Gilead and colleagues operationalize abstraction as a spectrum from the concrete and modality based, to the intermediate and multimodal, to the most abstract and theory-like. This aligns with earlier work on the spectrum of representations that people draw upon, from episodic to semantic (Tulving Reference Tulving1972), or from model-free to model-based (Gershman & Daw Reference Gershman and Daw2017). Gilead et al. further suggest that higher-level abstractions are representationally and neurally distinct from lower-level ones, and that different types of mental travels rely on different levels of this abstraction spectrum. Specifically, Gilead et al. limit simulation to more concrete representations. This formulation successfully captures some cases of prospection. For example, it accurately predicts that people will rely on abstract and theory-based sources in order to mentally travel to distal events (e.g., events in the far future, or about dissimilar others) and more concrete sources for proximal events (e.g., events in the near future, or about similar and familiar others). However, this formulation, like much prior work on simulation, underspecifies the dynamic ways in which different levels of abstraction interact during simulation. We propose that simulation requires agents to bridge representations at multiple levels of abstraction, in a dynamic interplay between abstract and concrete knowledge.

Higher-level abstracta scaffold lower-level ones, across multiple domains. During visual perception, category cues shape perceptual expectations (Gandolfo & Downing Reference Gandolfo and Downing2019). During social perception, the structure of emotion concepts shapes the perception of emotion in faces (Brooks & Freeman Reference Brooks and Freeman2018; Brooks et al. Reference Brooks, Chikazoe, Sadato and Freeman2019). During simulation, abstracta such as event knowledge and personal semantics (Renoult et al. Reference Renoult, Davidson, Palombo, Moscovitch and Levine2012) shape which specific, concrete details people will use to fill in the blanks (Conway & Pleydell-Pearce Reference Conway and Pleydell-Pearce2000; D'Argembeau & Mathy Reference D'Argembeau and Mathy2011). This suggests that abstract and semantic knowledge support episodic simulation.

To test this perspective, our lab studied individuals who excel at simulation: creative experts. Most people produce distal simulations with highly abstract, schematic content, as predicted by Gilead et al. However, creative experts continue to generate rich, vivid, detailed simulations even about very distal events. How do they do so? We find that during distal simulations, creative experts preferentially recruit the dorsomedial subsystem of the default mode network (Meyer et al. Reference Meyer, Hershfield, Waytz, Mildner and Tamir2019), a network associated with semantic knowledge and high-level construals (Baetens et al. Reference Baetens, Ma, Steen and Overwalle2014; Binder et al. Reference Binder, Desai, Graves and Conant2009; Fairhall & Caramazza Reference Fairhall and Caramazza2013; Gilead et al. Reference Gilead, Liberman and Maril2014; Simony et al. Reference Simony, Honey, Chen, Lositsky, Yeshurun, Wiesel and Hasson2016; Spunt et al. Reference Spunt, Kemmerer and Adolphs2016); both groups recruit the medial temporal subsystem, associated with episodic detail, to the same extent. This suggests that creative experts are able to generate vivid and concrete distal events by harnessing abstract knowledge. In fact, the system that organizes abstract information may facilitate creative ability: semantic memory organization in creative experts is less hierarchical than in typical thinkers (Kenett et al. Reference Kenett, Anaki and Faust2014; Mednick Reference Mednick1962). Together, these findings suggest that successful simulation requires the ability to flexibly integrate abstract knowledge with concrete representations.

Not only do abstract representations shape concrete ones, concrete representations also shape abstract ones. Abstracta are not stable knowledge stores; they are dynamic, probabilistic distributions over concreta (Griffiths et al. Reference Griffiths, Vul and Sanborn2012). Abstracta shift as the availability of concreta ebbs and flows. This has been demonstrated across different domains of cognition. During decision making, an incidental reminder of a single past reward episode biases decisions away from abstract summary reward values (Bornstein et al. Reference Bornstein, Khaw, Shohamy and Daw2017). During impression formation, people's general impression of others’ moral character rapidly updates after observing a single new moral act (Siegel et al. Reference Siegel, Mathys, Rutledge and Crockett2018). During trait judgments, people retrieve behavioral exemplars in order to estimate new traits in the absence of pre-existing trait representations (Klein et al. Reference Klein, Loftus, Trafton and Fuhrman1992; Markus Reference Markus1977). During spontaneous thought, concrete perceptual thoughts cue subsequent abstract thoughts (Bar et al. Reference Bar, Aminoff, Mason and Fenske2007; Klinger Reference Klinger2013; Mildner & Tamir Reference Mildner and Tamir2019; Northoff Reference Northoff, Christoff and Fox2018). Finally, during event simulation, heightened access to specific, concrete details shifts people's evaluation of abstract event features such as valence (Jing et al. Reference Jing, Madore and Schacter2017; Madore et al. Reference Madore, Jing and Schacter2019). Thus, concrete representations alter existing abstract representations, and can even be used to generate new ones on the fly.

When the ability to successfully bridge abstract knowledge with concrete knowledge is impaired, distal simulation is unsuccessful. For example, patients with depression show deficits in retrieving concrete, specific details of positive episodic memories (Williams & Scott Reference Williams and Scott1988). These patients also show deficits in updating negative abstract representations from new, concrete instances to the contrary (Korn et al. Reference Korn, Sharot, Walter, Heekeren and Dolan2014), and in producing vivid prospections about future positive events (Gamble et al. Reference Gamble, Moreau, Tippett and Addis2019). These patients have access to abstract knowledge, but without concrete details, they are stymied in their attempts to simulate successfully. In contrast, patients with deficits in semantic memory have access to concrete episodic details (Irish & Piguet Reference Irish and Piguet2013). Yet, these patients likewise struggle to simulate novel events. Instead, they tend to simply recall past events when asked to imagine a future event (Irish et al. Reference Irish, Addis, Hodges and Piguet2012). These patients have access to concrete details, but, without abstract knowledge, they fail to recombine these details into novel simulations (Irish & Piguet Reference Irish and Piguet2013). Together, these findings suggest that successful prospection relies on the flexible interplay between abstract and concrete representations.

Although some simulations draw upon only one level of abstraction, as Gilead et al. propose, rich simulations often require a dynamic interplay between multiple levels of abstraction. Abstract representations guide the search for detailed concrete representations, and concrete representations are used to construct abstract representations. From this perspective, we interpret the overlapping brain regions associated with semantic cognition and prospection as a manifestation of such reciprocal, dynamic processes. We hope future research will build toward a more complete model of simulation, one that can accommodate a mind that dynamically bridges the perceptual with the conceptual, the episodic with the semantic, and the concrete with the abstract.