Gilead et al. transfer the opposition of theory versus simulation from the mindreading debate to prediction more generally. Against previous approaches, they emphasize that prediction may draw on both simulation and theory-based inference. I shall raise two interrelated issues to guide further research: (i) What is the role of simulation as distinguished from theory-based inference? (ii) How is simulation to be defined as opposed to theory?
Gilead et al. describe simulation as projecting oneself into a specific spatiotemporal context. Yet, there are doubts that simulation requires self-projection. Any simulation more or less detaches from one's present self. Complete detachment may be a boundary case. A relevant debate revolves around Berkeley's famous claim that imagining a tree requires imagining perceiving the tree (Berkeley (Reference Berkeley1710/1975), sect. 23; Noordhof Reference Noordhof2002; Peacocke Reference Peacocke, Forster and Robinson1985, pp. 22–23; Williams Reference Williams1973, p. 35). Yet, even if one agrees with Berkeley that simulating things requires to simulate them as they appear perceived from a spatiotemporal viewpoint, it does not follow that one projects oneself as occupying that viewpoint. In writing a detective story, we may imagine an unwitnessed murder without projecting ourselves into the murderer, the murderee, or adding a witness (Currie Reference Currie1995, p. 170), though there is some debate on that (Gaut Reference Gaut1997). For these reasons, I also doubt the claim that the usefulness of simulation decreases the less similar simulated others are to oneself. First, one may simulate not only persons, but also things like the trajectory of a rolling boulder (Williamson Reference Williamson2007, p. 143). Second, overall similarity of the simulated situation to familiar environments indeed tends to facilitate simulation (Strohminger & Yli-Vakkuri Reference Strohminger, Yli-Vakkuri, Benton, Hawthorne and Rabinowitz2018, pp. 318–19). Yet, this result should not be restricted to the relationship between the imagining self and simulated persons.
The imaginer may not only detach from herself, she may also largely detach from any spatiotemporal context. In imagining a particular shade of blue, one may leave the spatiotemporal location unspecified.
Gilead et al. are overly restrictive about the availability of “representative scenarios” to be used in simulation, for example in predicting the future divorce rates in Britain. The role of such scenarios in prediction may be highly indirect. Consider a lawyer who has handled many divorces over the decades but never tried to extrapolate general tendencies. Recent cases being more vivid in her memory, re-enacting them may lead her to judge that the divorce rate is on the rise. It may even be contested that theory would be a better guide here. The lawyer may have tacit experiential knowledge, for example, on the changing factors leading to divorce that is only retrievable by simulation (Mach Reference Mach1897; Williamson Reference Williamson2007, pp. 145, 170).
Gilead et al. are overly restrictive either in their demands on episodic memory, purporting simulation to be of no avail if there are no relevant episodes (e.g., “how successful will I be as a professional wrestler?”). Yet, even if there are no pertinent episodic memories, one will tend to build representative scenarios from any resources available. A natural way of addressing the wrestler issue involves imagining oneself enmeshed in a wrestling match. Even if one has no remembrances of wrestling and never watched a match, one will assemble any information available in fleshing out the scene, for example, by an analogy to tavern brawls in Dutch paintings. In sum, the demarcations considered mark tendencies at best.
Coming to the choice between theory and simulation in prediction, Gilead et al. claim that simulation is more likely to be used when theory would require complex computations, cognitive resources are depleted, or when relevant socially acquired knowledge is lacking. Moreover, simulation tends to recruit less abstract representations. This suggests that simulation is as a rule less demanding in such respects than theory. The suggestion would need additional support. Theory-based inference can be easy and simple, whereas simulation can be very complicated, effortful, and resource-intensive in terms of socially acquired knowledge. It may recruit highly abstract representations. A simple folk-sociological hypothesis “nowadays people on a first date split the check” may allow me to easily derive how to act. In contrast, a psychologist's simulation of the dating situation may partly build on her past experiences and partly on arbitrarily complex empirical theories of dating behaviour. To illustrate how complicated simulation can be: Philosophers engaged in modal epistemology have us run simulations of whole worlds, partly by descriptive means (Chalmers Reference Chalmers, Gendler and Hawthorne2002; Yablo Reference Yablo1993). I also mention computer-aided simulations in science.
In light of these considerations, I only venture a minimum characterization: The general function of simulation consists of representing a situation that is relevantly like the target situation with regards to the feature predicted. It is a matter of further research how to extend this minimum condition towards a full definition. Some tendencies may be observed: Simulation tends towards sensorimotor representations of spatiotemporally concrete “situation models” (Zwaan Reference Zwaan1999; Reference Zwaan2016), though it may recruit any mental resources, including theory (Williamson Reference Williamson2007, p. 143).
The decision whether to use simulation or theory depends on the informational and representational resources available to an individual. When choosing between two radio stations, I may apply a theory about which station plays classical music or simulate pleasurable experiences based on remembrances of listening to one station or the other, depending on which resource is more readily available. Complicated issues, as in the divorce rate and the wrestler example, will often elicit a case-specific combination of simulation and theory. It is natural to address the wrestler issue by both imagining oneself performing in a match and by drawing on theoretical knowledge (if available) about the planned routine of such events. The lawyer in the divorce rate example is well-advised to balance her tendency of simulating salient cases by consulting statistics.
In sum, the use of simulation and theory in prediction is flexible, depending on the cognitive resources available. Simulation and theory may not only be combined in prediction, they may build on each other as one resource among others. Simulation is more complex and more versatile than Gilead et al. predict it to be, and the same goes for its combination with theory.
