RESUMEN

BACKGROUND: Mortgage discrimination refers to the systematic withholding of home mortgages from minoritized groups. In recent years, there has been an increase in empirical research investigating associations of historical and contemporary mortgage discrimination on contemporary  outcomes. Investigators have used a variety of measurement methods and approaches, which may have implications for results and interpretation. PURPOSE: We conducted a systematic review of peer-reviewed literature that has quantified links between both historical and current mortgage discrimination with contemporary adverse environmental, social, and health outcomes. Our goals were to document the methodology used to measure and assign mortgage discrimination, to assess implications for results and interpretation, and to make recommendations for future work. METHODS: Following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines, literature searches were conducted in September 2022 using terms that combined concepts of mortgage discrimination, health, and neighborhood environment. RESULTS: In total, 45 papers fit the eligibility criteria. In these, researchers investigated associations between mortgage discrimination and: (1) health outcomes (N = 28); (2) environmental and social exposures including heat, air pollution, greenspace, soil lead levels, and crime (N = 12); and (3) built environment features, including presence of retail alcohol, fast food, and tobacco stores (N = 5). Eleven included studies used Home Mortgage Discrimination Act (HMDA) data to identify racialized bias in mortgage discrimination or redlining, and 34 used Homeowner Loan Corporation (HOLC) maps. The construction and parametrization of mortgage discrimination or redlining and the spatial assignment of HOLC grades to contemporary addresses or neighborhoods varied substantially across studies. CONCLUSIONS: Results from our review suggest the need for careful consideration of optimal methods to analyze mortgage discrimination such as HOLC spatial assignment or HMDA index parametrization, contemplation of covariates, and place-based knowledge of the study location.

RESUMEN

Background: Young men who have sex with men and young transgender women (YMSM-YTW) use online spaces to meet sexual partners with increasing regularity, and research shows that experiences of racism online mimics the real world. Objective: We analyzed differences by race and ethnicity in web-based and mobile apps used to meet sexual partners as reported by Chicago-based YMSM-YTW in 2016-2017. Methods: A racially and ethnically diverse sample of 643 YMSM-YTW aged 16-29 years were asked to name websites or mobile apps used to seek a sexual partner in the prior 6 months, as well as provide information about sexual partnerships from the same period. We used logistic regression to assess the adjusted association of race and ethnicity with (1) use of any website or mobile apps to find a sexual partner, (2) use of a "social network" to find a sexual partner compared to websites or mobile apps predominantly used for dating or hookups, (3) use of specific websites or mobile apps, and (4) reporting successfully meeting a sexual partner online among website or mobile app users. Results: While most YMSM-YTW (454/643, 70.6%) used websites or mobile apps to find sexual partners, we found that Black non-Hispanic YMSM-YTW were significantly less likely to report doing so (comparing White non-Hispanic to Black non-Hispanic: adjusted odds ratio [aOR] 1.74, 95% CI 1.10-2.76). Black non-Hispanic YMSM-YTW were more likely to have used a social network site to find a sexual partner (comparing White non-Hispanic to Black non-Hispanic: aOR 0.20, 95% CI 0.11-0.37), though this was only reported by one-third (149/454, 32.8%) of all app-using participants. Individual apps used varied by race and ethnicity, with Grindr, Tinder, and Scruff being more common among White non-Hispanic YMSM-YTW (93/123, 75.6%; 72/123, 58.5%; and 30/123, 24.4%, respectively) than among Black non-Hispanic YMSM-YTW (65/178, 36.5%; 25/178, 14%; and 4/178, 2.2%, respectively) and Jack'd and Facebook being more common among Black non-Hispanic YMSM-YTW (105/178, 59% and 64/178, 36%, respectively) than among White non-Hispanic YMSM-YTW (6/123, 4.9% and 8/123, 6.5%, respectively). Finally, we found that while half (230/454, 50.7%) of YMSM-YTW app users reported successfully meeting a new sexual partner on an app, Black non-Hispanic YMSM-YTW app users were less likely to have done so than White non-Hispanic app users (comparing White non-Hispanic to Black non-Hispanic: aOR 2.46, 95% CI 1.50-4.05). Conclusions: We found that Black non-Hispanic YMSM-YTW engaged with websites or mobile apps and found sexual partners systematically differently than White non-Hispanic YMSM-YTW. Our findings give a deeper understanding of how racial and ethnic sexual mixing patterns arise and have implications for the spread of sexually transmitted infections among Chicago's YMSM-YTW.

