Human neurocomputational mechanisms of guilt-driven and shame-driven altruistic behavior
- Department of Psychology, Sun Yat-sen University, China
- Department of Psychology, The Ohio State University, United States
- Key Laboratory of Brain, Cognition and Education Sciences, Ministry of Education, School of Psychology, Center for Studies of Psychological Application, Guangdong Key Laboratory of Mental Health and Cognitive Science, South China Normal University, China
- State Key Laboratory of Cognitive Neuroscience and Learning, and IDG/McGovern Institute for Brain Research, Beijing Normal University, China
- Beijing Key Laboratory of Safe AI and Superalignment, China
- Beijing Institute of AI Safety and Governance, China
- Brain-inspired Cognitive AI Lab, Institute of Automation, Chinese Academy of Sciences, China
- University of Chinese Academy of Sciences, China
- Long-term AI, China
Figures
Figure 1
Experimental procedure.
(A) Timeline for the whole experiment. (B) Timeline for the interpersonal game. In this example, the participant, who was one of the four deciders, made an incorrect estimation, as did one of the other deciders. The participants subsequently decided to allocate three monetary tokens to the receiver. ISI, inter-stimulus interval.
Figure 2
Behavioral results.
(A) Harm had a stronger effect on guilt than on shame. (B) Responsibility revealed a stronger effect on shame than on guilt. (A, B) We created regression equations for guilt and shame ratings using the coefficient estimates from the linear mixed-effect regression analyses. To illustrate the impacts of harm and responsibility, we visualized the regression lines based on these equations. (C) The coefficient estimates from the linear mixed-effect regression model showed that, compared with shame, guilt exerted a larger effect on compensation. (D) Participants’ average guilt and shame feelings showed no significant difference. (C, D) Data are shown as the mean ± standard error with overlaid dot plots. *p < 0.05, ***p < 0.001; NS, not significant.
Figure 3
Computational modeling results.
(A, B) Model simulations reproduced the behavioral patterns of compensatory decisions as influenced by harm (A) and responsibility (B). Data are shown as the mean ± standard error with overlaid dot plots. (C) The compensatory sensitivity (κ) and compensatory baseline (η) had no significant correlation. The line represents the least squares fit with shading showing the 95% confidence interval.
Figure 4
Neural representation of cognitive antecedents and neural basis of emotion sensitivity.
(A) The quotient of harm divided by the number of wrongdoers (i.e. average harm per person) is represented by the striatum and posterior insula (pINS). (B) The neural responses to average harm per person in the two clusters containing temporoparietal junction (TPJ)/ superior temporal sulcus (STS) and precentral cortex (PRC)/postcentral cortex (POC)/ supplementary motor area (SMA) were negatively correlated with responsibility-driven shame sensitivity. (A, B) Negative T values indicate negative correlation. Whole-brain FWE-cluster correction at p<0.05 after cluster-forming-threshold at p<0.001. (C) The TPJ/STS and PRC/POC/SMA clusters showed a significantly stronger negative correlation with shame-driven sensitivity than with guilt-driven sensitivity. Each line represents the least squares fit with shading showing the 95% confidence interval.
Figure 5
Neural basis of compensatory sensitivity.
(A) The dorsomedial prefrontal cortex (DMPFC), supplementary motor area (SMA), and temporal pole (TP) showed significant activity associated with guilt-driven compensatory sensitivity. A red circle marked the region that showed significant activity associated with guilt-driven compensatory sensitivity but not with shame-driven compensatory sensitivity. (B) The DMPFC, SMA, right and left inferior parietal lobe (IPL), and left lateral prefrontal cortex (LPFC) showed significant activity associated with shame-driven compensatory sensitivity. Blue circles marked the region showed significant activity associated with shame-driven compensatory sensitivity but not with guilt-driven compensatory sensitivity. (C) The left LPFC showed a marginally stronger positive correlation with shame-driven sensitivity than with guilt-driven sensitivity. Each line represents the least squares fit with shading showing the 95% confidence interval. (A, B, D) The neural correlates of the parameter κ largely overlapped with the regions linked to guilt-driven and shame-driven compensatory sensitivities. The parameter κ is additionally associated with activation in the bilateral anterior insula (aINS). r, right; l, left; Whole-brain FWE-cluster correction at p<0.05 after cluster-forming-threshold at p<0.001.
