Social Learning Biases Research Articles

Formalising prestige bias: Differences between models with first-order and second-order cues.

Knowledge and behaviour are transmitted from one individual to another through social learning and eventually disseminated across the population. People often learn useful behaviours socially through selective bias rather than random selection of targets. Prestige bias, or the tendency to selectively imitate prestigious individuals, has been considered an important factor in influencing human behaviour. Although its importance in human society and culture has been recognised, the formulation of prestige bias is less developed than that of other social learning biases. To examine the effects of prestige bias on cultural evolution theoretically, it is imperative to formulate prestige and investigate its basic properties. We reviewed two definitions: one based on first-order cues, such as the demonstrator's appearance and job title, and the other based on second-order cues, such as people's behaviour towards the demonstrator (e.g. people increasingly pay attention to prestigious individuals). This study builds a computational model of prestige bias based on these two definitions and compares the cultural evolutionary dynamics they generate. Our models demonstrate the importance of distinguishing between the two types of formalisation, because they can have different influences on cultural evolution.

Algorithm-mediated social learning in online social networks.

Human social learning is increasingly occurring on online social platforms, such as Twitter, Facebook, and TikTok. On these platforms, algorithms exploit existing social-learning biases (i.e., towards prestigious, ingroup, moral, and emotional information, or 'PRIME' information) to sustain users' attention and maximize engagement. Here, we synthesize emerging insights into 'algorithm-mediated social learning' and propose a framework that examines its consequences in terms of functional misalignment. We suggest that, when social-learning biases are exploited by algorithms, PRIME information becomes amplified via human-algorithm interactions in the digital social environment in ways that cause social misperceptions and conflict, and spread misinformation. We discuss solutions for reducing functional misalignment, including algorithms promoting bounded diversification and increasing transparency of algorithmic amplification.

Brain mediators of biased social learning of self-perception in social anxiety disorder

Social anxiety disorder (SAD) is characterized by an excessive fear of social evaluation and a persistently negative view of the self. Here we test the hypothesis that negative biases in brain responses and in social learning of self-related information contribute to the negative self-image and low self-esteem characteristic of SAD. Adult participants diagnosed with social anxiety (N = 21) and matched controls (N = 23) rated their performance and received social feedback following a stressful public speaking task. We investigated how positive versus negative social feedback altered self-evaluation and state self-esteem and used functional Magnetic Resonance Imaging (fMRI) to characterize brain responses to positive versus negative feedback. Compared to controls, participants with SAD updated their self-evaluation and state self-esteem significantly more based on negative compared to positive social feedback. Responses in the frontoparietal network correlated with and mirrored these behavioral effects, with greater responses to positive than negative feedback in non-anxious controls but not in participants with SAD. Responses to social feedback in the anterior insula and other areas mediated the effects of negative versus positive feedback on changes in self-evaluation. In non-anxious participants, frontoparietal brain areas may contribute to a positive social learning bias. In SAD, frontoparietal areas are less recruited overall and less attuned to positive feedback, possibly reflecting differences in attention allocation and cognitive regulation. More negatively biased brain responses and social learning could contribute to maintaining a negative self-image in SAD and other internalizing disorders, thereby offering important new targets for interventions.

The cost of social influence: Own-gender and gender-stereotype social learning biases in adolescents and adults.

Pervasive gender gaps in academic subject and career choices are likely to be underpinned by social influences, including gender stereotypes of competence in academic and career domains (e.g., men excel at engineering, women excel at care), and model-based social learning biases (i.e., selective copying of particular individuals). Here, we explore the influence of gender stereotypes on social learning decisions in adolescent and adult males and females. Participants (Exp 1: N = 69 adolescents; Exp 2: N = 265 adults) were presented with 16 difficult multiple-choice questions from stereotypically feminine (e.g., care) and masculine (e.g., engineering) domains. The answer choices included the correct response and three incorrect responses paired with a male model, a female model, or no model. Participants' gender stereotype knowledge and endorsement were measured, and adolescents (Exp. 1) listed their academic subject choices. As predicted, there was a bias towards copying answers paired with a model (Exp.1: 74%, Exp. 2: 65% ps < .001). This resulted in less success than would be expected by chance (Exp. 1: 12%, Exp. 2: 16% ps < .001), demonstrating a negative consequence of social information. Adults (Exp 2) showed gender stereotyped social learning biases; they were more likely to copy a male model in masculine questions and a female model in feminine questions (p = .012). However, adolescents (Exp 1) showed no evidence of this stereotype bias; rather, there was a tendency for male adolescents to copy male models regardless of domain (p = .004). This own-gender bias was not apparent in female adolescents. In Exp 1, endorsement of masculine stereotypes was positively associated with selecting more own-gender typical academic subjects at school and copying significantly more male models in the male questions. The current study provides evidence for the first time that decision-making in both adolescence and adulthood is impacted by gender biases.

