Internality and the internalisation of failure: Evidence from a novel task

Federico Mancinelli, Jonathan P. Roiser, Peter DayanPLOS Computational Biology 2021 · DOI · OpenReview

TL;DR

This paper introduces a novel task to investigate how individual differences in perceived control (internality) influence learning from failure. They found that individuals with higher internality learn faster from negative outcomes, leading to more adaptive decision-making and reduced vulnerability to failure, especially in challenging environments.

What Problem Does This Paper Address?

How do we learn from our mistakes? This fundamental question is crucial for adaptation and survival. While we know that people differ in how much they believe they control their outcomes (their "locus of control"), and that this belief impacts behavior, the precise mechanisms by which these internal beliefs shape learning, particularly from failure, remain unclear. Existing studies often rely on simplified tasks or self-report measures. This paper aims to bridge this gap by designing a more ecologically valid task that disentangles different aspects of control and allows for computational modeling of learning from success and failure.

Methods

The researchers developed a novel videogame task where participants had to learn to achieve goals using different vehicles with varying levels of "guidability" (microscopic control) and in environments with different levels of "influence" (effort-outcome contingency).

  • Novel Task Design: The task involved selecting and maneuvering vehicles to reach goals, with varying rewards and penalties. Key manipulations included:
  • Influence: High influence blocks meant effort (pressing keys) directly translated to better outcomes, while low influence blocks decoupled effort from reward.
  • Guidability: Vehicles differed in how reliably they moved in the intended direction.
  • Goal Choice: Participants chose between more rewarding but less attainable goals (g+) and less rewarding but more attainable goals (g-).
  • Computational Modeling: They employed a hierarchical graphical model framework to analyze participant behavior. A "Vehicle dependent RW" (Rescorla-Wagner) model emerged as the best fit, suggesting that participants learned the achievability of goals based on their past successes and failures, with separate learning rates for positive and negative outcomes.
  • Statistical Analysis: Model-agnostic analyses (correlations) and model-dependent analyses (parameter recovery, random forests) were used to link individual differences in locus of control (specifically "internality") to behavioral outcomes and model parameters.

Datasets

The study used data collected from 35 human participants performing the novel videogame task. No external datasets were used for training or evaluation.

Key Insights and Ideas

  • Internality as a "Failure-Sensitive" Learning Trait: The most striking finding is that higher internality is strongly associated with faster learning from failure. This suggests that individuals who believe they are responsible for outcomes are more attuned to negative feedback, leading them to update their expectations of achievability more rapidly.
  • Adaptive "Timidity" in Internals: Internality was correlated with a preference for less risky, more attainable goals (g-). This "timidity" wasn't necessarily about avoiding effort, but rather about making more robust choices, especially in low-influence environments where failure is more likely.
  • Influence Matters for Learning Dynamics: The "influence" manipulation significantly altered learning. In high-influence conditions, learning was more balanced between success and failure. In low-influence conditions, where outcomes were less predictable, learning from failure became even more critical and more strongly predicted internality.
  • "Chance" Beliefs Less Predictive: Interestingly, beliefs about "chance" (C-scores) were not strongly predictive of behavior or learning in this task, suggesting participants largely perceived the task as skill-based rather than luck-driven.
  • "Powerful Others" and Performance in Low Control: Exploratory analyses revealed a surprising positive correlation between "Powerful Others" (P-scores) and earnings in low-influence conditions. This suggests that when individuals feel they have little control, attributing outcomes to external powerful agents might lead to more adaptive (or less loss-prone) behavior.

Key Results

  • Negative Correlation between Internality and g+ Choice: Higher internality (I-scores) was consistently associated with a lower proportion of attempts at the riskier, more rewarding goals (g+), across both high and low influence conditions (e.g., r = -0.50, p = 0.002 in high influence).
  • Faster Learning from Failure for High Internals: High I-scorers showed a steeper decrease in perceived achievability after failures, indicating faster learning from negative outcomes. This effect was more pronounced in low-influence conditions.
  • Model Fit: The "Vehicle dependent RW" model, which incorporated separate learning rates for success and failure and considered vehicle-specific learning, best explained the data.
  • Predictors of Internality: Random forest analyses identified learning rates from failure and distance sensitivity as key predictors of internality.

Limitations and Future Work

  • Task Complexity: The authors acknowledge that the task, while more ecologically valid than previous ones, still lacks the full complexity of real-world decision-making, particularly in distinguishing between model-based and model-free learning.
  • Influence Conditions: The study used only two influence conditions (high and low). Future work could explore intermediate levels to better disentangle parameter estimates.
  • Range of Losses: The study primarily focused on a fixed loss amount. Varying losses could provide a more nuanced understanding of sensitivity to negative outcomes.
  • Clinical Populations: The authors suggest that applying this task to clinical populations could shed light on how deficits in learning from failure might contribute to disorders characterized by perseverative dysfunctional thoughts or behaviors.
  • Neural Correlates: While the paper discusses potential neural underpinnings (e.g., vmPFC, serotonin), the study itself was not designed to reveal neural substrates. Future research could investigate these links.
  • Replication of "Powerful Others" Findings: The strong correlation between P-scores and performance in low-influence conditions was unexpected and warrants replication.

Summary updated

Abstract

Author summary How success in tasks depends on our efforts is largely a feature of what is known as the controllability of the situation or environment. This quantity should determine the way we approach, adapt to, and perform tasks. In novel settings, it can only be our expectations about controllability that exert an effect, for instance determining the balance of our focus between features of achievability and potential reward; or affecting prior notions about what exactly is achievable or what we can or cannot do. All of these might be different between aversive and appetitive domains. To study these issues, we designed a novel task in which subjects have to learn about a new environment, and analyzed their behavior using a rich computational model. We found that expectations about controllability played a particularly important role in influencing learning, but in a way that differed between positive and negative outcomes. In particular, high expectations about controllability were tied to higher learning rates of unachievability given negative outcomes, guarding subjects from further loss, and preserving their subjective optimistic expectations about control. Our findings can be interpreted within a theoretical framework which implicates control expectations in individual learning differences, and fit well within modern theories of learning in aversive contexts and serotonergic function.
BibTeX
@article{mancinelli2021internality,
  author = {Federico Mancinelli and Jonathan P. Roiser and Peter Dayan},
  title = {{Internality and the internalisation of failure: Evidence from a novel task}},
  journal = {PLOS Computational Biology},
  year = {2021},
  doi = {10.1371/journal.pcbi.1009134},
}
Figure