Keith Holyoak’s career has been devoted to understanding the representation of relational knowledge in the human mind and brain. The central focus of his work has been on how people learn and use abstract relations that depend on more than direct similarity of the entities they relate—relations of the sort required to grasp analogies, social regulations, moral principles, and causality.
Analogies and Schemas
Working together with graduate student Mary Gick at the University of Michigan, Keith introduced an experimental paradigm that provided a way to investigate the use of analogies to solve complex problems (Gick & Holyoak, 1980, 1983). This paradigm (inspired by ideas from Gestalt psychology) allows investigators to bring a kind of creative thinking under experimental control by first “planting” a potential source analog (e.g., a story introduced in the incidental context of a memory study), and then assessing when or if participants would use a solution suggested in the story to solve an ostensibly unrelated (but analogous) target problem. These studies established a clear dissociation between the retrieval of distant analogies and their potential to support inferences and the acquisition of abstract schemas.
In the decades since, Keith and his collaborators, as well as many other researchers, have worked to pin down the factors that determine when/if people use remote analogies successfully. His group pioneered investigation of developmental changes in the ability to solve problems by analogy (Holyoak, Junn & Billman, 1984; Richland, Morrison, & Holyoak, 2006). Their work illuminated the conditions under which analogical reasoning can be effective in education, especially in teaching mathematics (e.g., Richland, Zur & Holyoak, 2007). A key early finding was that analogical transfer can be greatly enhanced by comparing multiple source analogs, thereby creating a more abstract schema for a class of problems. A related line of work showed that relational schemas facilitate everyday reasoning about novel situations (Cheng & Holyoak, 1985). Augmented by instructions to compare and/or contrast multiple examples, the basic experimental paradigm that Keith originated became an educational tool that has been used to teach students important relational schemas in many contexts, including math, science, and business negotiations.
Neural and Computational Investigations
Keith was one of the first researchers to investigate the neural basis of higher cognition, focusing on the substrate for relational thinking in the human brain. A theoretical paper by Robin and Holyoak (1995) set the stage by relating the cognitive processes underlying relational reasoning to functions of subregions within the prefrontal cortex. Working with neuropsychologist Barbara Knowlton and others, Keith investigated relational reasoning in patients suffering from Frontotemporal Lobar Degeneration. These studies revealed that patients with extensive damage to the prefrontal cortex are selectively impaired at integrating multiple relations (Waltz et al., 1999) and handling interference (Morrison et al., 2004). In related work with college students, neuroimaging showed that the rostrolateral subregion of the prefrontal cortex is especially important in reasoning with multiple relations (Christoff et al., 2001; Kroger et al., 2002), whereas more posterior frontal subregions are involved in inhibitory control during relational reasoning (Cho et al., 2010). In recent years many other investigators, using a broad range of imaging methods, have been able to provide a more detailed picture of how the brain supports relational reasoning (reviewed by Holyoak & Monti, 2021).
Over the course of his career, Keith and his collaborators have formulated several computational models that perform tasks requiring relational reasoning. (Holyoak & Thagard, 1989; Hummel & Holyoak, 1997, 2003). These models are able to account for numerous empirical phenomena related to relational reasoning, both at the behavioral and neural levels. Recent work has addressed the questions of how distributed representations of semantic relations can be acquired from non-relational inputs (Lu, Wu, & Holyoak, 2019), and how distributed representations can be used to find mappings between the key concepts in complex analogs (Lu, Ichien, & Holyoak, 2022).
Reasoning by Constraint Satisfaction
Keith’s research tested the hypothesis that complex human reasoning can emerge from interactions between multiple local constraints. Together with law professor Dan Simon, Holyoak provided experimental evidence that complex decisions about legal cases are made in accord with principles of constraint satisfaction (Holyoak & Simon, 1999). Consumer choices operate in a similar fashion, with product attributes being reevaluated to cohere with the eventual choice (Lee & Holyoak, 2021). The constraint satisfaction framework guided work on methods to help overcome misinformation that has made many people hesitant to have their children vaccinated against childhood diseases (Horne et al., 2015).
Other Research Contributions
Keith’s research has addressed numerous other topics related to human thinking. These include:
(1) Semantic verification. With fellow Stanford graduate student Arnold Glass (Glass & Holyoak, 1975; Holyoak & Glass, 1975), Keith identified some of the mechanisms by which people evaluate semantic generalizations, focusing on ways in which sentences can be rejected on the basis of internal contradictions (“All kings are queens”) or apparent counterexamples (“All birds are robins”).
(2) Causal learning and inference. Waldmann and Holyoak (1992; Waldmann, Holyoak & Fratianne, 1995) performed a series of experiments demonstrating that the psychological representation of cause-effect relations can be dissociated from the overt temporal order of presented cues and their outcomes (e.g., a doctor may use symptoms to infer diseases, even though the causal arrow runs from diseases to symptoms). Keith and his collaborators applied a Bayesian framework to model uncertainty about cause-effect relations, providing compelling evidence that associative learning (even when formulated within the same Bayesian framework) is inadequate as an account of human causal learning (Lu et al., 2008).
(3) Comparative judgment. In several projects over his career, Keith explored the special properties of concepts that can be ordered along a unidimensional continuum of magnitude (e.g., size, intelligence), which allow comparative judgments to be made based on semantic knowledge. Holyoak (1978) proposed that the endpoints of continua routinely are used as implicit reference points, in effect stretching subjective distances between concepts close to the reference point, thus influencing comparative judgments. In related work, Holyoak and Gordon (1983) showed that the self serves as a reference point in social judgments, leading to asymmetries in perceived similarity. Chen, Lu, and Holyoak (2014) developed a computational model of how magnitude continua can be learned from examples and then used to make comparative judgments.
(4) Category learning. Keith has also contributed to understanding the mechanisms of human category learning. He developed and tested the earliest Bayesian model of human category learning, the category density model (Fried & Holyoak, 1984). This early line of work on category learning anticipated subsequent dramatic advances in Bayesian approaches to human induction.
In a parallel career as a poet and translator, Keith has published several poetry collections. Along the way, he has also drawn connections between literary symbolism and cognitive science. He recently developed this theme in greater depth in his monograph, The Spider’s Thread: Metaphor, in Mind, Brain, and Poetry (Holyoak, 2019).
Honors and Service Contributions to the Field
Keith has been a recipient of a John Simon Guggenheim Fellowship and a James McKeen Cattell Fellowship. He is a Fellow of the American Academy of Arts and Sciences, American Association for the Advancement of Science, the Association for Psychological Science, the Cognitive Science Society, and the Society of Experimental Psychologists. Keith served as Editor of Psychological Review from 2016-2021. He previously served as Editor of Cognitive Psychology, Senior Editor of Cognitive Science, Associate Editor of Psychological Science, and as editorial board member for numerous other journals.