Changes in Learning From Social Feedback After Web-Based Interpretation Bias Modification: Secondary Analysis of a Digital Mental Health Intervention Among Individuals With High Social Anxiety Symptoms.

Biases in social reinforcement learning, or the process of learning to predict and optimize behavior based on rewards and punishments in the social environment, may underlie and maintain some negative cognitive biases that are characteristic of social anxiety. However, little is known about how cognitive and behavioral interventions may change social reinforcement learning in individuals who are anxious. This study assessed whether a scalable, web-based cognitive bias modification for interpretations (CBM-I) intervention changed social reinforcement learning biases in participants with high social anxiety symptoms. This study focused on 2 types of social reinforcement learning relevant to social anxiety: learning about other people and learning about one's own social performance. Participants (N=106) completed 2 laboratory sessions, separated by 5 weeks of ecological momentary assessment tracking emotion regulation strategy use and affect. Approximately half (n=51, 48.1%) of the participants completed up to 6 brief daily sessions of CBM-I in week 3. Participants completed a task that assessed social reinforcement learning about other people in both laboratory sessions and a task that assessed social reinforcement learning about one's own social performance in the second session. Behavioral data from these tasks were computationally modeled using Q-learning and analyzed using mixed effects models. After the CBM-I intervention, participants updated their beliefs about others more slowly (P=.04; Cohen d=-0.29) but used what they learned to make more accurate decisions (P=.005; Cohen d=0.20), choosing rewarding faces more frequently. These effects were not observed among participants who did not complete the CBM-I intervention. Participants who completed the CBM-I intervention also showed less-biased updating about their social performance than participants who did not complete the CBM-I intervention, learning similarly from positive and negative feedback and from feedback on items related to poor versus good social performance. Regardless of the intervention condition, participants at session 2 versus session 1 updated their expectancies about others more from rewarding (P=.003; Cohen d=0.43) and less from punishing outcomes (P=.001; Cohen d=-0.47), and they became more accurate at learning to avoid punishing faces (P=.001; Cohen d=0.20). Taken together, our results provide initial evidence that there may be some beneficial effects of both the CBM-I intervention and self-tracking of emotion regulation on social reinforcement learning in individuals who are socially anxious, although replication will be important.

Kin selection favors religious traditions: ancestor worship as a cultural descendant-leaving strategy

ABSTRACT Recent years have seen renewed interest in the role of religious systems as drivers of the evolution of cooperation in human societies. One suggestion is that a cultural tradition of ancestor worship might have evolved as a “descendant-leaving strategy” of ancestors by encouraging increased altruism particularly between distant kin. Specifically, Coe and others have suggested a mechanism of cultural transmission exploiting social learning biases, whereby ancestors have been able to establish parental manipulation of kin recognition and perceived relatedness as a traditional behavior, leading to increased altruism among co-descendants and thereby maximizing the ancestor’s long-term inclusive fitness. Here, we develop a demographically explicit model in order to quantify the resulting increase in altruism and concomitant “ancestor-descendant conflict”, and to determine the evolutionary feasibility of religiously motivated cultural norms that promote altruism among co-descendants. Our analysis reveals that such norms could indeed drive an overall increase in altruism with potential for ancestor-descendant conflict, particularly in low-dispersal settings. Moreover, we find that natural selection can favor traditions encouraging increased altruism towards co-descendants under a range of conditions, with the inclusive-fitness costs of enacting an inappropriately high level of altruism being offset by inclusive-fitness benefits derived from the cultural tradition facilitating kin recognition.