Figure 6
Neural correlates of trait guilt and compensation.
(A) A small-volume correction analysis showed that participants with higher guilt trait scores (i.e., repair action tendencies) have more positive parametric responses to the quotient of harm divided by the number of wrongdoers in the anterior middle cingulate cortex (aMCC). The scatter plot is for presenting the positive correlation relationship between repair action tendencies and neural responses of the aMCC. The line represents the least squares fit with shading showing the 95% confidence interval. (B) The neural finding of the aMCC remained significant after whole-brain correction. Whole-brain FWE-cluster correction at p<0.05 after cluster-forming-threshold at p<0.001. (A, B) A circle of dots indicated the position of the aMCC mask. (C) The aMCC parametric responses mediated the relationship between repair action tendencies and compensation. β, path coefficient; CI, 95% confidence interval of β; bold font, significant indirect effect.
Appendix 2—figure 1
Manipulation checks.
(A) The differences in pain ratings across different harm levels were all significant. (B) The differences in responsibility ratings across different responsibility levels were all significant. ***p < 0.001.
Tables
Key resources table
| Reagent type (species) or resource | Designation | Source or reference | Identifiers | Additional information |
|---|---|---|---|---|
| Software, algorithm | MATLAB | MATLAB | RRID:SCR_001622 | |
| Software, algorithm | SPM 12 | SPM 12 | RRID:SCR_007037 | |
| Software, algorithm | R Project for Statistical Computing | R Project for Statistical Computing | RRID:SCR_001905 |
Appendix 3—table 1
Means (and standard deviations) of participants’ pain ratings across different harm levels.
| Conditions | Pain ratings |
|---|---|
| Harm level 1 | 2.52 (0.94) |
| Harm level 2 | 4.14 (1.07) |
| Harm level 3 | 6.14 (1.34) |
| Harm level 4 | 7.81 (1.38) |
Appendix 3—table 2
Means (and standard deviations) of participants’ responsibility ratings across different responsibility levels.
| Conditions | Responsibility ratings |
|---|---|
| Making correct estimates | 2.02 (1.37) |
| Responsibility level 1 | 3.98 (1.89) |
| Responsibility level 2 | 4.90 (1.74) |
| Responsibility level 3 | 6.02 (1.54) |
| Responsibility level 4 | 8.36 (1.28) |
Appendix 3—table 3
Linear mixed-effect regression results for emotion ratings by harm and emotion type.
| Fixed-effect regressor | Estimate | SE | T (df) | p |
|---|---|---|---|---|
| Harm | 0.231 | 0.083 | 2.77 (67) | 0.007 |
| Emotion type | –1.259 | 0.327 | 3.85 (41) | <0.001 |
| Harm ×Emotion type | 0.512 | 0.086 | 5.96 (41) | <0.001 |
| Intercept | 3.738 | 0.358 | 10.45 (61) | <0.001 |
-
Note: Dependent variable: emotion ratings; SE: standard error; df: degrees of freedom; Regarding emotion type, ‘shame’ was defined as the reference category.
Appendix 3—table 4
Linear mixed-effect regression results for emotion ratings by responsibility and emotion type.
| Fixed-effect regressor | Estimate | SE | T (df) | p |
|---|---|---|---|---|
| Responsibility | 0.927 | 0.109 | 8.54 (61) | <0.001 |
| Emotion type | 0.903 | 0.366 | 2.47 (41) | 0.018 |
| Responsibility ×Emotion type | –0.355 | 0.099 | 3.60 (41) | <0.001 |
| Intercept | 1.996 | 0.365 | 5.47 (66) | <0.001 |
-
Note: Dependent variable: emotion ratings; SE: standard error; df: degrees of freedom; Regarding emotion type, ‘shame’ was defined as the reference category.