Asunto(s)

RESUMEN

Pulse oximetry (SpO2) is a critical monitor for assessing oxygenation status and guiding therapy in critically ill patients. Race has been identified as a potential source of SpO2 error, with consequent bias and inequities in healthcare. This study was designed to evaluate the incidence of occult hypoxemia and accuracy of pulse oximetry associated with the Massey-Martin scale and characterize the relationship between Massey scores and self-identified race. This retrospective single institute study utilized the Massey-Martin scale as a quantitative assessment of skin pigmentation. These values were recorded peri-operatively in patients enrolled in unrelated clinical trials. The electronic medical record was utilized to obtain demographics, arterial blood gas values, and time matched SpO2 values for each PaO2 ≤ 125 mmHg recorded throughout their hospitalizations. Differences between SaO2 and SpO2 were compared as a function of both Massey score and self-reported race. 4030 paired SaO2-SpO2 values were available from 579 patients. The average error (SaO2-SpO2) ± SD was 0.23 ± 2.6%. Statistically significant differences were observed within Massey scores and among races, with average errors that ranged from - 0.39 ± 2.3 to 0.53 ± 2.5 and - 0.55 ± 2.1 to 0.37 ± 2.7, respectively. Skin color varied widely within each self-identified race category. There was no clinically significant association between error rates and Massey-Martin scale grades and no clinically significant difference in accuracy observed between self-reported Black and White patients. In addition, self-reported race is not an appropriate surrogate for skin color.

RESUMEN

BACKGROUND AND OBJECTIVES: Early detection of hepatocellular carcinoma (HCC) is associated with improved survival. However, a greater proportion of patients treated at safety net hospitals (SNHs) present with late-stage disease compared to those at academic medical centers (AMCs). This study aims to identify barriers to diagnosis of HCC, highlighting differences between SNHs and AMCs. METHODS: The US Safety Net Collaborative-HCC database was queried. Patients were stratified by facility of diagnosis (SNH or AMC). Patient demographics and HCC screening rates were examined. The primary outcome was stage at diagnosis (AJCC I/II-"early"; AJCC III/IV-"late"). RESULTS: 1290 patients were included; 50.2% diagnosed at SNHs and 49.8% at AMCs. At SNHs, 44.4% of patients were diagnosed late, compared to 27.6% at AMCs. On multivariable regression, Black race was associated with late diagnosis in both facilities (SNH: odds ratio 1.96, p = 0.03; AMC: 2.27, <0.01). Screening was associated with decreased odds of late diagnosis (SNH: 0.46, p = 0.04; AMC: 0.37, p < 0.01). CONCLUSIONS: Black race was associated with late diagnosis of HCC, while screening was associated with early diagnosis across institutional types. These results suggest socially constructed racial bias in screening and diagnosis of HCC. Screening efforts targeting SNH patients and Black patients at all facilities are essential to reduce disparities.

RESUMEN

Schizotypy involves schizophrenia-like traits and symptoms, with the Wisconsin Schizotypy Scales (WSS) being frequently used in previous research. There is some evidence that schizophrenia-spectrum symptom levels, including when using the WSS, might vary both by gender and by race and ethnicity. However, previous research has rarely examined to what extent the WSS show gender and racial bias. Further, this has not been previously examined for the Cognitive Slippage Scale, a measure of disorganized schizotypy. In this study, we examined biases for a subset of items from the WSS in a large sample of undergraduate students (n = 21,829). Using item response theory to test for levels of differential item functioning (DIF), we found some evidence of problematic DIF for all scales, including for negative, positive, and disorganized schizotypy scales. There was evidence of problematic DIF especially by gender and for Black and Multiracial participants. Overall, the current results suggest that gender and/or racial bias on these scales should be an important consideration in using these scales and our results could have implications for assessment of schizophrenia-spectrum symptoms.

