The Neural Basis of the Human Justice Intuition
Introduction
As a social species, the success of humankind relies on our ability to cooperate on a large scale (Buckholtz et al., 2008). Critically, it hinges on individuals working with and relying on strangers on a daily basis (Buckholtz et al., 2015; Buckholtz & Marois, 2012; Sznycer et al., 2016). For example, in order for one person to devote themself to agriculture, others must practice medicine, teach our children, protect our safety, and contribute to shared efforts in every domain of life. This cooperative pattern has been evident throughout human history, dating back to the division of labor in hunter-gatherer societies. With social coherence as a fundamental cornerstone of human resilience, it is thus argued that social norms are the key to our success (Buckholtz & Marois, 2012; Buckholtz et al., 2015). Norms – such as those that prohibit lying and theft – serve as a communal contract, allowing us to rely on strangers to meet our most fundamental needs (Sznycer et al., 2023). For this socially reliant system to function properly, evolutionary psychologists posit that over time, our species built upon domain-general neural networks to develop a shared set of cognitive tools for evaluating what is “right” for the greater good of society. (Buckholtz et al., 2015; Sell and Sznycer, 2023). In other words, natural selection favored cognitive mechanisms that promote agreement in the human species (Sznycer et al., 2023). We propose that these cognitive adaptations form the basis of what Sznycer and Patrick (2020) call the “human justice intuition” – our shared, innate sense of what is right and wrong.
1.1. An adaptationist theory
It is theorized, from an adaptationist perspective, that evolutionary pressures have favored human traits which promote interpersonal agreement about norms (Sznycer & Patrick, 2020). That is, we share cognitive mechanisms that have helped us establish norms, harmoniously appraise the morality of actions, and respond to norm violations in ways that benefit societal well-being (Sznycer et al., 2016).
1.1.1 Consensus on appropriate punishments for moral violations
We suggest that a shared human justice intuition explains a surprising phenomenon; across different demographics, cultures, and countries, people generally agree about the magnitude of punishment deserved for different criminal offenses (Robinson et al., 2007; Sznycer et al. 2020). Robinson et al. (2007) reviewed empirical research investigating shared intuitions about appropriate punishments for wrongdoing, and found remarkable agreement among individuals across all studies. One study asked 64 participants to rank the deserved punishments of 24 criminal offenses, and the researchers reported a Kendall’s W correlation of 0.95 across subjects (for reference, a correlation of 1 signifies perfect agreement across all participants). Most notably, other studies included in the review demonstrated that high levels of agreement persists across countries with considerable cultural differences, including Yugoslavia, the United States, Iran, and Ireland (Newman & Wolfgang, 2008; O’Connell & Whelan, 1996; Robinson et al., 2007). This significant interpersonal agreement despite cultural differences supports our proposition of a human justice intuition as a universal cognitive mechanism through which all humans establish and uphold norms.
1.1.2. Justice intuition as the source of legal codes
In practice, societies often codify their shared values into criminal law codes, with legal punishments set to deter individuals from disrupting social harmony. Sznycer and colleagues (2023) posit that legal codes are a byproduct of our underlying shared justice intuition, and thus aim to handle wrongdoing in a manner that yields the maximum benefit for the greater good of society. Under these premises, fair punishment requires a balance; it must effectively reform wrongful action without deterring offenders from positively contributing to society after punishment (Sznycer & Patrick, 2020; Sznycer et al., 2023).
1.1.3. A common valuation grammar
Drawing on adaptationist theories, Sznycer and Patrick (2020) introduce what they term a “common valuation grammar” as the foundation of our human justice intuition. This valuation grammar refers to the shared cognitive formula that allows all humans to evaluate social transgressions in a consistent, goal-directed manner aimed at fair punishment and the promotion of overall social well-being. The proposed formula has three key variables: (1) Devaluation - the extent to which a person loses social value for committing an offense; (2) Shame - the degree of shame experienced for committing an offense; and (3) Wrongness - the perceived moral severity of an offense.
1.1.3.1. Devaluation
Given that an individual’s social value often dictates their success, social devaluation is extremely costly (Sznycer and Patrick, 2020). Therefore, evolutionary pressures are believed to have favored cognitive mechanisms that allow us to accurately assess how our actions may impact our social standing (Sznycer et al., 2016). For instance, we can intuitively recognize that killing another person would harm our social standing and, in turn, reduce our chances of leading a successful life. From this perspective, the psychological mechanisms that guide our actions are thought to recruit cognitive tools specifically for evaluating the social consequences of our behavior.
1.1.3.2. Shame
Appraisal theory suggests that emotions, such as shame, arise as the result of a person’s appraisal of situations rather than as a direct product of situations themself (Tracy and Robins, 2006). According to information threat theory, shame specifically arises in response to situations that pose a threat of devaluation (Sznycer et al., 2016). That is, we feel shame about behaviors – such as theft and infidelity – that violate social norms and could damage our social standing. This connection between shame and devaluation is supported empirically, as Sznycer et al. (2016) demonstrate that an individual’s reported level of shame for an action can be predicted by the extent to which others report they would devalue someone who commits the act. For example, participant ratings of how much shame they would feel for stealing goods correlated with the average amount that other participants reported they would “devalue” a person for stealing. Notably, the researchers found that levels of shame tracked actual social devaluation ratings across cultures. That is, ratings of shame from participants in the United States correlated with social devaluation ratings assigned by participants in Israel and vice versa. Given that shame consistently and universally predicts social devaluation, principles of parsimony suggest that both variables are likely derived from a singular valuation system – our common valuation grammar – which is shared by all humans (Sznycer et al., 2016).
1.1.3.3. Wrongness
The severity of punishments is intended to intrinsically reflect the wrongness of offenses committed (Yang et al., 2019). Empirical studies support this, showing that participants punish crimes proportionally to the perceived moral wrongness of offenses (Buckholtz et al., 2008). Thus, it is thought that perceptions of wrongness are formulaically used to calculate appropriate punishments for criminal offenses.