Appendix 3—table 5
Linear mixed-effect regression results for emotion ratings by harm, responsibility, and emotion type.
| Fixed-effect regressor | Estimate | SE | T (df) | p |
|---|---|---|---|---|
| Harm | –0.079 | 0.136 | 0.58 (78) | 0.564 |
| Responsibility | 0.615 | 0.167 | 3.69 (74) | <0.001 |
| Emotion type | –0.157 | 0.576 | 0.274 (68) | 0.785 |
| Harm ×Responsibility | 0.124 | 0.052 | 2.39 (81) | 0.019 |
| Harm ×Emotion type | 0.423 | 0.171 | 2.47 (68) | 0.016 |
| Responsibility ×Emotion type | –0.443 | 0.194 | 2.29 (63) | 0.026 |
| Harm ×Responsibility × Emotion type | 0.036 | 0.069 | 0.53 (71) | 0.601 |
| Intercept | 2.198 | 0.434 | 5.06 (81) | <0.001 |
-
Note: Dependent variable: emotion ratings; SE: standard error; df: degrees of freedom; Regarding emotion type, ‘shame’ was defined as the reference category.
Appendix 3—table 6
Linear mixed-effect regression models of compensation.
| Model | Dependent variable | Fixed-effect regressors | BIC value |
|---|---|---|---|
| I | Compensation | Guilt | 9136 |
| II | Compensation | Shame | 9420 |
| III | Compensation | Guilt and shame | 9116* |
| IV | Compensation | Guilt, shame, and guilt ×shame | 9147 |
-
Note: BIC, Bayesian information criterion.
-
*
Best model.
Appendix 3—table 7
Linear mixed-effect regression results for compensation by guilt and shame.
| Fixed-effect regressor | Estimate | SE | T (df) | p |
|---|---|---|---|---|
| Guilt | 0.215 | 0.042 | 5.16 (41) | <0.001 |
| Shame | 0.095 | 0.026 | 3.69 (36) | <0.001 |
| Intercept | 2.988 | 0.316 | 9.47 (43) | <0.001 |
-
Note: Dependent variable: compensation; SE: standard error; df: degrees of freedom.
Appendix 3—table 8
Model comparison for compensatory behavior.
| Model | Brief model description | BIC values |
|---|---|---|
| Model 1.1 | Integration in the form of a quotient, self-interest, and compensation baseline | 8095 |
| Model 1.2 | Integration in the form of a quotient and self-interest | 10566 |
| Model 1.3 | Integration in the form of a quotient and compensation baseline | 8001* |
| Model 1.4 | Integration in the form of a quotient | 10408 |
| Model 2.1 | Integration in the form of a product, self-interest, and compensation baseline | 8138 |
| Model 2.2 | Integration in the form of a product and self-interest | 10472 |
| Model 2.3 | Integration in the form of a product and compensation baseline | 8046 |
| Model 2.4 | Integration in the form of a product | 10314 |
-
Note: BIC, Bayesian information criterion.
-
*
Best model.
Appendix 3—table 9
Brain regions responding parametrically to the level of harm and the number of wrongdoers during outcome evaluation (results based on GLM 1).
| Modulator/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Modulator: H | ||||||
| None | ||||||
| Modulator: W | ||||||
| Precentral and postcentral cortex | 484 | –21 | –27 | 66 | 4.70 | <0.001 |
| Precentral and postcentral cortex | 168 | 21 | –33 | 66 | 4.23 | 0.023 |
-
Note: H denotes the level of harm; W denotes the number of wrongdoers; MNI, Montreal Neurological Institute; FWE, family-wise error correction.
Appendix 3—table 10
Brain regions responding parametrically to the level of harm and the number of wrongdoers during outcome evaluation (results based on GLMs 2 and 3).
| Modulator/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Modulator: H | ||||||
| None | ||||||
| Modulator: W | ||||||
| Precentral and postcentral cortex | 514 | –21 | –27 | 66 | 4.70 | <0.001 |
| Precentral and postcentral cortex | 171 | 21 | –33 | 66 | 4.22 | 0.022 |
-
Note: H denotes the level of harm; W denotes the number of wrongdoers; MNI, Montreal Neurological Institute; FWE, family-wise error correction.
Appendix 3—table 11
Brain regions responding parametrically to the quotient of the level of harm divided by the number of wrongdoers during outcome evaluation (results based on GLM 4).
| Modulator/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Modulator: | ||||||
| Striatum | 545 | 24 | -6 | 21 | –6.84 | <0.001 |
| Posterior insula | 181 | –48 | -3 | 15 | –4.43 | 0.021 |
-
Note: denotes the quotient of the level of harm divided by the number of wrongdoers. MNI, Montreal Neurological Institute; FWE, family-wise error correction. Negative T values indicate that the activity in the brain region is inversely related to the modulator.