Asunto(s)

RESUMEN

Significance: Our goal is to understand the root cause of reported oxygen saturation ( SpO 2 ) overestimation in heavily pigmented skin types to devise solutions toward enabling equity in pulse oximeter designs. Aim: We aim to gain theoretical insights into the effect of skin tone on SpO 2 - R curves using a three-dimensional, four-layer tissue model representing a finger. Approach: A finger tissue model, comprising the epidermis, dermis, two arteries, and a bone, was developed using a Monte Carlo-based approach in the MCmatlab software. Two skin tones-light and dark-were simulated by adjusting the absorption and scattering properties within the epidermal layer. Following this, SpO 2 - R curves were generated in various tissue configurations, including transmission and reflection modes using red and infrared wavelengths. In addition, the influence of source-detector (SD) separation distances on both light and dark skin tissue models was studied. Results: In transmission mode, SpO 2 - R curves did not deviate with changes in skin tones because both pulsatile and non-pulsatile terms experienced equal attenuation at red and infrared wavelengths. However, in reflection mode, measurable variations in SpO 2 - R curves were evident. This was due to differential attenuation of the red components, which resulted in a lower perfusion index at the red wavelength in darker skin. As the SD separation increased, the effect of skin tone on SpO 2 - R curves in reflection mode became less pronounced, with the largest SD separation exhibiting effects similar to those observed in transmission mode. Conclusions: Monte Carlo simulations have demonstrated that different light pathlengths within the tissue contribute to the overestimation of SpO 2 in people with darker skin in reflection mode pulse oximetry. Increasing the SD separation may mitigate the effect of skin tone on SpO 2 readings. These trends were not observed in transmission mode; however, further planned research using more complex models of the tissue is essential.

Asunto(s)

RESUMEN

OBJECTIVE: Rates of ADHD are lowest among Asian American children (1-6.1%) compared to all other major ethnic and racial groups in the US, but there is limited literature on reasons for the disparity in estimated prevalence rates. METHOD: We conducted a narrative review to integrate the literature on ADHD in children in Asian countries with that on ADHD among Asian American youth to highlight potential explanations for disparities in ADHD diagnosis and treatment among Asian American children relative to other racial and ethnic groups. RESULTS: Factors possibly contributing to the low estimated prevalence rates of ADHD among Asian American children include: a higher proportion of Inattentive ADHD presentation among Chinese, Malaysian, and Indian children; racial bias and the influence of the Model Minority Myth; cultural differences in classroom identification; mental health stigma in Asian American communities; parent perception of ADHD as misbehavior rather than a neurodevelopmental disorder; and parent support for children's academic activities that may mask impairment. CONCLUSION: We offer recommendations to inform individual and community-level psychoeducation, and new directions for research to address this health disparity.

Asunto(s)

RESUMEN

BACKGROUND: In mild to moderate lumbar spinal stenosis (LSS) where conservative care treatments fail, minimally invasive treatments, such as interspinous spacers without decompression or fusion (ISD), may be appropriate. While previous studies have demonstrated racial and socioeconomic disparities in the surgical treatment of LSS, there are limited data on how those factors impact accessibility to these procedures. This study explored demographic, socioeconomic, and geographic differences in the use of ISD. METHODS: Using the Medicare 100% files from 2017 through 2022, this retrospective claims analysis identified when and if patients diagnosed with LSS received ISD implantation. Cox proportional hazards regression was used to examine the association between racial and socioeconomic characteristics and the rate of ISD implantation, stratified by geographic region. RESULTS: A total of 1,316,622 individuals met the inclusion criteria; 4730 (0.4%) underwent ISD implantation, with a mean (standard deviation) time to treatment of 11.9 (13.2) months after diagnosis. The likelihood of ISD implantation was higher for older patients (except for the oldest group), males, those with lower disease burden, and White patients. Cox regression revealed that the associations of racial and socioeconomic factors with ISD implantation varied by U.S. region. In the Midwest and Northeast, lower median household income was associated with a decreased likelihood of ISD implantation regardless of race, while in the South, Black patients were less likely to undergo ISD implantation regardless of income. CONCLUSIONS: The observed disparities in access to ISD implantation mirror existing trends in surgical interventions for LSS, suggesting further study and interventions are needed to address inequities.

RESUMEN

Watching movies is among the most popular entertainment and cultural activities. How do viewers react when a movie sequel increases racial minority actors in the main cast ("minority increase")? On the one hand, such sequels may receive better evaluations if viewers appreciate racially inclusive casting for its novel elements (the value-in-diversity perspective) and moral appeal (the fairness perspective on diversity). On the other hand, discrimination research suggests that if viewers harbor biases against racial minorities, sequels with minority increase may receive worse evaluations. To examine these competing possibilities, we analyze a unique panel dataset of movie series released from 1998 to 2021 and conduct text analysis of 312,457 reviews of these movies. Consistent with discrimination research, we find that movies with minority increase receive lower ratings and more toxic reviews. Importantly, these effects weaken after the advent of the Black Lives Matter (BLM) movement, especially when the movement's intensity is high. These results are reliable across various robustness checks (e.g., propensity score matching, random implementation test). We conceptually replicate the bias mitigation effect of BLM in a preregistered experiment: Heightening the salience of BLM increases White individuals' acceptance of racial minority increase in a movie sequel. This research demonstrates the power of social movements in fostering diversity, equality, and inclusion.

