Tobias Gerstenberg
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Tobias Gerstenberg has a total of 178 co-authors. Here are the top co-authors by number of shared publications:
- Joshua B. Tenenbaum
Massachusetts Institute of Technology
13 shared publications Thomas Icard
10 shared publications- Sarah A. Wu
Department of Brain & Cognitive Sciences Massachusetts Institute of Technology
9 shared publications David Rose
7 shared publicationsSiying Zhang
Department of Psychology, University of Washington, Seattle, USA
6 shared publications- Erik Brockbank
Department of Psychology, Stanford University, Stanford, USA.
5 shared publications
Latest Publications
- Open Access
- 11 October 2026
- Published by Center for Open Science in PsyArXiv
Making good decisions with counterfactuals
Making a good decision often requires knowing something counterfactual: whether a person would be better off if we acted than if we didn't. Formal work shows how both experimental and observational data ought to be combined to inform good decisions. Here, we ask whether people's intuitions follow this normative standard. In three experiments (N = 354), participants decided whether or not to act based on observational and experimental data. Participants performed better than chance but far from optimally. Trial-by-trial feedback about the consequences of their choices, including what would have happened had they chosen otherwise, didn't improve performance. The breakdown was at the inference step: participants' estimates of whether a person would benefit from the treatment, be harmed by it, or be left unaffected were almost unrelated to normative predictions. Yet, participants' decisions followed coherently from those (incorrect) estimates. Experts in statistics, economics, and data science made better decisions than novices, but their counterfactual inferences were no more accurate. Counterfactual reasoning is critical for making good decisions from experimental and observational data, but its logic doesn't come naturally to people.
- 11 October 2026
- Published by Center for Open Science in PsyArXiv
Open AccessMaking good decisions with counterfactuals
Making a good decision often requires knowing something counterfactual: whether a person would be better off if we acted than if we didn't. Formal work shows how both experimental and observational data ought to be combined to inform good decisions. Here, we ask whether people's intuitions follow this normative standard. In three experiments (N = 354), participants decided whether or not to act based on observational and experimental data. Participants performed better than chance but far from optimally. Trial-by-trial feedback about the consequences of their choices, including what would have happened had they chosen otherwise, didn't improve performance. The breakdown was at the inference step: participants' estimates of whether a person would benefit from the treatment, be harmed by it, or be left unaffected were almost unrelated to normative predictions. Yet, participants' decisions followed coherently from those (incorrect) estimates. Experts in statistics, economics, and data science made better decisions than novices, but their counterfactual inferences were no more accurate. Counterfactual reasoning is critical for making good decisions from experimental and observational data, but its logic doesn't come naturally to people.
- Open Access
- 3 September 2026
- Published by Center for Open Science in PsyArXiv
Constructing causal stories: How mental models shape memory, prediction, and generalization
How do people learn to predict what happens next? On one account, people do so by building mental models that mirror aspects of the causal structure of the world. Accordingly, people tell a story of how the data was generated, focusing on goal-relevant information. On another account, people make predictions by learning simple mappings from relevant features of the situation to the outcome. Here, we provide evidence for the causal account. Across three experiments and two paradigms, we find that people misremember what happened, predict incorrectly what will happen, and generalize to novel situations in a way that's more consistent with the causal account than a feature-based alternative. People spontaneously construct causal models that compress experience to privilege causally relevant information. These models organize how we remember the past, predict the future, and generalize to novel situations.
- 3 September 2026
- Published by Center for Open Science in PsyArXiv
Open AccessConstructing causal stories: How mental models shape memory, prediction, and generalization
How do people learn to predict what happens next? On one account, people do so by building mental models that mirror aspects of the causal structure of the world. Accordingly, people tell a story of how the data was generated, focusing on goal-relevant information. On another account, people make predictions by learning simple mappings from relevant features of the situation to the outcome. Here, we provide evidence for the causal account. Across three experiments and two paradigms, we find that people misremember what happened, predict incorrectly what will happen, and generalize to novel situations in a way that's more consistent with the causal account than a feature-based alternative. People spontaneously construct causal models that compress experience to privilege causally relevant information. These models organize how we remember the past, predict the future, and generalize to novel situations.
