An Agent Independent Axis for Executed and Modeled Choice in Medial Prefrontal Cortex

Antoinette Nicolle, Miriam C. Klein-Flügge, Laurence T. Hunt, Ivo Vlaev, Raymond J. Dolan, Timothy E.j. BehrensNeuron 2012 · DOI

TL;DR

This study reveals that the medial prefrontal cortex (mPFC) doesn't just represent "self" vs. "other" but rather distinguishes between choices that are executed (acted upon) and those that are modeled (considered but not acted upon). This "executed vs. modeled" axis is a fundamental organizational principle in the mPFC, with distinct sub-regions dynamically switching roles depending on whether a choice is for oneself or for another.

What Problem Does This Paper Address?

Understanding how the brain makes decisions is a central theme in neuroscience and economics. While we know that regions like the ventromedial prefrontal cortex (vmPFC) are involved in evaluating options and making choices based on subjective value, and the dorsomedial prefrontal cortex (dmPFC) is implicated in social cognition and inferring others' intentions, the precise computational relationship between these functions has been unclear. A key question is whether the brain has separate systems for "self" and "other" or if there's a more unified computational architecture. This paper tackles this by investigating how the brain represents value when making choices for oneself versus when making choices on behalf of another, aiming to disentangle whether the neural organization is based on the agent (self vs. other) or the nature of the computation (executed vs. modeled).

Methods

The researchers employed functional magnetic resonance imaging (fMRI) to observe brain activity in participants performing a novel intertemporal choice task. This task involved subjects choosing between a smaller reward delivered sooner and a larger reward delivered later. Crucially, in some trials, participants made choices for themselves, while in others, they made choices as if they were their assigned partner.

To ensure that the "self" and "other" value computations were as distinct as possible, participants were carefully paired: one with a "low discounter" preference (valuing future rewards more) and one with a "high discounter" preference (valuing immediate rewards more). This pairing aimed to decorrelate the subjective value signals of the two individuals.

The core analytical approach involved identifying brain regions whose activity correlated with the value difference between chosen and unchosen options. They then tested for a functional gradient along the medial wall of the prefrontal cortex, specifically examining whether this gradient aligned with a "self vs. other" axis or an "executed vs. modeled" axis. This was achieved by analyzing activity in specific brain regions under different choice conditions (self-choice vs. partner-choice) and looking for systematic shifts in neural representation.

Datasets

The study did not introduce new benchmark datasets. Instead, it utilized data collected from 20 participants (10 pairs) who underwent fMRI scanning. Participants were prescreened using an online intertemporal choice questionnaire to estimate their individual discount rates.

Key Insights and Ideas

  • Executed vs. Modeled, Not Just Self vs. Other: The most significant insight is that the medial prefrontal cortex's functional organization is primarily driven by whether a choice is executed (acted upon) or modeled (considered but not acted upon), rather than simply by whether the choice is for the self or another agent.
  • Dynamic Role Switching: vmPFC and dmPFC dynamically swap their functional roles. When making a choice for oneself, vmPFC represents the values relevant for that executed choice, while dmPFC represents the values of the partner (modeled). Conversely, when choosing for the partner, vmPFC represents the partner's executed choice values, and dmPFC represents the subject's own values (now modeled).
  • Unified Computational Machinery: The study suggests that the brain uses similar computational mechanisms (value difference calculations) for both self-referential and social inference, but these computations are implemented in distinct, yet dynamically interacting, brain regions.
  • A New Functional Axis: This research establishes a novel "executed vs. modeled" functional axis along the medial wall of the prefrontal cortex, offering a new way to understand its organization beyond the traditional self/other distinction.

Key Results

  • Gradient Analysis: A significant ventral-dorsal gradient was found in the medial prefrontal cortex, with more ventral regions (vmPFC) showing activity related to executed choice values and more dorsal regions (dmPFC) showing activity related to modeled choice values. This gradient was statistically significant (t[18] = 6.28, p < 0.00001).
  • No Self/Other Gradient: In contrast, no significant gradient was found along a "self vs. other" axis in the same regions (t[18] = -1.06, p > 0.30).
  • Role Reversal: Formal tests confirmed that vmPFC specifically tracked executed values (self values during self-choice, partner values during partner-choice), while dmPFC tracked modeled values (partner values during self-choice, self values during partner-choice). This role reversal was statistically significant.
  • Temporoparietal Cortex: Similar "executed vs. modeled" gradients and role-switching were observed in the temporoparietal cortex, suggesting this organizational principle might extend beyond the mPFC.

Limitations and Future Work

  • Computational Specificity: While the study shows that similar computational signals (value difference) are used for self and other, it doesn't fully detail the precise algorithms or how these signals are integrated.
  • Interpretation of dmPFC Activity: The authors suggest that dmPFC activity when modeling one's own irrelevant preferences might reflect projecting one's own values into the partner's mind or simulating how one's own values would influence the partner. This opens avenues for exploring higher-order belief inference and iterated reasoning.
  • Generalizability: The study used a specific intertemporal choice task. Future work could explore whether this "executed vs. modeled" axis applies to other types of decisions and social interactions.
  • Caudal dmPFC: The paper briefly mentions a more caudal region of dmPFC that shows opposing coding to vmPFC for executed value difference, suggesting further complexity within the medial prefrontal cortex that warrants deeper investigation.

Summary updated

Abstract

Adaptive success in social animals depends on an ability to infer the likely actions of others. Little is known about the neural computations that underlie this capacity. Here, we show that the brain models the values and choices of others even when these values are currently irrelevant. These modeled choices use the same computations that underlie our own choices, but are resolved in a distinct neighboring medial prefrontal brain region. Crucially, however, when subjects choose on behalf of a partner instead of themselves, these regions exchange their functional roles. Hence, regions that represented values of the subject's executed choices now represent the values of choices executed on behalf of the partner, and those that previously modeled the partner now model the subject. These data tie together neural computations underlying self-referential and social inference, and in so doing establish a new functional axis characterizing the medial wall of prefrontal cortex.
BibTeX
@inproceedings{nicolle2012agent,
  author = {Antoinette Nicolle and Miriam C. Klein-Flügge and Laurence T. Hunt and Ivo Vlaev and Raymond J. Dolan and Timothy E.J. Behrens},
  title = {{An Agent Independent Axis for Executed and Modeled Choice in Medial Prefrontal Cortex}},
  booktitle = {Neuron},
  year = {2012},
  doi = {10.1016/j.neuron.2012.07.023},
}
Figure