1.1.4. Prior study and rationale for replication
Sznycer and Patrick (2020) conducted a behavioral experiment to investigate the role of these three proposed variables – Devaluation, Shame, and Wrongness – in the process of evaluating criminal offenses. Participants from India and the U.S. were asked to rate various criminal offenses based on each of the three variables before assigning a prison time punishment for the offenses. The ratings were framed as follows: (1) Devaluation - How negatively would you view another person for killing a person, (2) Shame - How much shame would you feel if you killed a person, (3) Wrongness - How morally wrong do you think it is to kill a person, and (4) Time - How much prison time, if any, should killing a person be punished with by law. The offenses used in the study were selected from real legal codes (one modern and one ancient). This study design allowed the researchers to examine the relationship of the valuation grammar variables to both participant punishment decisions and objective punishments prescribed in legal codes. Based on the hypothesis that these variables constitute the cognitive algorithm used to evaluate wrongdoing, the researchers predicted that all three variables would systematically track punishment assignments. Further, the use of both modern and ancient legal codes allowed the researchers to test the assumption that all humans – including past and present legal code creators – rely on the same shared cognitive mechanisms to evaluate wrongdoing and assign punishments. Thus, the researchers predicted that the valuation grammar variables would also track the actual legal punishments.
Sznycer and Patrick (2020) found that participant ratings of shame, devaluation, wrongness and prison time demonstrated high intersubject agreement within their own countries as well as across countries. That is, participant ratings were highly correlated with other participants, regardless of cultural background. Perhaps more notably, the researchers found that participant ratings positively correlated with the actual punishments in both the modern and ancient legal code. These findings support the theory that all humans (including study participants as well as ancient and modern legal code creators) use a shared cognitive mechanism to evaluate criminal offenses. While this study supports the idea of a shared human justice intuition, the study methods did not provide insight into the cognitive mechanisms responsible for this high interpersonal agreement. The current study will replicate the design of Sznycer and Patrick’s (2020) study, while collecting fMRI data to examine the neural basis of devaluation, shame, wrongness and time ratings. We refer now to prior neuroimaging studies that examined the cognitive process of assigning punishment for wrongdoing.
1.2. Neural basis of evaluating wrongdoing
1.2.1. Third-party punishment in fMRI studies
The enforcement of social norms through legal punishment is an example of third-party punishment (TPP), which involves individuals – such as judges – assigning punishments for actions that do not directly affect them (Buckholtz & Marois, 2012). Researchers interested in the cognitive processing behind TPP often use fMRI neuroimaging to measure neural activity as participants assign punishments for various acts. Across fMRI studies of TPP, research converges to highlight three distinct cognitive steps involved in TPP: (1) the detection of harm, (2) the evaluation of a perpetrator’s intentions and blameworthiness, and (3) the selection of an appropriate punishment, accordingly (Bellucci et al., 2017; Buckholtz et al., 2008, 2015; Ginther et al., 2016; Glass et al., 2016). These three processing steps are attributed to distinct neural networks, with harm detection stemming from the salience network, blame assessments engaging elements of the mentalizing network and subsequent punishment decisions recruiting regions in the central-executive network (Bellucci et al., 2017, 2020; Buckholtz & Marois, 2012; Krueger & Hoffman, 2016; Yang et al., 2019, 2021).
1.2.1.1. The Salience Network
The salience network is thought to constantly monitor and prioritize the stimuli in a person’s environment that require attention. In regards to TPP, it is believed to detect harmful acts that violate norms, create an adverse emotional response to the situation, and alert other cognitive systems to the threat (Kreuger and Hoffman, 2016). The anterior insula (AI) is a key region in the salience network and is believed to be specifically responsible for detecting harmful acts (Bellucci et al., 2020). Neuroimaging studies of TPP consistently show activation in the AI and other salient network regions during punishment decisions, with AI activation specifically correlating with levels of shame and guilt related to immoral behavior (Bellucci et al., 2020; Roth et al., 2014; Weber et al., 2024; Yang et al., 2019; Zhu et al., 2019)
1.2.1.2. The Mentalizing Network
The mentalizing network is thought to be responsible for self-referential processes as well as our ability to infer the mental states, intentions and beliefs of others (Kreuger & Hoffmann, 2016). That is, it allows us to reflect on our own mental states, recognize that other people have internal thoughts and beliefs that differ from our own, and infer the intentions behind others’ actions (Glass et al., 2016). When engaging in TPP, this network helps us evaluate the perpetrators’ intentions in order to assess their responsibility for crimes (Bellucci et al., 2017). The mentalizing network includes the posterior cingulate cortex (PCC), the dorsomedial prefrontal cortex (dmPFC), the amygdala, and the temporoparietal junction (TPJ), among others (Buckholtz & Marois, 2012; Glass et al., 2016). These regions are thought to work together to establish blameworthiness, before sending that information to higher-order regions in the central-executive network, which is thought to ultimately determine the appropriate punishment (Bellucci et al., 2017).
1.2.1.3. The Central-Executive Network
The central-executive network (CEN) is responsible for higher-order cognitive functions, including attentional control and decision making. In the context of TPP, it is suggested that the CEN integrates streams of information from the salience network and mentalizing network before selecting an appropriate punishment accordingly (Bellucci et al., 2020; Buckholtz et al., 2015; Treadway et al., 2014; Yang et al., 2019). Particularly, neuroimaging studies identify the dorsomedial prefrontal cortex (DLPFC) as the “hub” in the CEN that consolidates evaluations of responsibility and mediates punishment decisions (Bellucci et al., 2017).This proposed role of the DLPFC – referred to as the “integration-and-selection hypothesis” – is supported by evidence showing that disrupting DLPFC activity using transcranial magnetic stimulation (TMS) reduces participants’ punishment assignments without affecting their judgments of blameworthiness (Buckholtz et al., 2015). This suggests that when the DLPFC is inhibited, participants remain capable of evaluating blameworthiness – a process attributed to the mentalizing network – but struggle to integrate that information into their punishment decisions. Thus, the CEN, and the DLPFC in particular, is thought to support the final stage of TPP: translating blame into a calibrated punitive response.