- Open Access
- 30 August 2026
- Published by Center for Open Science in PsyArXiv
Anticipating the risks and benefits of counterfactual world simulation models
This paper examines the transformative potential of Counterfactual World Simulation Models (CWSMs). CWSMs use pieces of multi-modal evidence, such as the CCTV footage or sound recordings of a road accident, to build a high-fidelity 3D reconstruction of the scene. They can also answer causal questions, such as whether the accident happened because the driver was speeding, by simulating what would have happened in relevant counterfactual situations. CWSMs will enhance our capacity to envision alternate realities and investigate the outcomes of counterfactual alterations to how events unfold. This also, however, raises questions about what alternative scenarios we should be considering and what to do with that knowledge. We present a normative and ethical framework that guides and constrains the simulation of counterfactuals. We address the challenge of ensuring fidelity in reconstructions while simultaneously preventing stereotype perpetuation during counterfactual simulations. We anticipate different modes of how users will interact with CWSMs and discuss how their outputs may be presented. Finally, we address the prospective applications of CWSMs in the legal domain, recognizing both their potential to revolutionize legal proceedings as well as the ethical concerns they engender. Anticipating a new type of AI, this paper seeks to illuminate a path forward for responsible and effective use of CWSMs.
- 30 August 2026
- Published by Center for Open Science in PsyArXiv
Open AccessAnticipating the risks and benefits of counterfactual world simulation models
This paper examines the transformative potential of Counterfactual World Simulation Models (CWSMs). CWSMs use pieces of multi-modal evidence, such as the CCTV footage or sound recordings of a road accident, to build a high-fidelity 3D reconstruction of the scene. They can also answer causal questions, such as whether the accident happened because the driver was speeding, by simulating what would have happened in relevant counterfactual situations. CWSMs will enhance our capacity to envision alternate realities and investigate the outcomes of counterfactual alterations to how events unfold. This also, however, raises questions about what alternative scenarios we should be considering and what to do with that knowledge. We present a normative and ethical framework that guides and constrains the simulation of counterfactuals. We address the challenge of ensuring fidelity in reconstructions while simultaneously preventing stereotype perpetuation during counterfactual simulations. We anticipate different modes of how users will interact with CWSMs and discuss how their outputs may be presented. Finally, we address the prospective applications of CWSMs in the legal domain, recognizing both their potential to revolutionize legal proceedings as well as the ethical concerns they engender. Anticipating a new type of AI, this paper seeks to illuminate a path forward for responsible and effective use of CWSMs.
- Open Access
- 17 June 2026
- Published by Center for Open Science in PsyArXiv
“I would not have done that”: Outcome knowledge distorts memory for decisions made minutes ago
Remembering one’s past decisions is critical for learning from delayed feedback and maintaining an accurate self-model. Across two gamified experimental paradigms and three experiments (total N = 500), we find that a selective forgetting of unsuccessful guesses operates within minutes, that it evades introspection, and that it affects confidence ratings in memory judgments. Analysis of memory errors reveals that, in reconstructing their past decisions, people integrate what they remember deciding, what they believe they would have decided, and what they would decide now if asked to make the same decision again.
- 17 June 2026
- Published by Center for Open Science in PsyArXiv
Open Access“I would not have done that”: Outcome knowledge distorts memory for decisions made minutes ago
Remembering one’s past decisions is critical for learning from delayed feedback and maintaining an accurate self-model. Across two gamified experimental paradigms and three experiments (total N = 500), we find that a selective forgetting of unsuccessful guesses operates within minutes, that it evades introspection, and that it affects confidence ratings in memory judgments. Analysis of memory errors reveals that, in reconstructing their past decisions, people integrate what they remember deciding, what they believe they would have decided, and what they would decide now if asked to make the same decision again.
- Open Access
- 22 May 2026
- Published by Center for Open Science in PsyArXiv
Seeing, hearing, and feeling causation
How do people decide whether one event caused another? While previous research has focused on visual and auditory cues in causal perception, the role of touch remains underexplored. Here, we investigate how haptic feedback contributes to causal judgments across three psychophysical experiments. In Experiment 1, we introduced force-based haptic feedback to a visual launching paradigm and found that haptic information increased causal judgments compared to visual feedback alone. Experiment 2 combined vision, audio, force-based haptics, and vibrotactile haptics, revealing that additional sensory cues increase causal judgments with diminishing returns—the largest benefit comes from adding a second modality. In Experiment 3, we examined how both the number and physical realism of multisensory cues affect causal perception, finding that both factors boosted causal judgments. We present a Bayesian multimodal inference model that captures human judgments by integrating visual, auditory, and haptic information based on their relative timing, uncertainty, and realism. Taken together, these experiments show that haptic information contributes to causal judgments by shaping the multisensory evidence observers use when deciding whether one event physically caused another, including how realistic and physically coherent the event appears. More broadly, we find that temporal alignment and cross-modal coherence are key constraints for how multisensory evidence shapes causal judgments, with implications for virtual reality, robotics, and human-computer interaction systems.