Different content biases affect fidelity of disease transmission along experimental diffusion chains

Although many theoretical studies have investigated which social learning biases are involved in the diffusion of information in human groups, many of these biases are still unknown. We investigated, for the first time, whether factors such as ancestry, regularity in evolutionary time, regularity in the current environment, and previous experience with health conditions act as biases along experimental transmission chains. We performed an experiment based on recall products from 80 chains of four generations each (N = 320). The content transmitted between the participants involved fictional stories containing information about different diseases (causes, symptoms, and forms of treatment) grouped according to their regularity in evolutionary time (acute or chronic), regularity in the current environment (high incidence or low incidence), period of origin (ancestral or modern), and previous experience (yes or no). The stories were passed along the linear chains of four participants who read and recalled them, and the product of their recall was passed on to the next participant within their chain. As predicted, we found that information about ancestral and more regularly affecting diseases throughout human (acute) evolutionary history acted as content biases, guiding the more accurate diffusion of information. In contrast, we found no evidence to support the influence of context biases for regularity in the current environment (high-incidence diseases) and from previous experiences when these were simulated along the experimental transmission chains. We argue that content bias may have driven the biased dissemination of information about ancestral and acute diseases. Acute illnesses were more regular over evolutionary time and may have generated greater selective pressure on human cognition. Furthermore, our behavioral immune systems may have evolved to respond more efficiently to clues to diseases we have faced thousands of years ago than clues to recent illnesses. This fact may have led to information about ancestral diseases becoming more memorable and better diffused in experimental cultural chains.

Content bias in the cultural evolution of house finch song

We used three years of house finch, Haemorhous mexicanus , song recordings spanning four decades in the introduced eastern range to assess how individual level cultural transmission mechanisms drive population level changes in birdsong. First, we developed an agent-based model (available as a new R package called ‘TransmissionBias’) that simulates the cultural transmission of house finch song given different parameters related to transmission biases, or biases in social learning that modify the probability of adoption of particular cultural variants. Next, we used approximate Bayesian computation and machine learning to estimate what parameter values likely generated the temporal changes in diversity in our observed data. We found evidence that strong content bias, likely targeted towards syllable complexity, plays a central role in the cultural evolution of house finch song in the New York metropolitan area. Frequency and demonstrator biases appear to be neutral or absent. Additionally, we estimated that house finch song is transmitted with extremely high fidelity. Future studies can use our simulation framework to better understand how cultural transmission and population declines influence song diversity in wild populations. • We analysed house finch songs from three years spanning four decades in New York. • We compared real data against simulations to make inferences about cultural transmission. • Content bias, likely for syllable complexity, drives cultural evolutionary changes. • Frequency and demonstrator biases were absent and transmission fidelity was high.

Aggression‐based social learning in the zebra finch (Taeniopygia guttata)

AbstractSelectively learning from specific types of individuals may be adaptive if demonstrator characteristics can be used to identify more beneficial sources of social information. Such “social learning biases” have been experimentally demonstrated in a number of species, but these experiments generally involve restricted laboratory conditions using a limited number of potential demonstrators and tend to consider only the characteristics of demonstrators rather than the importance of pairwise relationships on information transfer between individuals. In this study, we presented a novel foraging task to a large population of zebra finches (Taeniopygia guttata) housed in a free‐flying aviary and used multinetwork Network‐Based Diffusion Analysis (NBDA) to establish whether birds learned from individuals they shared particular relationships with. Specifically, we investigated whether task solves followed social learning pathways representing the following relationships between individuals: feeding associations, aggressive interactions, positive associations (e.g. grooming) and mating pairs. We found strong evidence that zebra finches learn from their aggressors, irrespective of the outcome of that aggressive encounter. This has been previously suggested in laboratory‐based studies on zebra finches, but never conclusively documented in a freely interacting population. We also found some weaker evidence to suggest that zebra finches learn from their mates—a social learning bias that has previously received little to no attention. However, we found that mates‐based learning occurred infrequently and was secondary to aggression‐based social learning biases. Our results therefore additionally highlight the importance of including combinations of multiple potential information pathways in social learning analyses to account for secondary learning pathways that may otherwise be missed.