Asunto(s)

RESUMEN

We examined whether the enforcement of phantom rules-frequently broken and rarely enforced codified rules-varies by the race of the rule breaker. First, we analyzed whether race affects when 311 calls, a nonemergency service, end in arrest in New York City. Across 10 years, we found that calls from census blocks of neighborhoods consisting of mostly White individuals were 65% less likely to escalate to arrest than those where White people were the numerical minority. Next, we experimentally manipulated transgressor race and found that participants (N = 393) who were high in social dominance orientation were more likely to route 311 calls to 911 when the transgressor was Black (vs. White). We also explored the subjective experience of phantom rule enforcement; People of color report they are more likely to be punished for violating phantom rules compared to White people. Overall, we find evidence of racism in the enforcement of phantom rules.

RESUMEN

BACKGROUND: Over recent years, technological advances in wearables have allowed for continuous home monitoring of heart rate and oxygen saturation. These devices have primarily been used for sports and general wellness and may not be suitable for medical decision-making, especially in saturations below 90% and in patients with dark skin color. Wearable clinical-grade saturation of peripheral oxygen (SpO2) monitoring can be of great value to patients with chronic diseases, enabling them and their clinicians to better manage their condition with reliable real-time and trend data. OBJECTIVE: This study aimed to determine the SpO2 accuracy of a wearable ring pulse oximeter compared with arterial oxygen saturation (SaO2) in a controlled hypoxia study based on the International Organization for Standardization (ISO) 80601-2-61:2019 standard over the range of 70%-100% SaO2 in volunteers with a broad range of skin color (Fitzpatrick I to VI) during nonmotion conditions. In parallel, accuracy was compared with a calibrated clinical-grade reference pulse oximeter (Masimo Radical-7). Acceptable medical device accuracy was defined as a maximum of 4% root mean square error (RMSE) per the ISO 80601-2-61 standard and a maximum of 3.5% RMSE per the US Food and Drug Administration guidance. METHODS: We performed a single-center, blinded hypoxia study of the test device in 11 healthy volunteers at the Hypoxia Research Laboratory, University of California at San Francisco, under the direction of Philip Bickler, MD, PhD, and John Feiner, MD. Each volunteer was connected to a breathing apparatus for the administration of a hypoxic gas mixture. To facilitate frequent blood gas sampling, a radial arterial cannula was placed on either wrist of each participant. One test device was placed on the index finger and another test device was placed on the fingertip. SaO2 analysis was performed using an ABL-90 multi-wavelength oximeter. RESULTS: For the 11 participants included in the analysis, there were 236, 258, and 313 SaO2-SpO2 data pairs for the test device placed on the finger, the test device placed on the fingertip, and the reference device, respectively. The RMSE of the test device for all participants was 2.1% for either finger or fingertip placement, while the Masimo Radical-7 reference pulse oximeter RMSE was 2.8%, exceeding the standard (4% or less) and the Food and Drug Administration guidance (3.5% or less). Accuracy of SaO2-SpO2 paired data from the 4 participants with dark skin in the study was separately analyzed for both test device placements and the reference device. The test and reference devices exceeded the minimum accuracy requirements for a medical device with RMSE at 1.8% (finger) and 1.6% (fingertip) and for the reference device at 2.9%. CONCLUSIONS: The wearable ring meets an acceptable standard of accuracy for clinical-grade SpO2 under nonmotion conditions without regard to skin color. TRIAL REGISTRATION: ClinicalTrials.gov NCT05920278; https://clinicaltrials.gov/study/NCT05920278.

