Pretending not to know reveals a capacity for model-based self-simulation
1matan.mazor@all-souls.ox.ac.uk 2
2
Abstract
Pretending not to know requires appreciating how one would behave without a given piece of knowledge, and acting accordingly. Here, two game-based experiments reveal a capacity to simulate decision-making under such counterfactual ignorance. 1001 English-speaking adults saw the solution to a game (ship locations in Battleship, the hidden word in Hangman) but attempted to play as though they never had this information. Pretenders accurately mimicked broad aspects of genuine play, including the number of guesses required to reach a solution, as well as subtle patterns, such as effects of decision uncertainty on decision time. While peers were unable to detect pretense, statistical analysis and computational modeling uncovered traces of ‘over-acting’ in pretenders’ decisions, suggesting a schematic simulation of their minds. Opening up a new approach to studying self-simulation, our results reveal intricate metacognitive knowledge about decision-making, drawn from a rich—but simplified—internal model of cognition.
1 Research Transparency Statement
1.1 General Disclosures
The authors declare no conflicts of interest. Funding: This study was supported by an NSF BCS grant #2021053 awarded to C.F. M.M is supported by a post-doctoral research fellowship at All Souls College. Artificial intelligence: No artificial intelligence assisted technologies were used in this research or the creation of this article. Ethics: The research complied with all relevant ethical regulations and was approved by the Institutional Review Board of Johns Hopkins University.
1.2 Experiment 1 (Battleship)
Preregistration: The hypotheses and methods were preregistered on 2021-12–21, prior to data collection, and following the analysis of an independent pilot sample. A detailed pre-registration can be accessed at https://osf.io/v9zsb. The pre-registration was time-locked using cryptographic randomisation-based time-locking (Mazor, Mazor, & Mukamel, 2019) (protocol sum: 60c270410375e8a192468fc1a0e9c93da60d5e203eb2760b621a8631a26f4c5c; link to relevant lines in experimental code, making experimental randomisation causally dependent on the content of the pre-registration and thus ensuring that all data were collected after pre-registration. All pre-registered analyses are publicly available, including the report-generating R script (https://self-model.github.io/pretendingNotToKnow/docs/exp.-1-battleship.html). Exploratory analyses are flagged as such. Materials: All study materials, including demos of analysis experiments, are publicly available (https://github.com/self-model/pretendingNotToKnow). Data: All primary data are publicly available (https://osf.io/zma9b). Analysis scripts: All analysis scripts are publicly available (https://osf.io/zma9b).
1.3 Experiment 2 (Hangman)
Preregistration: The hypotheses and methods were preregistered on 2022-06–22, prior to data collection, and following the analysis of an independent pilot sample. A detailed pre-registration can be accessed at osf.io/3thry. The pre-registration was time-locked using cryptographic randomisation-based time-locking, making experimental randomisation causally dependent on the content of the pre-registration and thus ensuring that all data were collected after pre-registration (Mazor, Mazor, and Mukamel 2019). Due to an error in the experiment code, time-locking took effect only from player number 221 (batch 3) and on (protocol sum: c4929c7fe33df1b7b52f15c789d98eab30a9cee09a8121807a3c59e28e7430a4;relevant lines in experimental code. All pre-registered analyses are publicly available, including the report-generating R script (https://self-model.github.io/pretendingNotToKnow/docs/exp.-2-hangman.html). Exploratory analyses are flagged as such. Materials: All study materials, including demos of the experiments, are publicly available (https://github.com/self-model/pretendingNotToKnow). Data: All primary data are publicly available (https://osf.io/zma9b). Analysis scripts: All analysis scripts are publicly available (https://osf.io/zma9b).