You have
Access
Gilead et al. transfer the opposition of theory versus simulation from the mindreading debate to prediction more generally. Against previous approaches, they emphasize that prediction may draw on both simulation and theory-based inference. I shall raise two interrelated issues to guide further research: (i) What is the role of simulation as distinguished from theory-based inference? (ii) How is simulation to be defined as opposed to theory?
Gilead et al. describe simulation as projecting oneself into a specific spatiotemporal context. Yet, there are doubts that simulation requires self-projection. Any simulation more or less detaches from one's present self. Complete detachment may be a boundary case. A relevant debate revolves around Berkeley's famous claim that imagining a tree requires imagining perceiving the tree (Berkeley (Reference Berkeley1710/1975), sect. 23; Noordhof Reference Noordhof2002; Peacocke Reference Peacocke, Forster and Robinson1985, pp. 22–23; Williams Reference Williams1973, p. 35). Yet, even if one agrees with Berkeley that simulating things requires to simulate them as they appear perceived from a spatiotemporal viewpoint, it does not follow that one projects oneself as occupying that viewpoint. In writing a detective story, we may imagine an unwitnessed murder without projecting ourselves into the murderer, the murderee, or adding a witness (Currie Reference Currie1995, p. 170), though there is some debate on that (Gaut Reference Gaut1997). For these reasons, I also doubt the claim that the usefulness of simulation decreases the less similar simulated others are to oneself. First, one may simulate not only persons, but also things like the trajectory of a rolling boulder (Williamson Reference Williamson2007, p. 143). Second, overall similarity of the simulated situation to familiar environments indeed tends to facilitate simulation (Strohminger & Yli-Vakkuri Reference Strohminger, Yli-Vakkuri, Benton, Hawthorne and Rabinowitz2018, pp. 318–19). Yet, this result should not be restricted to the relationship between the imagining self and simulated persons.
The imaginer may not only detach from herself, she may also largely detach from any spatiotemporal context. In imagining a particular shade of blue, one may leave the spatiotemporal location unspecified.
Gilead et al. are overly restrictive about the availability of “representative scenarios” to be used in simulation, for example in predicting the future divorce rates in Britain. The role of such scenarios in prediction may be highly indirect. Consider a lawyer who has handled many divorces over the decades but never tried to extrapolate general tendencies. Recent cases being more vivid in her memory, re-enacting them may lead her to judge that the divorce rate is on the rise. It may even be contested that theory would be a better guide here. The lawyer may have tacit experiential knowledge, for example, on the changing factors leading to divorce that is only retrievable by simulation (Mach Reference Mach1897; Williamson Reference Williamson2007, pp. 145, 170).
Gilead et al. are overly restrictive either in their demands on episodic memory, purporting simulation to be of no avail if there are no relevant episodes (e.g., “how successful will I be as a professional wrestler?”). Yet, even if there are no pertinent episodic memories, one will tend to build representative scenarios from any resources available. A natural way of addressing the wrestler issue involves imagining oneself enmeshed in a wrestling match. Even if one has no remembrances of wrestling and never watched a match, one will assemble any information available in fleshing out the scene, for example, by an analogy to tavern brawls in Dutch paintings. In sum, the demarcations considered mark tendencies at best.
Coming to the choice between theory and simulation in prediction, Gilead et al. claim that simulation is more likely to be used when theory would require complex computations, cognitive resources are depleted, or when relevant socially acquired knowledge is lacking. Moreover, simulation tends to recruit less abstract representations. This suggests that simulation is as a rule less demanding in such respects than theory. The suggestion would need additional support. Theory-based inference can be easy and simple, whereas simulation can be very complicated, effortful, and resource-intensive in terms of socially acquired knowledge. It may recruit highly abstract representations. A simple folk-sociological hypothesis “nowadays people on a first date split the check” may allow me to easily derive how to act. In contrast, a psychologist's simulation of the dating situation may partly build on her past experiences and partly on arbitrarily complex empirical theories of dating behaviour. To illustrate how complicated simulation can be: Philosophers engaged in modal epistemology have us run simulations of whole worlds, partly by descriptive means (Chalmers Reference Chalmers, Gendler and Hawthorne2002; Yablo Reference Yablo1993). I also mention computer-aided simulations in science.
In light of these considerations, I only venture a minimum characterization: The general function of simulation consists of representing a situation that is relevantly like the target situation with regards to the feature predicted. It is a matter of further research how to extend this minimum condition towards a full definition. Some tendencies may be observed: Simulation tends towards sensorimotor representations of spatiotemporally concrete “situation models” (Zwaan Reference Zwaan1999; Reference Zwaan2016), though it may recruit any mental resources, including theory (Williamson Reference Williamson2007, p. 143).
The decision whether to use simulation or theory depends on the informational and representational resources available to an individual. When choosing between two radio stations, I may apply a theory about which station plays classical music or simulate pleasurable experiences based on remembrances of listening to one station or the other, depending on which resource is more readily available. Complicated issues, as in the divorce rate and the wrestler example, will often elicit a case-specific combination of simulation and theory. It is natural to address the wrestler issue by both imagining oneself performing in a match and by drawing on theoretical knowledge (if available) about the planned routine of such events. The lawyer in the divorce rate example is well-advised to balance her tendency of simulating salient cases by consulting statistics.
In sum, the use of simulation and theory in prediction is flexible, depending on the cognitive resources available. Simulation and theory may not only be combined in prediction, they may build on each other as one resource among others. Simulation is more complex and more versatile than Gilead et al. predict it to be, and the same goes for its combination with theory.