1.3. The current study
The current study aims to bridge evolutionary theories of the human justice intuition with neuroscientific understandings of TPP. We replicate the Sznycer and Patrick (2020) study by having participants complete ratings for devaluation, shame, wrongness and punishment (prison time assignment) while undergoing fMRI scanning. By examining neural activation in regions of the salience, mentalizing and central-executive networks during these ratings, we investigate neural mechanisms recruited when assigning punishment and when evaluating the three proposed components of the human justice intuition formula. We hypothesize that the cognitive mechanisms recruited for TPP are also recruited for judgements related to the three formulaic elements proposed by Sznycer and Patrick (2020).
1.3.1. Hypotheses
- Neural activation patterns for shame, devaluation, wrongness, and punishment time will correlate with participant ratings for those four categories, and the same neural regions will be recruited across all four rating types.
- Neural activation patterns for shame, devaluation, and wrongness will correlate with participants’ assigned punishments (time ratings).
- Neural activation patterns for shame, devaluation, wrongness, and punishment time will correlate with the objective punishments assigned in the legal codes from which the crimes were selected.
2. Methods
2.1. Participants
Twenty-two U.S. college students completed the fMRI study. One participant was excluded from data analysis because they failed to submit ratings for all trials. There was reasonable diversity among study participants (mean age = 20.10, SD = 1.76, range = 18-23; 13 female, 8 male; 11 Caucasian/White, 8 Hispanic/LatinX, 2 Mixed). The study was designed as a pilot; therefore, a formal power analysis was not performed. However, our sample size was similar to previous neuroimaging studies examining TPP and related emotional processing (Bellucci et al., 2016, N = 26; Michl et al., 2014, N=14; Buckholtz et al., 2015, N=10; Buckholtz et al., 2008, N=16). The study was approved by the Human Subjects Committee of University of California Santa Barbara (protocol number: 6-24-0621), and participants were recruited via flyers distributed on campus at the university. Potential participants were screened for MRI safety requirements and the study’s exclusionary criteria through an online form. Participants that met both study criteria and MRI safety criteria were then contacted for enrollment in the study. Each participant received a compensation of $20 per hour for the MRI scans and $10 per hour for other participation time, (including time required for survey completion, fMRI preparation, and post-task debrief). All participants provided informed consent prior to beginning the study and were debriefed upon completion.
2.2. Procedure
The study involved two rounds of data collection. Participants first completed an online survey through Qualtrics that gathered demographic data and included eight socio-psychological scales (see Appendix). Upon completion of the survey, participants scheduled their fMRI data collection appointments. fMRI data was collected using a 3T Siemens Prisma scanner at the Brain Imaging Center at the University of California, Santa Barbara. Prior to beginning the experiment, participants were introduced to the study design and practiced the experiment task on a laboratory computer. Once familiar with the task, participants provided informed consent and completed fMRI safety screening. Participants then entered the fMRI scanner and we collected anatomical scans for each participant before beginning the experimental task, which was divided into four sessions. Participants were offered brief breaks between sessions of the task. Upon completion of all four sessions, participants exited the scanner. Participants then used a laboratory computer to re-attempt any trials that they failed to complete during the task, providing behavioral data for those missed trials. One participant failed to complete all missed trials, and their data was excluded from analysis. Participants then completed a post-task survey to gauge their level of engagement with the task as well as their understanding of the experiment, and to identify whether they were familiar with the legal codes used. We planned to exclude data of participants who correctly identified the legal codes used, but no participants did so successfully. We then debriefed participants, thanked them, and provided them with their compensation.
2.3. Rating task
The rating task for the experiment was directly derived from the original study that was replicated (Sznycer and Patrick, 2020). As in the original experiment, participants were asked to rate criminal offenses for each of four experimental conditions: (1) Shame: How much shame would you feel for committing the act; (2) Devaluation: How negatively would you view another person for committing the act; (3) Wrongness: How morally wrong do you think the act is; and (4) Time: How much prison time, if any, would you assign for the act (See Figure 1). Participants rated all 55 crimes for each condition, for a total of 220 crime-condition combinations. The 220 combinations were randomly divided into four sessions, with each session containing a roughly equal distribution of crimes from each legal code and condition per session.
Figure 1: Summary of the questions, scale ranges and scale ends for all four conditions. The Devaluation condition uniquely used gender-specific pronouns based on the participant’s identified gender (e.g. I wouldn’t view her negatively at all).
Each session began with a 10-second fixation cross to capture baseline neural activity, followed by a 10-second condition prompt (ex: “How much SHAME would you feel if you were to take the following actions?”), and a 2-second wait screen (see Figure 2). Immediately after the wait screen, participants were presented with 13-14 randomized criminal offense situations, one at a time. For each situation, participants viewed the situation description for 4 seconds, then a sliding scale appeared for them to provide their rating. Participants had approximately 9 seconds to complete their ratings, then a fixation cross was presented for 4-6 seconds to re-model baseline neural activity before the next situation (see Figure 3). This sequence was repeated for all four condition blocks, each containing 13-14 randomized questions.
Figure 2. Sequence used to introduce each condition block within a session. Each session comprised four question blocks (one per condition), with approximately 13-14 situations in each block. Each condition block was introduced using the sequence shown above: fixation cross, condition question, and wait screen.
Figure 3. Sequence used to present criminal offenses as stimuli. Within each condition block, situations were presented in the following order: the situation was shown, followed by a slider for participants to rate the situation, and then a fixation cross appeared to re-establish baseline neural activity before the next situation.