- 22 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessSeeing, hearing, and feeling causation
How do people decide whether one event caused another? While previous research has focused on visual and auditory cues in causal perception, the role of touch remains underexplored. Here, we investigate how haptic feedback contributes to causal judgments across three psychophysical experiments. In Experiment 1, we introduced force-based haptic feedback to a visual launching paradigm and found that haptic information increased causal judgments compared to visual feedback alone. Experiment 2 combined vision, audio, force-based haptics, and vibrotactile haptics, revealing that additional sensory cues increase causal judgments with diminishing returns—the largest benefit comes from adding a second modality. In Experiment 3, we examined how both the number and physical realism of multisensory cues affect causal perception, finding that both factors boosted causal judgments. We present a Bayesian multimodal inference model that captures human judgments by integrating visual, auditory, and haptic information based on their relative timing, uncertainty, and realism. Taken together, these experiments show that haptic information contributes to causal judgments by shaping the multisensory evidence observers use when deciding whether one event physically caused another, including how realistic and physically coherent the event appears. More broadly, we find that temporal alignment and cross-modal coherence are key constraints for how multisensory evidence shapes causal judgments, with implications for virtual reality, robotics, and human-computer interaction systems.
- Open Access
- 19 May 2026
- Published by Center for Open Science in PsyArXiv
Young children use mental simulation to reason about their performance
Young children can use observed performance outcomes (e.g., successes and failures) to decide when to persist and which tasks to pursue. However, learners often face situations where outcomes either cannot be observed or are uninformative. Using a simple tablet game, we asked whether preschool-aged children can use mentally simulated outcomes to guide their decisions. In Experiment 1, the game froze mid-trial so the final outcome was unavailable. Children preferred to repeat the same game when their attempt would have resulted in success (versus failure). In Experiment 2, an on-screen agent intervened mid-trial, rendering the outcome uninformative about the child's performance. Children preferred to play the game without (versus with) the agent when their attempts would have been successful without the agent's intervention. These findings suggest that children can simulate alternative outcomes of their actions and use them to guide how they pursue future tasks.
- 19 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessYoung children use mental simulation to reason about their performance
Young children can use observed performance outcomes (e.g., successes and failures) to decide when to persist and which tasks to pursue. However, learners often face situations where outcomes either cannot be observed or are uninformative. Using a simple tablet game, we asked whether preschool-aged children can use mentally simulated outcomes to guide their decisions. In Experiment 1, the game froze mid-trial so the final outcome was unavailable. Children preferred to repeat the same game when their attempt would have resulted in success (versus failure). In Experiment 2, an on-screen agent intervened mid-trial, rendering the outcome uninformative about the child's performance. Children preferred to play the game without (versus with) the agent when their attempts would have been successful without the agent's intervention. These findings suggest that children can simulate alternative outcomes of their actions and use them to guide how they pursue future tasks.
- Open Access
- 15 May 2026
- Published by Center for Open Science in PsyArXiv
Causal judgments in complex situations
The Counterfactual Simulation Model (CSM) proposes that people make causal judgments by mentally simulating what would have happened in relevant counterfactual scenarios. The model assumes causal judgments incorporate two aspects: whether-causation (would the outcome have occurred without the candidate cause?) and how-causation (did the candidate cause affect how the outcome occurred?). The CSM successfully predicts judgments in simple billiard ball scenarios (2 or 3 balls). Here, we explore whether it captures people's judgments in more complex situations (up to 12 balls). We measured whether-causation (Experiment 1), how-causation (Experiment 2), and causal judgments (Experiment 3) for billiard ball scenarios varying in complexity (2-12 balls). A model combining both whether- and how-causation best predicted causal judgments, though a how-causation only model performed almost as well. However, the CSM's whether-causation predictions diverged from human judgments in complex scenarios, while how-causation predictions remained robust. This suggests that when participants make causal judgments, they might shift from counterfactual simulation toward simpler force-tracking heuristics as complexity increases.
- 15 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessCausal judgments in complex situations
The Counterfactual Simulation Model (CSM) proposes that people make causal judgments by mentally simulating what would have happened in relevant counterfactual scenarios. The model assumes causal judgments incorporate two aspects: whether-causation (would the outcome have occurred without the candidate cause?) and how-causation (did the candidate cause affect how the outcome occurred?). The CSM successfully predicts judgments in simple billiard ball scenarios (2 or 3 balls). Here, we explore whether it captures people's judgments in more complex situations (up to 12 balls). We measured whether-causation (Experiment 1), how-causation (Experiment 2), and causal judgments (Experiment 3) for billiard ball scenarios varying in complexity (2-12 balls). A model combining both whether- and how-causation best predicted causal judgments, though a how-causation only model performed almost as well. However, the CSM's whether-causation predictions diverged from human judgments in complex scenarios, while how-causation predictions remained robust. This suggests that when participants make causal judgments, they might shift from counterfactual simulation toward simpler force-tracking heuristics as complexity increases.