RESUMEN

Recent work has emphasized the disproportionate bias faced by minorities when interacting with law enforcement. However, research on the topic has been hampered by biased sampling in administrative data, namely that records of police interactions with citizens only reflect information on the civilians that police elect to investigate, and not civilians that police observe but do not investigate. In this work, we address a related bias in administrative police data which has received less empirical attention, namely reporting biases around investigations that have taken place. Further, we investigate whether digital monitoring tools help mitigate this reporting bias. To do so, we examine changes in reports of interactions between law enforcement and citizens in the wake of the New York City Police Department's replacement of analog memo books with mobile smartphones. Results from a staggered difference in differences estimation indicate a significant increase in reports of citizen stops once the new smartphones are deployed. Importantly, we observe that the rise is driven by increased reports of "unproductive" stops, stops involving non-White citizens, and stops occurring in areas characterized by a greater concentration of crime and non-White residents. These results reinforce the recent observation that prior work has likely underestimated the extent of racial bias in policing. Further, they highlight that the implementation of digital monitoring tools can mitigate the issue to some extent.

Asunto(s)

RESUMEN

This paper exploits the potential of Global Positioning System datasets sourced from mobile phones to estimate the racial composition of road users, leveraging data from their respective Census block group. The racial composition data encompasses approximately 46 million trips in the Chicago metropolitan region. The research focuses on the relationship between camera tickets and racial composition of drivers vs. police stops for traffic citations and the racial composition in these locations. Black drivers exhibit a higher likelihood of being ticketed by automated speed cameras and of being stopped for moving violations on roads, irrespective of the proportion of White drivers present. The research observes that this correlation attenuates as the proportion of White drivers on the road increases. The citation rate measured by cameras better matches the racial composition of road users on the links with cameras than do stops by police officers. This study therefore presents an important contribution to understanding racial disparities in moving violation stops, with implications for policy interventions and social justice reforms.

RESUMEN

Racial disparities in pediatric pain care are prevalent across a variety of health care settings, and likely contribute to broader disparities in health, morbidity, and mortality. The present research expands on prior work demonstrating potential perceptual contributions to pain care disparities in adults and tests whether racial bias in pain perception extends to child targets. We examined the perception and hypothetical treatment of pain in Black and White boys (experiment 1), Black and White boys and girls (experiment 2), Black and White boys and adult men (experiment 3), and Black, White, Asian, and Latinx boys (experiment 4). Across this work, pain was less readily perceived on Black (vs White) boys' faces-though this bias was not observed within girls. Moreover, this perceptual bias was comparable in magnitude to the same bias measured with adult targets and consistently predicted bias in hypothetical treatment. Notably, bias was not limited to Black targets-pain on Hispanic/Latinx boys' faces was also relatively underperceived. Taken together, these results offer strong evidence for racial bias in pediatric pain perception. PERSPECTIVE: This article demonstrates perceptual contributions to racial bias in pediatric pain recognition. Participants consistently saw pain less readily on Black boys' faces, compared with White boys, and this perceptual bias consistently predicted race-based gaps in treatment. This work reveals a novel factor that may support pediatric pain care disparities.

Asunto(s)

RESUMEN

An extensive literature shows that race information can impact cognitive performance. Two key findings include an attentional bias to Black racial cues in U.S. samples and diminished recognition of other-race faces compared to same-race faces in predominantly White adult samples. Yet face stimuli are increasingly used in psychological research often unrelated to race (Conley et al., 2018) or without consideration for how race information may influence cognitive performance, especially among developmental participants from different racial groups. In the current study we used open-access data from the Adolescent Brain Cognitive DevelopmentSM (ABCD) Study® 4.0.1 release to test for developmentally similar other- and same-race effects of Black and White face stimuli on attention, working memory, and recognition memory in 9- and 10-year-old Black and White children (n=5,659) living in the U.S. Black and White children showed better performance when attending to Black versus White faces. We also show an advantage in recognition memory of same-race compared to other-race faces in White children that did not generalize to Black children. Together the findings highlight how race information, even when irrelevant to an experiment, may indirectly lead to misinterpretation of group differences in cognitive performance in children of different racial backgrounds.

Asunto(s)

RESUMEN

Racial disparities in opioid overdose have increased in recent years. Several studies have linked these disparities to health care providers' inequitable delivery of opioid use disorder (OUD) services. In response, health care policymakers and systems have designed new programs to improve equitable OUD care delivery. Racial bias training has been 1 commonly utilized program. Racial bias training educates providers about the existence of racial disparities in the treatment of people who use drugs and the role of implicit bias. Our study evaluates a pilot racial bias training delivered to 25 hospital emergency providers treating patients with OUDs in 2 hospitals in Detroit, Michigan. We conducted a 3-part survey, including a baseline assessment, post-training assessment, and a 2-month follow-up to evaluate the acceptability and feasibility of scaling the racial bias training to larger audiences. We also investigate preliminary data on changes in self-awareness of implicit bias, knowledge of training content, and equity in care delivery to patients with OUD. Using qualitative survey response data, we found that training participants were satisfied with the content and quality of the training and especially valued the small-group discussions, motivational interviewing, and historical context.