Attention checks were randomly displayed during the second and fourth session to ensure active participation during the task. We planned to exclude the data of participants who failed, but all participants passed the attention checks. Each of the four sessions lasted approximately 18m, for a total of ~72 minutes of active task engagement.
2.4. Stimuli
The 55 criminal offenses used in the study were selected from the original study conducted by Sznycer and Patrick (2020), and originate from two legal codes: the U.S. Code Title 18 (n = 25) and the Tang Code of Imperial China (n = 30). Title 18 is an active legal code and contains modern criminal offenses (e.g., “Causing bodily injury to a child younger than 12”). In contrast, the Tang Code has been inactive for over 1,000 years and contains ancient criminal offenses (e.g., “Recklessly racing horses where there are groups of people without any reason”). These legal codes allowed us to examine whether the proposed human justice intuition is a shared cognitive feature of both modern and ancient human societies.
2.5. Behavioral measurements
Participants used a hand-held clicker to complete their ratings from inside the fMRI machine. The response slider always started in the middle of the scale, and participants used right and left buttons to move the slider accordingly. A central button was used to submit the ratings. The Shame, Wrongness, and Devaluation conditions were rated on a continuous scale of 0-100, while the Time condition was rated on a scale of 0-50 years (see Figure 1).
2.6. Imaging
Neural activation was measured using the blood-oxygenation level dependent (BOLD) effect. We obtained BOLD contrast using a gradient echo-planar imaging sequence (General Electric Signa HDX 3.0T; field strength of 3 Tesla; whole brain coverage with 54 interleaved slices, slice size 2.50mm; TR = 1900 ms; TE = 30.0 ms; flip angle = 65◦, field of view = 21 x 21 cm2, matrix size 78 x 93 x 78). Raw DICOMs were organized based on the Brain Imaging Data Structure (BIDS; Weber et al., 2024).
2.7 Regions of interest
In line with the literature, the most recent fMRI study on TPP included regions spanning the salience, mentalizing and central-executive networks (Weber et al., 2024). From these relevant regions, we selected a representative region for the salience network (Dorsal Anterior Insula; see Figure 4) and the mentalizing network (TPJ Posterior Supramarginal Angular Gyrus; see Figure 5) for analysis.
Figure 4. Visualization of the Dorsal Anterior Insula (dAI) from the Neurosynth Parcellation (k =50), overlaid on a standard anatomical template.
Figure 5. Visualization of the TPJ Posterior Supramarginal Angular Gyrus (TPJ-P) from the Neurosynth Parcellation (k =50), overlaid on a standard anatomical template.
2.8. fMRI data preprocessing
`The fMRI data underwent minimal preprocessing using fMRIprep, a tool that adapts preprocessing workflows based on specific datasets, addressing reproducibility issues commonly encountered with fMRI preprocessing (Esteban et al., 2019). Following our preprocessing with fMRIprep, we completed additional post-processing steps consistent with established guidelines for imaging data. We applied spatial smoothing (fwhm=5mm) and carried out basic voxelwise denoising using a General Linear Model (GLM) that incorporated the six realignment parameters.
2.9. Representational similarity analysis
Representational Similarity Analysis (RSA) is a computational technique that allows for comparison between neural activation and behavioral ratings (Kriegeskorte, 2008; Popal et al., 2019). For neuroimaging data, an individual’s specific neural response to one stimulus is compared to their neural response to other stimuli, allowing us to capture the distinct (or similar) ways that the individual neurally represents the different stimuli. For behavioral data, participant responses to different stimuli are similarly compared. RSA then assesses the correspondence between these two representational spaces – neural and behavioral – by correlating the patterns observed in neuroimaging and behavioral data. When a participant’s neural representations of stimuli significantly correlate with their behavioral responses, it is suggested that their behavioral responses are encoded in the neural substrates examined (Kriegeskorte, 2008). In the context of the current study, say a participant gives similar ratings for theft and lying, but a markedly different rating for murder. If their neural activation patterns during the evaluation of theft and lying are also similar – but the pattern for murder is distinct – this suggests that the participant’s neural activation reflects the structure of their behavioral responses. In this case, neural patterns can be interpreted as encoded representations of the participant’s subjective evaluations. In short, RSA allows us to examine whether patterns of neural activation mirror patterns of behavioral responses across stimuli.
We used RSA to compare neural activation patterns within our two ROIs to participants’ ratings of crimes across all four conditions. This approach allowed us to assess whether these regions represent information in a way that reflects participants’ evaluative judgements.
2.9.1. Participant-Level RSA
RSA compares neural and behavioral representations of stimuli by constructing representational dissimilarity matrices (RDMs) for each data type. A correlation is then computed between neural and behavioral RDMs to assess the similarity in their representational structures (See Figure 6). All RDMs were constructed using MATLAB within the Visual Studio Code environment.
Figure 6. Visualization of Spearman correlation analysis between RDMs. Neural and Behavioral RDMs were created as 55 x 55 matrices. The lower triangle of each RDM was then flattened into a single column vector, and a Spearman correlation was conducted between the flattened RDMs.
2.9.1.1. Neural RDM Construction
We created eight neural RDMs for each participant (2 ROIs x 4 Conditions; e.g., TPJ-P Shame Neural RDM). These RDMs were generated by computing pairwise Spearman correlations between multivoxel neural activation patterns for each pair of stimuli. For example, we calculated the Spearman correlation between the neural response to Crime 1 and Crime 2, then between Crime 1 and Crime 3, and so on, across all stimulus pairs. Because RDMs capture dissimilarity, we then computed 1 - ρ (Spearman’s rho) for each pair of stimuli and organized the resulting values into a 55 x 55 (number of crimes assessed) dissimilarity matrix. RDMs are symmetrical across the diagonal, so we then extracted the lower triangle of each matrix (excluding the diagonal) and flattened it into a single column vector for analysis.