Appendix 3—table 12
Neural activity in the brain regions correlated with the parameter estimates of responsibility-driven shame sensitivity.
| Covariate/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Covariate: responsibility-driven shame sensitivity | ||||||
| TPJ/STS | 386 | –69 | –24 | 12 | 4.68 | 0.001 |
| PRC/POC/SMA | 2372 | 15 | –30 | 75 | 6.27 | <0.001 |
-
Note: TPJ, temporoparietal junction; STS, superior temporal sulcus; PRC, precentral cortex; POC, postcentral cortex; SMA, supplementary motor area. MNI, Montreal Neurological Institute; FWE, family-wise error correction.
Appendix 3—table 13
The differences in correlations between the brain activities and different types of emotion sensitivity.
| Brain region | Emotion sensitivity | Correlation 1 | Correlation 2 | Correlation Difference | |||||
|---|---|---|---|---|---|---|---|---|---|
| R | p | R | p | Z | p | ||||
| TPJ/STS | RDSS vs. RDGS | –0.51 | 0.001 | 0.04 | 0.797 | 2.44 | 0.015 | ||
| TPJ/STS | RDSS vs. HDSS | –0.51 | 0.001 | 0.21 | 0.211 | 3.38 | <0.001 | ||
| TPJ/STS | RDSS vs. HDGS | –0.51 | 0.001 | –0.17 | 0.308 | 1.87 | 0.062 | ||
| PRC/POC/SMA | RDSS vs. RDGS | –0.68 | <0.001 | 0.20 | 0.206 | 4.36 | <0.001 | ||
| PRC/POC/SMA | RDSS vs. HDSS | –0.68 | <0.001 | 0.12 | 0.457 | 4.17 | <0.001 | ||
| PRC/POC/SMA | RDSS vs. HDGS | –0.68 | <0.001 | –0.15 | 0.354 | 3.30 | 0.001 | ||
-
Note: TPJ, temporoparietal junction; STS, superior temporal sulcus; PRC, precentral cortex; POC, postcentral cortex; SMA, supplementary motor area; RDSS, responsibility-driven shame sensitivity; RDGS, responsibility-driven guilt sensitivity; HDSS, harm-driven shame sensitivity; HDGS, harm-driven guilt sensitivity. During the leave-one-subject-out cross-validation procedure, TPJ/STS and PRC/POC/SMA merged into a single inseparable cluster in two folds, and no suprathreshold cluster was detected within TPJ/STS in another fold. These cases were coded as missing, resulting in 39 participants for TPJ/STS and 40 participants for PRC/POC/SMA.
Appendix 3—table 14
Neural activity in the brain regions correlated with the parameter estimates of guilt-driven or shame-driven compensatory sensitivities.
| Covariate/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Covariate: guilt-driven compensatory sensitivity | ||||||
| DMPFC/SMA | 394 | 12 | 21 | 57 | 5.02 | <0.001 |
| TP | 220 | –57 | 6 | –27 | 4.99 | 0.005 |
| lIPL | 131 | –36 | –60 | 54 | 4.44 | 0.036 |
| Covariate: shame-driven compensatory sensitivity | ||||||
| DMPFC/SMA | 783 | -9 | 21 | 48 | 5.28 | <0.001 |
| lLPFC | 234 | –33 | 51 | 12 | 4.54 | 0.003 |
| rIPL | 285 | 36 | –54 | 54 | 5.32 | 0.001 |
| lIPL | 283 | –33 | –51 | 39 | 5.25 | 0.001 |
-
Note: DMPFC, dorsomedial prefrontal cortex; LPFC, lateral prefrontal cortex; SMA, supplementary motor area; TP, temporal pole; IPL, inferior parietal lobe; ITC, inferior temporal cortex. r, right; l, left; MNI, Montreal Neurological Institute; FWE, family-wise error correction.
Appendix 3—table 15
The differences in correlations between the brain activity and different types of compensatory sensitivity.
| Brain region | Emotion sensitivity | Correlation 1 | Correlation 2 | Correlation Difference | |||||
|---|---|---|---|---|---|---|---|---|---|
| R | p | R | p | Z | p | ||||
| rIPL | GDCS vs. SDCS | 0.43 | 0.005 | 0.56 | <0.001 | 1.01 | 0.312 | ||
| lIPL | GDCS vs. SDCS | 0.42 | 0.005 | 0.47 | 0.002 | 0.37 | 0.709 | ||
| lLPFC | GDCS vs. SDCS | 0.15 | 0.335 | 0.44 | 0.004 | 1.93 | 0.053 | ||
-
Note: IPL, inferior parietal lobe; LPFC, lateral prefrontal cortex; r, right; l, left; GDCS, guilt-driven compensatory sensitivity; SDCS, shame-driven compensatory sensitivity.
