Large language models show human-like content biases in transmission chain experiments.

As the use of large language models (LLMs) grows, it is important to examine whether they exhibit biases in their output. Research in cultural evolution, using transmission chain experiments, demonstrates that humans have biases to attend to, remember, and transmit some types of content over others. Here, in five preregistered experiments using material from previous studies with human participants, we use the same, transmission chain-like methodology, and find that the LLM ChatGPT-3 shows biases analogous to humans for content that is gender-stereotype-consistent, social, negative, threat-related, and biologically counterintuitive, over other content. The presence of these biases in LLM output suggests that such content is widespread in its training data and could have consequential downstream effects, by magnifying preexisting human tendencies for cognitively appealing and not necessarily informative, or valuable, content.

Curious about threats: Morbid curiosity and interest in conspiracy theories in US adults.

Conspiracy theories allege secret plots between two or more powerful actors to achieve an outcome, sometimes explaining important events or proposing alternative understandings of reality in opposition to mainstream accounts, and commonly highlight the threat presented by the plot and its conspirators. Research in psychology proposes that belief in conspiracy theories is motivated by a desire to understand threats and is predicted by increased anxiety. Morbid curiosity describes the tendency to seek out information about threatening or dangerous situations and is associated with an interest in threat-related entertainment and increased anxiety. Across three studies, we investigated the relationship between morbid curiosity and conspiracy theories in US-based samples. We found that higher trait morbid curiosity was associated with higher general conspiracist beliefs (Study 1) and the perceived threat of conspiratorial explanations of events (Study 2). Using a behavioural choice paradigm, we found that participants who chose to investigate morbidly curious stimuli were more likely to choose to learn about conspiratorial explanations for events (Study 3). Greater curiosity about the minds of dangerous people was consistently the strongest predictor of conspiratorial ideation and interest. These results suggest that morbid curiosity is an important but hitherto unstudied predictor of conspiratorial interest and belief.

Belief correlations with parental vaccine hesitancy: Results from a national survey.

We conducted a nationally representative survey of parents' beliefs and self-reported behaviors regarding childhood vaccinations. Using Bayesian selection among multivariate models, we found that beliefs, even those without any vaccine or health content, predicted vaccine-hesitant behaviors better than demographics, social network effects, or scientific reasoning. The multivariate structure of beliefs combined many types of ideation that included concerns about both conspiracies and side effects. Although they are not strongly related to vaccine-hesitant behavior, demographics were key predictors of beliefs. Our results support some of the previously proposed pro-vaccination messaging strategies and suggest some new strategies not previously considered.

Realizamos una encuesta nacionalmente representativa sobre las creencias y comportamientos autodeclarados por los padres con relación a la vacunación infantil. Usando selección bayesiana entre modelos multivariados, encontramos que las creencias, aun aquellas sin ningún contenido sobre vacunas o salud, predijeron comportamientos indecisos sobre la vacuna mejor que las características demográficas, los efectos de las redes sociales o el razonamiento científico. La estructura multivariada de las creencias combinó muchos tipos de ideación que incluyó preocupaciones tanto sobre conspiraciones como efectos secundarios. Aunque no están relacionados fuertemente con los comportamientos de indecisión, las características demográficas fueron predictores centrales de las creencias. Nuestros resultados apoyan algunas de las estrategias de mensajes pro­vacunación propuestas previamente y sugieren algunas nuevas estrategias no consideradas anteriormente. [vacunación, encuesta, cultura acumulativa, bayesiana].

An experimental investigation into the transmission of antivax attitudes using a fictional health controversy.

RATIONALE: Although vaccines are an invaluable weapon in combatting diseases, they are often surrounded by controversy. Vaccine controversies usually arise with the claims of some parents or doctors who link vaccines to harmful outcomes. These controversies often negatively affect vaccination coverage. OBJECTIVES: This experiment simulated a vaccine controversy to understand which content features of vaccination-related information are well transmitted and how this transmission affects vaccine intention. METHOD: All participants (N = 64) read two conflicting views (pro- and anti-) about a fictional vaccine ('dipherpox vaccine'). These conflicting views were held by a parent and a doctor, whose views varied across conditions. This information was transmitted along linear chains of four participants who recalled it and the product of their recall was passed to the next participant within their chain. They also responded whether they would vaccinate or not. RESULTS: The experience-based view held by the parent was better transmitted than the medical-based view held by the doctor, while the pro-vaccine and anti-vaccine views were similarly transmitted. Despite all the participants having neutral or positive attitudes towards vaccines in general, 39.1% of them decided not to vaccinate. Nevertheless, vaccination attitude was the strongest predictor of vaccination intention. The less positive participants' attitudes were towards vaccines in general, the less likely they were to vaccinate against dipherpox after exposure to the controversy. CONCLUSION: The results suggest that vaccination campaigns may be made more effective by including personal experiences of the negative consequences of non-vaccination.

Serial killers, spiders and cybersex: Social and survival information bias in the transmission of urban legends.

This study uses urban legends to examine the effects of the social information bias and survival information bias on cultural transmission across three phases of transmission: the choose-to-receive phase, the encode-and-retrieve phase, and the choose-to-transmit phase. In line with previous research into content biases, a linear transmission chain design with 60 participants aged 18-52 was used to examine the encode-and-retrieve phase, while participants were asked to rank their interest in reading the story behind a headline and passing a story on for the other two phases. Legends which contained social information (Social Type), legends which contained survival information (Survival Type), and legends which contained both forms of information (Combined Type) were all recalled with significantly greater accuracy than control material, while Social and Combined Type legends were recalled with significantly greater accuracy than Survival Type legends. In another study with 30 participants aged 18-22, no significant differences were found between legend types in either the choose-to-receive phase or the choose-to-transmit phase.