2.9.1.2. Behavioral RDM Construction
We created four behavioral RDMs per participant – one for each condition. These were generated using a similar process as the neural RDMs. However, because the behavioral data contained only a single rating value per stimulus, we calculated the pairwise absolute difference between ratings for each stimulus pair instead of using Spearman correlations. Since absolute distance is already a measure of dissimilarity, the resulting values were directly organized into a 55 x 55 dissimilarity matrix. As with the neural RDMs, we extracted the lower triangle (excluding the diagonal) and flattened it into a single column vector for analysis.
2.9.1.3. Objective RDM Construction
We constructed an RDM of the objective prison sentences derived from the legal codes used as stimuli. This calculation followed the same procedure used for the behavioral RDMs; we computed pairwise absolute distance between prison sentence values for each stimulus pair, arranged the results into a 55 x 55 dissimilarity matrix, extracted the lower triangle (excluding the diagonal) and flattened it into a single column vector for analysis.
2.9.1.4 RSA Correlation Analyses
For each participant, we then calculated Spearman correlations between the flattened RDM vectors using the following three analysis approaches.
- Matched-Condition RSA
Our Matched-Condition RSA involved calculating the Spearman correlation between the flattened neural RDM and the corresponding flattened behavioral RDM for each condition (e.g. Behavioral Shame x Neural Shame). This analysis was conducted separately for each of the four conditions and across both ROIs, yielding eight Spearman correlation values per participant. A significant correlation for this analysis indicates that the pattern of neural activation across trials within a given ROI mirrors the pattern of behavioral responses, suggesting that the ROI may cognitively represent information relevant to the specific evaluation being made.
- Time Behavioral vs. Other Neural RSA
Our Behavioral Time vs. Other Neural RSA involved calculating the Spearman correlation between the flattened behavioral Time RDM and the flattened neural RDMs from the remaining three conditions (i.e., Behavioral Time x Neural Shame, Behavioral Time x Neural Devaluation, Behavioral Time x Neural Wrongness). This analysis was conducted separately for each of the three conditions and across both ROIs, yielding six Spearman correlation values per participant. A significant correlation for this analysis indicates that a participant’s prison time ratings correspond to their neural activation patterns in a given ROI during evaluations of the other judgement conditions. This suggests that the ROI may encode cognitive representations of shame, devaluation or wrongness that are functionally related to the cognitive processing underlying punishment decisions.
- Objective Vs. Neural RSA
Our Objective Vs. Neural RSA involved calculating the Spearman correlation between the flattened Objective Rating RDM and the flattened neural RDMs from all conditions (i.e., Objective ratings x Neural Devaluation). This analysis was conducted separately for all four conditions and across both ROIs, yielding eight Spearman correlation values per participant. A significant correlation for this analysis indicates that the objective punishments prescribed by legal codes correspond to participants’ neural activation patterns in a given ROI during evaluations for each condition. This suggests that the ROI may cognitively represent crimes in a manner that aligns with legally codified punishment decisions.
2.9.2. Group-Level RSA Analysis
After completing all RSA calculations for each participant, we grouped the resulting Spearman correlation values by analysis type, condition, and ROI. This resulted in eight Matched-Condition groupings (4 conditions x 2 ROIs), six Behavioral Time vs. Other Neural groupings (3 conditions x 2 ROIs), and eight Objective vs. Neural groupings (4 conditions x 2 ROIs). We then conducted one-sided, one-sample t-tests for each grouping of correlation values to assess whether the mean correlation was significantly greater than zero. T-tests were conducted utilizing Python in the Visual Studio Code environment. To account for multiple comparisons within each condition-analysis combination, we applied a Bonferroni correction based on the 2 ROIs tested per condition, adjusting the significance threshold to α = .025 (0.05/2).
3. Results
3.1. Matched-Condition RSA
This analysis assessed the relationship between neural activation patterns and behavioral ratings across four judgment conditions: Shame, Devaluation, Wrongness, and Time. Significant correlations indicate that in a given ROI, participants cognitively represent information in a way that tracks their evaluative decisions (see Figure 7).
No significant correlations were found between neural activation patterns and behavioral ratings in the Shame, Devaluation, or Wrongness conditions. However, in the Time condition, neural activation in the dAI (r = .028 , p < .05) and the TPJ-P (r = .030 , p < .05) was significantly correlated with participants’ prison time assignments. This suggests that these regions may cognitively represent information that is relevant to the process of assigning prison time punishment for wrongdoing.
Figure 7. Results of the RSA correlation analysis (neural-behavioral matched pairs). AI and TPJ neural representations of crimes in the Time condition are significantly correlated with Time ratings. Negative correlation values are not shown. *: p < .05 (Bonferroni corrected).
3.2. Time Behavioral vs. Other Neural RSA
This analysis examined whether neural activation patterns during the Devaluation, Shame, and Wrongness conditions were associated with participants’ prison time rating patterns. By examining these cross-condition correlations, we aimed to assess whether the neural encoding of each proposed element of the human justice intuition is related to the evaluative process of assigning punishment for crimes. Significant correlations suggest that the cognitive representations of devaluation, shame, and wrongness judgments are linked to participant punishment decisions (See Figure 8).
Participant time ratings were significantly correlated with neural activation patterns in the Shame condition within the dAI (r = .029, p < .001) and the TPJ-P (r = .018, p < .05), suggesting that cognitive representations in these regions during shame judgements are linked to the ultimate punishments assigned by participants. Similarly, in the Devaluation condition, neural activation in the dAI was associated with participant time ratings (r = .038, p < .05), suggesting that neural encoding of devaluation in the dAI is associated with punishment decisions.
Figure 8. Results of the RSA correlation analysis (time-behavioral vs. other neural). AI and TPJ neural representations of crimes in the Shame condition, and AI neural representation in the Devaluation condition, are significantly correlated with participants’ prison time estimate. *: p < .05, ***: p < .001 (Bonferroni corrected).