Static environments with limited resources select for multiple foraging strategies rather than conformity

AbstractSocial learning and conformity, or positive frequency‐biased transmission, are important components of the evolution of culture and group identities in animals and humans. Previous theoretical work on the evolution of social learning and conformity when environmental states change found that conformity evolves for a large range of environmental conditions. However, the foraging environment can affect the adaptive benefits of social learning and conformity. Conformity can cause a population to focus mainly on one resource while ignoring other available resources, which can incite competition if the first resource is limited. We study the evolution of social learning and frequency‐biased transmission in a limited‐resource environment using a replicator model of a forager population feeding on two resources in which foragers set their resource preferences during an early learning stage. In sharp contrast with previous models, anti‐conformity, or negative frequency‐biased transmission, rather than conformity, evolves from a population of non‐conformists under most environmental conditions. Numerical simulations suggest that both social learning and conformity cause the foragers to favor one resource over the other even though the resources provide the same benefit to foragers. Resource limitation favors anti‐conformity because anti‐conformity tends to force both resources to be exploited equally. Similarly, depletion of resources selects against social learning. However, increasing social learning is adaptive when independent discovery of foraging cues is difficult because it increases the rate at which individuals learn to find food. Consequently, in an environment with difficult‐to‐learn food cues, increased social learning is more likely to evolve. However, social learning does not emerge from a population of individual learners. Our results show that competition and consumer‐resource interactions can alter the evolutionary course of social learning and transmission bias. As there are many examples of conformity in the animal kingdom, and especially in humans, future studies should explore what extensions of the model would allow conformity to first evolve.

When does it pay to follow the crowd? Children optimize imitation of causally irrelevant actions performed by a majority

Cultural evolutionary theory posits that human cultural complexity rests on a set of adaptive learning biases that help to guide functionality and optimality in social learning, but this sits in contrast with the commonly held view that children are unselective “over-imitators.” Here, we tested whether 4- and 6-year-old children use social learning biases flexibly to fine-tune their copying of irrelevant actions. Children watched a video of a majority demonstrating causally irrelevant actions and a minority demonstrating only causally relevant actions. In one condition observers approved of the majority and disapproved of the minority, and in the other condition observers watched the majority and minority neutrally. Results showed that both 4- and 6-year-olds copied the inefficient majority more often than the efficient minority when the observers had approved of the majority’s actions, but they copied the efficient minority significantly more when the observers had watched neutrally. We discuss the implications of children’s optimal selectivity in copying and the importance of integrating social approval into majority-biased learning when acquiring norms and conventions and in broader processes of cultural evolution.

Convergence of knowledge in a stochastic cultural evolution model with population structure, social learning and credibility biases

Understanding how knowledge emerges and propagates within groups is crucial to explain the evolution of human populations. In this work, we introduce a mathematically oriented model that draws on individual-based approaches, inhomogeneous Markov chains and learning algorithms, such as those introduced in [F. Cucker and S. Smale, On the mathematical foundations of learning, Bull. Amer. Math. Soc. 39 (2002) 1–49; F. Cucker, S. Smale and D. X. Zhou, Modeling language evolution, Found. Comput. Math. 4 (2004) 315–343]. After deriving the model, we study some of its mathematical properties, and establish theoretical and quantitative results in a simplified case. Finally, we run numerical simulations to illustrate some properties of the model. Our main result is that, as time goes to infinity, individuals’ knowledge can converge to a common shared knowledge that was not present in the convex combination of initial individuals’ knowledge.

Honesty biases trustworthiness impressions.

Honesty is central to trust and trustworthiness. However, how a good reputation as honest person is learned and induces trustworthiness impressions is still unexplored. Developing a novel paradigm, we show in 3 consecutive experiments that individuals prefer trusting honest others who share truthful information, especially if honest behavior is consistent over time. Trust in honest others was independent of proximal benefits, and honest individuals were repaid for their honesty with higher trust in a subsequent interaction. Crucially, signs of dishonesty decreased trust but only in those who had not previously built a good reputation as honest partners. On the contrary, those who could establish a good reputation were trusted even when they were no longer trustworthy, suggesting that participants could not successfully track changes in trustworthiness of those with an established good reputation. These findings suggest that a good reputation biases the ability to learn the momentary trustworthiness of another person and impairs the updating of one's beliefs about the other's character for behavior revision. Computational modeling analyses indicate an asymmetry in information integration when interacting with honest individuals that likely underlies such learning impairment. By showing how a good reputation influences learning processes in trust-based interactions, our results provide a mechanistic account for biases in social learning and social interactions, advancing our understanding of social behaviors in particular and human cognition in general. (PsycInfo Database Record (c) 2020 APA, all rights reserved).

