2 Introduction

Pretense relies on an ability to simulate and mimic one’s own behavior under a counterfactual belief state. For example, in order to successfully deceive your friends into thinking that you were surprised by the birthday party they threw for you, it is not sufficient that you are able to reason about their mental states (“I know that they are planning a surprise party, but they don’t know that I know that…”) — you also need to convincingly simulate and mimic your hypothetical behavior had you not known about the party (“Where would I look first had I not known? What would I say? How long would it take me to recover from the surprise?”). Similar examples abound in higher-stakes contexts such as diplomacy, warcraft and law. This is not a trivial challenge: previous research on “hindsight biases” suggests that knowledge about the actual state of the world can interfere with our ability to correctly judge what we would have believed (Fischhoff 1975, 1977; Wood 1978; Roese and Vohs 2012) or perceived (Harley, Carlsen, and Loftus 2004; Bernstein and Harley 2007; Bernstein et al. 2012) without this knowledge. Such biases remain potent even when instructing participants to overcome them (Harley, Carlsen, and Loftus 2004; Pohl and Hell 1996). Moreover, even if pretenders can correctly determine what they would have believed, they must further accurately simulate how they would think and behave in this different belief state.

The reliance of this kind of epistemic pretense on self-simulation makes it a promising tool for revealing the structure and content of people’s internal models of their own minds. When directly asked, participants are able to provide relatively accurate descriptions of their own decision-making (Morris et al. 2023) and perception (Levin and Angelone 2008; Mazor, Siegel, and Tenenbaum 2023). Pretending not to know opens a new window into the structure and content of this metacognitive knowledge, with two important advantages. First, by not relying on explicit reports, pretense has the potential to reveal implicit self-knowledge – that is, structured knowledge about the self that is not reportable. And second, data obtained from pretense experiments can be analyzed and modeled using the same tools employed by cognitive scientists to study non-pretense behavior, affording a direct and finer-grained comparison between pretend and genuine decision-making.

Our research question is whether people can mentally simulate their actions under a counterfactual knowledge state of ignorance. To that end, we had participants “pretend not to know” in a game setting. Using an online version of the games Battleship and Hangman (in which players seek to uncover the locations of enemy ships or the identity of a word), participants played a ‘non-pretend’ (normal) version of the game, as well as a ‘pretend’ version where they were given complete information about the hidden ships / target word but were instructed to behave as if they didn’t have this information. Participants’ pretense behaviour mirrored broad patterns and subtle features of real players’ decisions and decision times. At the same time, epistemic pretense was characterized by over-acting, stereotypical behavior, and suboptimal incorporation of new information: all markers of model-based simulation. Together, we take these findings as evidence for a capacity to mentally simulate decisions and actions using a simplified and schematic self-model.