3.3. Objective vs. Neural RSA
No statistically significant correlations were found between neural activation patterns and the objective punishments prescribed in the legal codes.
4. Discussion
This study aimed to examine the neural basis of the human justice intuition and third party punishment (TPP) decisions. The results suggest that key elements of both the Mentalizing Network (TPJ-P) and the Salience Network (dAI) demonstrate meaningful activation patterns during prison time assignment decisions. Additionally, in both regions studied, we found that neural activation during Shame evaluations corresponded with participants’ punishment decisions. Finally, in the dAI, neural activation during Devaluation processing was correlated with prison time assignments as well. Together, these results support our proposal that cognitive mechanisms for evaluating Shame and Devaluation are recruited in a way that is meaningful for TPP and that those same mechanisms play a direct role in assigning punishments.
Looking at neural activation while participants assign prison times for various crimes, we see that activation in the dAI and the TPJ-P corresponds to prison time ratings, suggesting that both regions cognitively represent information relevant to the decisions made. Prior literature regarding TPP suggests that punishment decisions are primarily informed by two factors: (1) the amount of harm caused and (2) the intention and blameworthiness of the perpetrator (Buckholtz et al., 2015; Ginther et al., 2016; Glass et al., 2016; Kreuger & Hoffman, 2016). As part of the Salience Network, the Anterior Insula (AI) helps to detect pertinent external stimuli and direct attention towards them accordingly. In regards to TPP, previous neuroimaging studies suggest that the AI is specifically responsible for detecting acts that cause harm or violate norms (Bellucci et al., 2020; Yang et al., 2019). That is, the AI evaluates harm caused by actions and alerts other cognitive systems accordingly. Our findings support this hypothesis, as neural activation in the dAI corresponds with participants’ punishment decisions. Participants showed similar cognitive representations in the dAI for crimes that they rated similarly, and differing representations for those they rated differently. One interpretation of this finding is that the dAI cognitively represents information about the harm caused by criminal acts. Taken with prior literature, which indicates that harm evaluation is a key factor for assigning punishment, we suggest that the dAI contributes to TPP through assessing the harm caused by various actions (Bellucci et al., 2016; Buckholtz et al., 2015; Ginther et al., 2016; Kreuger and Hoffmann, 2016; Yang et al., 2019). We also demonstrate that activation in the TPJ-P correlates with punishment decisions in a similar manner. The TPJ, as a key player in the mentalizing network, is thought to be responsible for evaluating the mental states and intentions of others (Buckholtz et al., 2008). Prior neuroimaging studies suggest that the TPJ evaluates the intentions of perpetrators and assigns blameworthiness through producing “blame” signals that are used by other cognitive regions for punishment selection purposes (Bellucci et al., 2017; Bellucci et al., 2020; Buckholtz and Marois, 2012). This is supported by studies which have shown that during TPP, the TPJ activates before other regions in the brain, and is thought to send information about blame to other regions that use it to select appropriate punishments (Buckholtz et al., 2008; Buckholtz and Marois, 2012). Our results support this hypothesis, as activation in the TPJ-P was predictive of punishment decisions. Perhaps even stronger evidence for the roles of the TPJ-P and the dAI is captured by our analysis of neural activation during Shame, Devaluation and Wrongness judgements.
Beginning with the dAI, we demonstrate that neural activation in this region when considering the potential devaluative and shameful consequences of various acts corresponds to punishment decisions for those same crimes. The design of our study allowed us to identify that cognition in the dAI related to Devaluation and Shame – even when temporally separated from the act of assigning punishment – is linked to punishment decisions. We interpret these results as evidence in favor of adaptationist theories, which propose that shame functions as a proxy mechanism for tracking threats of social devaluation and, in doing so, deters individuals from engaging in norm-violating behavior. If shame and devaluation function as internal sensors that promote adherence to norms, we further suggest that legal punishments act as external manifestations of these evolved cognitive adaptations. Prior neuroscience research supports this view, with studies indicating that the dAI tracks norm violations and the harmful consequences of behavior as well as shame and guilt anticipation (Roth et al., 2014; Yu et al., 2020; Zhu et al., 2019). In line with this, our findings suggest that the dAI cognitively represents shame and devaluation in a way that is directly relevant to punishment assignments. Together with prior neuroimaging evidence, our results support the adaptationist theory of the human justice intuition and underscore the role of shame and devaluation in the cognitive processes used to evaluate crimes and assign punishment.
The human justice intuition is further supported by our finding that the TPJ-P cognitively represents shame evaluations in a manner that appears relevant to TPP ratings. As a region involved in understanding the mental states of others, the TPJ is thought to play a role in evaluating perpetrators’ intentions to assess blameworthiness. We propose that part of this process involves drawing on cognitive mechanisms related to shame. Adaptationist theories suggest that shame serves as an internal sensor for tracking social norms and that shame arises in response to violations of those norms. If shame functions as a universal internal gauge for what is “right” according to the social norms that shape legal codes, then levels of anticipated shame would also be referenced during the cognitive process of evaluating appropriate punishments. Our findings support this concept, as activation in the TPJ-P during shame judgements corresponded to participants’ prison time assignments. This suggests that participants’ mental representations of shame for various crimes correlated with their ultimate punishments. One interpretation of this finding is that the TPJ-P not only represents shame evaluations but also uses them as formulaic components when calculating punishments. Further support for this idea comes from prior work showing that when the TPJ is disrupted with TMS, participants place less weight on the perpetrator’s intention and more weight on the outcome when assigning punishments (Young et al., 2010). This suggests that the TPJ plays a critical role in incorporating intention into punishment decisions. Taken together with our findings, it appears likely that the TPJ-P contributes to TPP by evaluating perpetrators’ intentions, assessing blameworthiness, and interacting with other neural regions to integrate this information into a final punishment judgment.