Rethinking prestige bias

Some cultural evolution researchers have argued for the importance of prestige bias as a systematic and widespread social learning bias, that structures human social learning and cultural transmission patterns. Broadly speaking, prestige bias accounts understand it as a bias towards copying ‘prestigious’ individuals (which are typically described as high-status, due to a high level of skill or success in a socially valued domain, and so are treated by others with respect and deference). Prestige bias, along with other social learning biases, has been argued to pay a crucial role in allowing cumulative cultural selection to take place, thereby generating adaptations that are key to our success as a species. However, I argue for skepticism about the plausibility and scope of a prestige bias account. I argue that although an account of prestige bias seems plausible or compelling on their face, it is committed to a particular view of the cognition underpinning the bias, and therefore to predictions regarding its flexibility and context-sensitivity. Given this, current empirical evidence gives us reason to doubt the explanatory value of a prestige bias account over a naive, goal-directed agent account. Additionally, the way that prestige is defined in empirical work is in tension with a general understanding of prestige, casting doubt upon its status as evidence of prestige bias. I examine two studies cited as evidence of prestige bias, arguing that in these cases we cannot clearly favour a prestige bias explanation over a goal-directed agent explanation.

Modelling rapid online cultural transmission: evaluating neutral models on Twitter data with approximate Bayesian computation

As social media technologies alter the variation, transmission and sorting of online information, short-term cultural evolution is transformed. In these media contexts, cultural evolution is an intra-generational process with much ‘horizontal’ transmission. As a pertinent case study, here we test variations of culture-evolutionary neutral models on recently-available Twitter data documenting the spread of true and false information. Using Approximate Bayesian Computation to resolve the full joint probability distribution of models with different social learning biases, emphasizing context versus content, we explore the dynamics of online information cascades: Are they driven by the intrinsic content of the message, or the extrinsic value (e.g., as a social badge) whose intrinsic value is arbitrary? Despite the obvious relevance of specific learning biases at the individual level, our tests at the online population scale indicate that unbiased learning model performs better at modelling information cascades whether true or false.

Cultural evolutionary theory as a theory of forces

Cultural evolutionary theory has been alternatively compared to a theory of forces, such as Newtonian mechanics, or the kinetic theory of gases. In this article, I clarify the scope and significance of these metatheoretical characterisations. First, I discuss the kinetic analogy, which has been recently put forward by Tim Lewens. According to it, cultural evolutionary theory is grounded on a bottom-up methodology, which highlights the additive effects of social learning biases on the emergence of large-scale cultural phenomena. Lewens supports this claim by arguing that it is a consequence of cultural evolutionists’ widespread commitment to population thinking. While I concur with Lewens that cultural evolutionists often actually conceive cultural change in aggregative terms, I think that the kinetic framework does not properly account for the explanatory import of population-level descriptions in cultural evolutionary theory. Starting from a criticism of Lewens’ interpretation of population thinking, I argue that the explanatory role of such descriptions is best understood within a dynamical framework—that is, a framework according to which cultural evolutionary theory is a theory of forces. After having spelled out the main features of this alternative interpretation, I elucidate in which respects it helps to outline a more accurate characterisation of the overarching structure of cultural evolutionary theory.

Social attention biases in juvenile wild vervet monkeys: implications for socialisation and social learning processes.

