et al., 1997) we addressed the question of the functional anatomy of inductive and deductive reasoning and reported that both induction and deduction activated a similar left frontal-temporal system, and that induction differed from deduction in greater activation of medial dorsal prefrontal cortex (BA 8 and 9). In a more current fMRI study (Goel & Dolan, in review), we have elaborated and clarified these results. Both types of reasoning are characterized by activation of left lateral prefrontal and bilateral dorsal frontal, parietal, and occipital cortices. However, deduction results in greater activation of left inferior frontal gyrus (BA 44) than induction, while left dorsolateral (BA 8/9) prefrontal gyrus shows greater activity during induction than deduction. This pattern suggests a dissociation within prefrontal cortex for deductive and inductive reasoning.

In another study (Goel & Dolan, 2000) we investigated the neural substrate of inductive inference, particularly hypothesis selection, using fMRI. We discovered that rule inference was specifically associated with bilateral hippocampal activation while the task by difficulty interaction was associated with activation in right lateral orbital prefrontal cortex. We interpret the former in terms of semantic encoding of novel stimuli and the latter in terms of hypothesis selection. Thus we show an anatomical dissociation between task implementation and task difficulty that may correspond to a critical psychological distinction in the processes necessary for inductive inference.
Given that induction is content-based, we are currently trying to tease apart the respective contributions of memory and inference mechanisms.

Our studies of deductive and inductive reasoning are converging on the same anatomical hypothesis as Project 3. Specifically, Right and left prefrontal cortex are differentially involved in cognitive processing. The right PFC is preferentially involved in nonconceptual and incongruent reasoning, while the left PFC is critical for reasoning with conceptually meaningful, coherent, material.

Basic operating costs of these studies are funded by my McDonnell-Pew Award (Title of Project: “Imaging the Reasoning Brain”) and NSERC Operating Grants.


Nov. 10, 2003