5. Conclusion
Evolutionary psychology suggests that in order for humans to cooperate on a societal scale, there must be shared agreement about what is acceptable, established in the form of social norms. Underlying widespread agreement about norms, according to adaptationist theories, are shared cognitive mechanisms that allow all humans to evaluate actions in a consistent manner. It is proposed that these cognitive mechanisms form the neural basis of the human justice intuition – the shared and innate human sense of what is “right” and “wrong.” The human justice intuition is thought to consist of three formulaic elements: devaluation, shame and wrongness, which are used to inform and determine punishment decisions. In other words, when we assign a punishment for a crime, we use devaluation, shame and wrongness judgements to help calculate and select punishment severity. We used fMRI neuroimaging during devaluation, shame, wrongness and time punishments to examine the hypothesis that the neural mechanisms recruited for the three formulaic elements are also recruited for punishment decisions. We found that the TPJ-P and dAI activated in a meaningful manner during time punishments, suggesting that both regions cognitively encode information that is relevant to the decision making process. We also found that the dAI activates meaningfully during Devaluation and Shame judgements, cognitively representing information in a pattern that corresponds to prison time punishments. Neuroimaging research suggests that the dAI plays a key role in detecting harmful or norm-violating actions and alerting other cognitive systems accordingly. Here, we suggest that the dAI detects norm violations through mechanisms that serve as an internal sensor for anticipating shame and devaluation as a result of various actions. In a similar manner, we found that TPJ-P activation during Shame judgements corresponds to punishment decisions. As a key element of the mentalizing region, the TPJ-P is thought to evaluate the mental states of offenders and we suggest that it accomplishes this mental state evaluation through recruiting mechanisms for anticipating shame. Together, the dAI and TPJ-P activate and cognitively represent information about shame that is related to their punishment decisions, suggesting that the information encoded during shame evaluations is similarly recruited or accessed during punishment decisions. When actually making the decisions, these regions were again activated in a meaningful manner, further supporting our proposal that they play an important role in TPP. While these results provide support for the TPJ-P and dAI as cognitive elements involved in both TPP and evaluations related to devaluation, shame and wrongness, our study was limited by our small sample size (n=21). Neuroimaging studies are particularly impaired by small sample sizes, and the supporting evidence we found suggests that even more meaningful results may be uncovered with greater experimental power. Additionally, this study is limited by a region-based analysis approach, whereas full-brain analysis and functional connectivity analysis may provide greater insight into the role that various neural networks play in TPP. Future iterations of this work may allow researchers to better understand the specific roles that the salience, mentalizing and central-executive networks play in evaluating wrongdoing and assigning punishment. While we were limited to two representative regions, this study provides evidence that elements of the salience network (e.g. dAI) and the mentalizing network (e.g. TPJ-P) both play a meaningful role in assigning punishments for crimes and in evaluating the proposed elements of the human justice intuition. We propose that the TPJ-P and dAI meaningfully activate to evaluate shame and devaluation, and that those human justice intuitions are used to guide punishment decisions.
References
Bellucci, G., Camilleri, J. A., Iyengar, V., Eickhoff, S. B., & Krueger, F. (2020). The emerging neuroscience of social punishment: Meta-analytic evidence. Neuroscience & Biobehavioral Reviews, 113, 426–439. https://doi.org/10.1016/j.neubiorev.2020.04.011
Bellucci, G., Chernyak, S., Hoffman, M., Deshpande, G., Dal Monte, O., Knutson, K. M., Grafman, J., & Krueger, F. (2017). Effective connectivity of brain regions underlying third-party punishment: Functional MRI and Granger causality evidence. Social Neuroscience, 12(2), 124–134. https://doi.org/10.1080/17470919.2016.1153518
Buckholtz, J. W. (2015). Social norms, self-control, and the value of antisocial behavior. Current Opinion in Behavioral Sciences, 3, 122–129. https://doi.org/10.1016/j.cobeha.2015.03.004
Buckholtz, J. W., Asplund, C. L., Dux, P. E., Zald, D. H., Gore, J. C., Jones, O. D., & Marois, R. (2008). The Neural Correlates of Third-Party Punishment. Neuron, 60(5), 930–940. https://doi.org/10.1016/j.neuron.2008.10.016
Buckholtz, J. W., & Marois, R. (2012). The roots of modern justice: Cognitive and neural foundations of social norms and their enforcement. Nature Neuroscience, 15(5), 655–661. https://doi.org/10.1038/nn.3087
Buckholtz, J. W., Martin, J. W., Treadway, M. T., Jan, K., Zald, D. H., Jones, O., & Marois, R. (2015). From Blame to Punishment: Disrupting Prefrontal Cortex Activity Reveals Norm Enforcement Mechanisms. Neuron, 87(6), 1369–1380. https://doi.org/10.1016/j.neuron.2015.08.023
Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., Kent, J. D., Goncalves, M., DuPre, E., Snyder, M., Oya, H., Ghosh, S. S., Wright, J., Durnez, J., Poldrack, R. A., & Gorgolewski, K. J. (2019). fMRIPrep: A robust preprocessing pipeline for functional MRI. Nature Methods, 16(1), 111–116. https://doi.org/10.1038/s41592-018-0235-4
Ginther, M. R., Bonnie, R. J., Hoffman, M. B., Shen, F. X., Simons, K. W., Jones, O. D., & Marois, R. (2016a). Parsing the Behavioral and Brain Mechanisms of Third-Party Punishment. The Journal of Neuroscience, 36(36), 9420–9434. https://doi.org/10.1523/JNEUROSCI.4499-15.2016
Glass, L., Moody, L., Grafman, J., & Krueger, F. (2016). Neural signatures of third-party punishment: Evidence from penetrating traumatic brain injury. Social Cognitive and Affective Neuroscience, 11(2), 253–262. https://doi.org/10.1093/scan/nsv105
Kriegeskorte, N. (2008). Representational similarity analysis – connecting the branches of systems neuroscience. Frontiers in Systems Neuroscience. https://doi.org/10.3389/neuro.06.004.2008
Krueger, F., & Hoffman, M. (2016). The Emerging Neuroscience of Third-Party Punishment. Trends in Neurosciences, 39(8), 499–501. https://doi.org/10.1016/j.tins.2016.06.004
Michl, P., Meindl, T., Meister, F., Born, C., Engel, R. R., Reiser, M., & Hennig-Fast, K. (2014). Neurobiological underpinnings of shame and guilt: A pilot fMRI study. Social Cognitive and Affective Neuroscience, 9(2), 150–157. https://doi.org/10.1093/scan/nss114
Newman, G.R., & Wolfgang, M.E. (2008). Comparative Deviance: Perception and Law in Six Cultures (1st ed.). Routledge. https://doi.org/10.4324/9781315080840
O’Connell, M., & Whelan, A. (1996). TAKING WRONGS SERIOUSLY: Public Perceptions of Crime Seriousness. British Journal of Criminology, 36(2), 299–318. https://doi.org/10.1093/oxfordjournals.bjc.a014087
Popal, H., Wang, Y., & Olson, I. R. (2019). A Guide to Representational Similarity Analysis for Social Neuroscience. Social Cognitive and Affective Neuroscience, 14(11), 1243–1253. https://doi.org/10.1093/scan/nsz099
Robinson, P. H., & Kurzban, R. (2007). Concordance and Conflict in Intuitions of Justice.