RESUMEN

Natural Language Processing (NLP), a form of Artificial Intelligence, allows free-text based clinical documentation to be integrated in ways that facilitate data analysis, data interpretation and formation of individualized medical and obstetrical care. In this cross-sectional study, we identified all births during the study period carrying the radiology-confirmed diagnosis of fibroid uterus in pregnancy (defined as size of largest diameter of >5 cm) by using an NLP platform and compared it to non-NLP derived data using ICD10 codes of the same diagnosis. We then compared the two sets of data and stratified documentation gaps by race. Using fibroid uterus in pregnancy as a marker, we found that Black patients were more likely to have the diagnosis entered late into the patient's chart or had missing documentation of the diagnosis. With appropriate algorithm definitions, cross referencing and thorough validation steps, NLP can contribute to identifying areas of documentation gaps and improve quality of care.

Asunto(s)

RESUMEN

Deep learning (DL) has gained prominence in healthcare for its ability to facilitate early diagnosis, treatment identification with associated prognosis, and varying patient outcome predictions. However, because of highly variable medical practices and unsystematic data collection approaches, DL can unfortunately exacerbate biases and distort estimates. For example, the presence of sampling bias poses a significant challenge to the efficacy and generalizability of any statistical model. Even with DL approaches, selection bias can lead to inconsistent, suboptimal, or inaccurate model results, especially for underrepresented populations. Therefore, without addressing bias, wider implementation of DL approaches can potentially cause unintended harm. In this paper, we studied a novel method for bias reduction that leverages the frequency domain transformation via the Gerchberg-Saxton and corresponding impact on the outcome from a racio-ethnic bias perspective.

RESUMEN

Pulse oximeters' (POs) varying performance based on skin tones has been highly publicised. Compared to arterial blood gas analysis, POs tend to overestimate oxygen saturation (SpO2) values for people with darker skin (occult hypoxemia). The objective is to develop a test bench for assessing commercial home and hospital-based POs in controlled laboratory conditions. A laboratory simulator was used to mimic different SpO2 values (~ 70 to 100%). Different neutral density and synthetic melanin filters were used to reproduce low signal and varying melanin attenuation levels. Six devices consisting of commercial home (Biolight, N = 13; ChoiceMMed, N = 18; MedLinket, N = 9) and hospital-based (Masimo Radical 7 with Neo L, N = 1; GE B450 Masimo SET with LNCS Neo L, N = 1; Nonin 9550 Onyx II™, N = 1) POs were reviewed and their response documented. Significant variations were observed in the recorded SpO2 values among different POs when exposed to identical simulated signals. Differences were greatest for lower SpO2 (< 80%) where empirical data is limited. All PO responses under low signal and melanin attenuation did not change across various simulated SpO2 values. The bench tests do not provide conclusive evidence that melanin does not affect in vivo SpO2 measurements. Research in the areas of instrument calibration, theory and design needs to be further developed.

RESUMEN

Large language models (LLMs) have the potential to transform our lives and work through the content they generate, known as AI-Generated Content (AIGC). To harness this transformation, we need to understand the limitations of LLMs. Here, we investigate the bias of AIGC produced by seven representative LLMs, including ChatGPT and LLaMA. We collect news articles from The New York Times and Reuters, both known for their dedication to provide unbiased news. We then apply each examined LLM to generate news content with headlines of these news articles as prompts, and evaluate the gender and racial biases of the AIGC produced by the LLM by comparing the AIGC and the original news articles. We further analyze the gender bias of each LLM under biased prompts by adding gender-biased messages to prompts constructed from these news headlines. Our study reveals that the AIGC produced by each examined LLM demonstrates substantial gender and racial biases. Moreover, the AIGC generated by each LLM exhibits notable discrimination against females and individuals of the Black race. Among the LLMs, the AIGC generated by ChatGPT demonstrates the lowest level of bias, and ChatGPT is the sole model capable of declining content generation when provided with biased prompts.

Asunto(s)