The concept of directed social learning predicts that social learning opportunities for an individual will depend on social dynamics, context and demonstrator identity. However, few empirical studies have examined social attention biases in animal groups. Sex-based and kinship-based biases in social learning and social attention towards females have been shown in a despotic and female philopatric primate: the vervet monkey (Chlorocebus pygerythrus). The present study examined social attention during the juvenile period. Social attention was recorded through 5-min focal observations during periods of natural foraging. Kin emerged as the most important focus of social attention in juveniles, intensified by biased spatial proximity towards matrilineal related members. The highest-ranking conspecifics were more frequently observed by juveniles than low-ranking ones. Additionally, younger and orphaned juveniles showed higher levels of social attention overall, compared to other age categories. No effect of the juvenile's hierarchical rank was detected, suggesting that the variation in social attention recorded reflects different biases and stages of social learning and socialisation, rather than social anxiety. Juvenile females tended to exhibit a dominance-based bias more strongly than did males. This might be explained by a greater emphasis on attaining social knowledge during juvenile socialisation in the philopatric sex. Moreover, despite a preferred association between juveniles, social attention was more often directed to adults, suggesting that adults may still be more often chosen as a target of attention independent of their dominance rank.

Social learning and partisan bias in the interpretation of climate trends

Vital scientific communications are frequently misinterpreted by the lay public as a result of motivated reasoning, where people misconstrue data to fit their political and psychological biases. In the case of climate change, some people have been found to systematically misinterpret climate data in ways that conflict with the intended message of climate scientists. While prior studies have attempted to reduce motivated reasoning through bipartisan communication networks, these networks have also been found to exacerbate bias. Popular theories hold that bipartisan networks amplify bias by exposing people to opposing beliefs. These theories are in tension with collective intelligence research, which shows that exchanging beliefs in social networks can facilitate social learning, thereby improving individual and group judgments. However, prior experiments in collective intelligence have relied almost exclusively on neutral questions that do not engage motivated reasoning. Using Amazon's Mechanical Turk, we conducted an online experiment to test how bipartisan social networks can influence subjects' interpretation of climate communications from NASA. Here, we show that exposure to opposing beliefs in structured bipartisan social networks substantially improved the accuracy of judgments among both conservatives and liberals, eliminating belief polarization. However, we also find that social learning can be reduced, and belief polarization maintained, as a result of partisan priming. We find that increasing the salience of partisanship during communication, both through exposure to the logos of political parties and through exposure to the political identities of network peers, can significantly reduce social learning.

Payoff- and Sex-Biased Social Learning Interact in a Wild Primate Population

Social learning in animals is now well documented, but few studies have determined the contexts shaping when social learning is deployed. Theoretical studies predict copying of conspecifics gaining higher payoffs [1-4], a bias demonstrated in primates only in captivity [5]. In the wild, research has shown selective attention toward the philopatric sex, a group's stable core [6]. Here, we report the first rigorous experimental test of the existence of a payoff bias in wild primates and its interaction with the sex of the model. We created a payoff bias in which an immigrant alpha male in each of three groups of wild vervet monkeys received five times more food upon opening a foraging box than did the philopatric alpha female, whereas in two control groups, male and female models received the same amount of food. We tested whether this payoff asymmetry would override thepreviously documented selective learning from resident females. Group members were tested after having watched both models. When both models received the same amount of food, audience members copied the female model significantly more than the male model, confirming previous findings. However, when a marked payoff bias was introduced, male, but not female, vervet monkeys significantly more often copied the male model receiving a higher payoff. These results demonstrate behavioral flexibility in the dispersing sex in these primates and suggest that the philopatric sex can afford to be more conservative in their social learning. Our findings show that multiple social-learning biases can coexist and interact within the same species.

Sixteen-month-old infants are sensitive to competence in third-party observational learning

Observational learning is important to development, but not all adult models are equally informative and accurate. Selectivity is important in observational learning. Past research studies have not always differentiated competence and confidence, so the current study investigated infants’ selective imitation after observing third-party interactions, when confidence and competence were varied independently. Forty-eight 16-month-olds watched a model demonstrate the function of tools while displaying high or low levels of confidence and competence. Infants were significantly more likely to imitate individuals who were competent and were influenced less by their confidence level. Infants were more likely to reach for and use the working tool if the model was competent in her tool choice, but infant behavior was less affected by differences in confidence. Results suggest that by 16-months of age infants are sensitive to the competence of a demonstrator, suggesting a bias in social learning towards more valuable information.