Roth, L., Kaffenberger, T., Herwig, U., & Brühl, A. B. (2014). Brain Activation Associated with Pride and Shame. Neuropsychobiology, 69(2), 95–106. https://doi.org/10.1159/000358090
Sell, A., & Sznycer, D. (2023). Societal institutions echo evolved human nature: An analysis of the Western criminal justice system and its relation to anger. Evolution and Human Behavior, 44(3), 210–221. https://doi.org/10.1016/j.evolhumbehav.2023.01.007
Sznycer, D., & Patrick, C. (2020). The origins of criminal law. Nature Human Behaviour, 4(5), 506–516. https://doi.org/10.1038/s41562-020-0827-8
Sznycer, D., Sell, A., & Williams, K. E. G. (2023). Justice-making institutions and the ancestral logic of conflict. Evolution and Human Behavior, 44(3), 184–194. https://doi.org/10.1016/j.evolhumbehav.2022.12.003
Sznycer, D., Tooby, J., Cosmides, L., Porat, R., Shalvi, S., & Halperin, E. (2016). Shame closely tracks the threat of devaluation by others, even across cultures. Proceedings of the National Academy of Sciences, 113(10), 2625–2630. https://doi.org/10.1073/pnas.1514699113
Tracy, J. L., & Robins, R. W. (2006). Appraisal Antecedents of Shame and Guilt: Support for a Theoretical Model. Personality and Social Psychology Bulletin, 32(10), 1339–1351. https://doi.org/10.1177/0146167206290212
Treadway, M. T., Buckholtz, J. W., Martin, J. W., Jan, K., Asplund, C. L., Ginther, M. R., Jones, O. D., & Marois, R. (2014). Corticolimbic gating of emotion-driven punishment. Nature Neuroscience, 17(9), 1270–1275. https://doi.org/10.1038/nn.3781
Weber, R., Hopp, F. R., Eden, A., Fisher, J. T., & Lee, H.-E. (2024c). Vicarious punishment of moral violations in naturalistic drama narratives predicts cortical synchronization. NeuroImage, 292, 120613. https://doi.org/10.1016/j.neuroimage.2024.120613
Yang, Q., Bellucci, G., Hoffman, M., Hsu, K.-T., Lu, B., Deshpande, G., & Krueger, F. (2021). Intrinsic functional connectivity of the frontoparietal network predicts inter-individual differences in the propensity for costly third-party punishment. Cognitive, Affective, & Behavioral Neuroscience, 21(6), 1222–1232. https://doi.org/10.3758/s13415-021-00927-4
Yang, Q., Shao, R., Zhang, Q., Li, C., Li, Y., Li, H., & Lee, T. (2019). When morality opposes the law: An fMRI investigation into punishment judgments for crimes with good intentions. Neuropsychologia, 127, 195–203. https://doi.org/10.1016/j.neuropsychologia.2019.01.020
Young, L., Camprodon, J. A., Hauser, M., Pascual-Leone, A., & Saxe, R. (2010). Disruption of the right temporoparietal junction with transcranial magnetic stimulation reduces the role of beliefs in moral judgments. Proceedings of the National Academy of Sciences, 107(15), 6753–6758. https://doi.org/10.1073/pnas.0914826107
Yu, H., Koban, L., Chang, L. J., Wagner, U., Krishnan, A., Vuilleumier, P., Zhou, X., & Wager, T. D. (2020). A Generalizable Multivariate Brain Pattern for Interpersonal Guilt. Cerebral Cortex, 30(6), 3558–3572. https://doi.org/10.1093/cercor/bhz326
Zhu, R., Feng, C., Zhang, S., Mai, X., & Liu, C. (2019). Differentiating guilt and shame in an interpersonal context with univariate activation and multivariate pattern analyses. NeuroImage, 186, 476–486. https://doi.org/10.1016/j.neur... Exclusionary Criteria: History of metal welding; injury by metallic object; claustrophobia, seizures, or other medical conditions relevant to fMRI; pregnancy; inability to wear contact lenses if non-normal vision; tattoos with metallic pigments; metallic objects in body.
Study Exclusionary Criteria: Previously diagnosed with neurological, mental or psychological conditions; used illegal drugs or medication for mental disorders in the past 6 months; or had hearing loss or sensitivity.
Social-Psychological Scales: GASP, Individualism-Collectivism, Just World, Dark Factor, OUS, Moral Foundations, Moral Identity and Conviction, Essentialism.
Post a comment