- Open Access
- 13 May 2026
- Published by Center for Open Science in PsyArXiv
Asking the right questions? What people learn about strangers in conversation
When meeting somebody for the first time, how do we get to know them? In the current work, we investigate how people learn about others’ personalities through the questions they ask in conversation. Across two studies, participants completed a personality inventory then were paired with an online partner for a ten-minute chat. They were either instructed to get to know their partner in freeform conversation or were provided questions to discuss. The questions were either informative or uninformative for getting to know a stranger. Participants completed the same personality inventory about their partner afterwards. We test whether choosing from informative questions enabled participants to form a more accurate impression of their partner. We find that freeform conversation improved personality predictions overall, but differences in the informativeness of the questions discussed had minimal effects on accuracy; deep questions may only be as good as the disclosures they elicit.
- 13 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessAsking the right questions? What people learn about strangers in conversation
When meeting somebody for the first time, how do we get to know them? In the current work, we investigate how people learn about others’ personalities through the questions they ask in conversation. Across two studies, participants completed a personality inventory then were paired with an online partner for a ten-minute chat. They were either instructed to get to know their partner in freeform conversation or were provided questions to discuss. The questions were either informative or uninformative for getting to know a stranger. Participants completed the same personality inventory about their partner afterwards. We test whether choosing from informative questions enabled participants to form a more accurate impression of their partner. We find that freeform conversation improved personality predictions overall, but differences in the informativeness of the questions discussed had minimal effects on accuracy; deep questions may only be as good as the disclosures they elicit.
- Open Access
- 13 May 2026
- Published by Center for Open Science in PsyArXiv
Explaining Order Effects in Counterfactual Reasoning
Human reasoning is often affected by order, such that later judgments depend on earlier ones. For example, order effects have been observed when people answer counterfactual questions. The order in which questions are asked affects whether participants backtrack -- imagining how variables that are upstream of a counterfactual change might have been different. Some counterfactual theories predict backtracking while others do not. We build on this prior work and develop a computational model of counterfactual reasoning that captures order effects. We evaluate different versions of this model on existing empirical results, finding that a model which backtracks and produces systematic order effects best explains human judgments. We conclude by discussing implications of the finding that counterfactual reasoning requires a context-sensitive evidence source for both resource-rational and discourse-coherent explanations of reasoning.
- 13 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessExplaining Order Effects in Counterfactual Reasoning
Human reasoning is often affected by order, such that later judgments depend on earlier ones. For example, order effects have been observed when people answer counterfactual questions. The order in which questions are asked affects whether participants backtrack -- imagining how variables that are upstream of a counterfactual change might have been different. Some counterfactual theories predict backtracking while others do not. We build on this prior work and develop a computational model of counterfactual reasoning that captures order effects. We evaluate different versions of this model on existing empirical results, finding that a model which backtracks and produces systematic order effects best explains human judgments. We conclude by discussing implications of the finding that counterfactual reasoning requires a context-sensitive evidence source for both resource-rational and discourse-coherent explanations of reasoning.
- Open Access
- 11 May 2026
- Published by Center for Open Science in PsyArXiv
What You See is What You Regret
Regret is a counterfactual emotion that arises when one judges that an outcome would have been better if one had acted differently. Classic accounts distinguish between commission regret (regretting an action) and omission regret (regretting an inaction). Here, we re-examine this distinction through the lens of attention and counterfactual salience. We find that externally highlighting subsets of options reliably increased omission and commission regret even after action. Increasing the size of the salient set attenuated regret intensity. When participants actively constructed their own consideration sets, the same patterns emerged. These findings suggest that commission--omission asymmetries may depend less on action versus inaction per se, and more on how attention shapes the space of salient counterfactuals, with implications for when regret might support adaptive learning or disrupt it.
- 11 May 2026
- Published by Center for Open Science in PsyArXiv
Open AccessWhat You See is What You Regret
Regret is a counterfactual emotion that arises when one judges that an outcome would have been better if one had acted differently. Classic accounts distinguish between commission regret (regretting an action) and omission regret (regretting an inaction). Here, we re-examine this distinction through the lens of attention and counterfactual salience. We find that externally highlighting subsets of options reliably increased omission and commission regret even after action. Increasing the size of the salient set attenuated regret intensity. When participants actively constructed their own consideration sets, the same patterns emerged. These findings suggest that commission--omission asymmetries may depend less on action versus inaction per se, and more on how attention shapes the space of salient counterfactuals, with implications for when regret might support adaptive learning or disrupt it.