Appendix 3—table 16
Linear regression results for parameter κ by guilt-driven and shame-driven compensatory sensitivities.
| Fixed-effect regressor | Estimate | SE | T (df) | p |
|---|---|---|---|---|
| Guilt-driven compensatory sensitivity | 2.062 | 0.352 | 5.85 (38) | <0.001 |
| Shame-driven compensatory sensitivity | 2.426 | 0.865 | 2.82 (38) | 0.008 |
| Intercept | –0.048 | 0.107 | –0.45 (38) | 0.653 |
-
Note: Dependent variable: compensation; SE: standard error; df: degrees of freedom.
Appendix 3—table 17
Neural activity in the brain regions correlated with the parameter estimates of κ.
| Covariate/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Covariate: κ | ||||||
| DMPFC/SMA/LPFC | 1702 | 54 | 36 | 27 | 5.70 | <0.001 |
| aINS | 173 | 33 | 18 | 6 | 4.98 | 0.013 |
| TP/aINS | 771 | –54 | 9 | –30 | 5.31 | <0.001 |
| rIPL | 283 | 36 | –54 | 51 | 5.39 | 0.001 |
| lIPL | 249 | –33 | –57 | 57 | 5.22 | 0.003 |
| ITC | 215 | 57 | –48 | –18 | 5.86 | 0.005 |
-
Note: DMPFC, dorsomedial prefrontal cortex; SMA, supplementary motor area; LPFC, lateral prefrontal cortex; aINS, anterior insula; TP, temporal pole; IPL, inferior parietal lobe; ITC, inferior temporal cortex. r, right; l, left; MNI, Montreal Neurological Institute; FWE, family-wise error correction.
Appendix 3—table 18
The brain regions whose responses to the quotient of harm divided by the number of wrongdoers correlated with repair action tendencies (i.e., scores of a trait guilt subscale).
| Covariate/Region | Cluster Size | Peak MNI Coordinates | T | PFWE | ||
|---|---|---|---|---|---|---|
| x | y | z | ||||
| Covariate: repair action tendencies (trait guilt) | ||||||
| aMCC* | 12 | -3 | 30 | 15 | 4.21 | 0.003 |
| aMCC/DMPFC/SMA/precuneus | 4003 | 3 | 54 | 24 | 5.55 | <0.001 |
| LPFC | 166 | 33 | 27 | 36 | 4.37 | 0.021 |
| aINS | 418 | –30 | 15 | -9 | 4.93 | <0.001 |
| rTP/ITC | 182 | 48 | –12 | –27 | 4.69 | 0.015 |
| lTP/ITC | 241 | –42 | -6 | –33 | 4.67 | 0.005 |
| IPL/TPJ | 141 | 51 | –39 | 18 | 4.63 | 0.035 |
-
Note: aMCC, anterior middle cingulate cortex; DMPFC, dorsomedial prefrontal cortex; SMA, supplementary motor area; LPFC, lateral prefrontal cortex; aINS, anterior insula; TP, temporal pole; ITC, inferior temporal cortex; IPL, inferior parietal lobe; TPJ, temporoparietal junction. r, right; l, left; MNI, Montreal Neurological Institute; FWE, family-wise error correction. *small-volume correction.
Appendix 3—table 19
The number of filler trials with different outcomes.
| Pain levels | |||||
|---|---|---|---|---|---|
| Number of filler trials (48 in total) | No pain stimulation | 1 | 2 | 3 | 4 |
| Number of wrong deciders excluding the participant | |||||
| 0 | 24 | - | - | - | - |
| 1 | - | 3 | 3 | 3 | 3 |
| 2 | - | 2 | 2 | 2 | 2 |
| 3 | - | 1 | 1 | 1 | 1 |
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Human neurocomputational mechanisms of guilt-driven and shame-driven altruistic behavior
eLife 14:RP107223.
https://doi.org/10.7554/eLife.107223.3
