arXiv subjects

David Rozado

Publications and source records attributed to David Rozado.

5 recordsLinked to original sources

arXiv2025

Gender and Positional Biases in LLM-Based Hiring Decisions: Evidence from Comparative CV/R\'esum\'e Evaluations

This study examines the behavior of Large Language Models (LLMs) when evaluating professional candidates based on their resumes or curricula vitae (CVs). In an experiment involving 22 leading LLMs, each model was systematically given one job description along with a pair of profession-matched CVs, one bearing a male first name, the other a female first name, and asked to select the more suitable candidate for the job. Each CV pair was presented twice, with names swapped to ensure that any observed preferences in candidate selection stemmed from gendered names cues. Despite identical professional qualifications across genders, all LLMs consistently favored female-named candidates across 70 different professions. Adding an explicit gender field (male/female) to the CVs further increased the preference for female applicants. When gendered names were replaced with gender-neutral identifiers "Candidate A" and "Candidate B", several models displayed a preference to select "Candidate A". Counterbalancing gender assignment between these gender-neutral identifiers resulted in gender parity in candidate selection. When asked to rate CVs in isolation rather than compare pairs, LLMs assigned slightly higher average scores to female CVs overall, but the effect size was negligible. Including preferred pronouns (he/him or she/her) next to a candidate's name slightly increased the odds of the candidate being selected regardless of gender. Finally, most models exhibited a substantial positional bias to select the candidate listed first in the prompt. These findings underscore the need for caution when deploying LLMs in high-stakes autonomous decision-making contexts and raise doubts about whether LLMs consistently apply principled reasoning.

cs.CL↗

arXiv2025

Measuring Political Preferences in AI Systems: An Integrative Approach

Political biases in Large Language Model (LLM)-based artificial intelligence (AI) systems, such as OpenAI's ChatGPT or Google's Gemini, have been previously reported. While several prior studies have attempted to quantify these biases using political orientation tests, such approaches are limited by potential tests' calibration biases and constrained response formats that do not reflect real-world human-AI interactions. This study employs a multi-method approach to assess political bias in leading AI systems, integrating four complementary methodologies: (1) linguistic comparison of AI-generated text with the language used by Republican and Democratic U.S. Congress members, (2) analysis of political viewpoints embedded in AI-generated policy recommendations, (3) sentiment analysis of AI-generated text toward politically affiliated public figures, and (4) standardized political orientation testing. Results indicate a consistent left-leaning bias across most contemporary AI systems, with arguably varying degrees of intensity. However, this bias is not an inherent feature of LLMs; prior research demonstrates that fine-tuning with politically skewed data can realign these models across the ideological spectrum. The presence of systematic political bias in AI systems poses risks, including reduced viewpoint diversity, increased societal polarization, and the potential for public mistrust in AI technologies. To mitigate these risks, AI systems should be designed to prioritize factual accuracy while maintaining neutrality on most lawful normative issues. Furthermore, independent monitoring platforms are necessary to ensure transparency, accountability, and responsible AI development.

cs.CY↗

arXiv2024

The Political Preferences of LLMs

I report here a comprehensive analysis about the political preferences embedded in Large Language Models (LLMs). Namely, I administer 11 political orientation tests, designed to identify the political preferences of the test taker, to 24 state-of-the-art conversational LLMs, both closed and open source. When probed with questions/statements with political connotations, most conversational LLMs tend to generate responses that are diagnosed by most political test instruments as manifesting preferences for left-of-center viewpoints. This does not appear to be the case for five additional base (i.e. foundation) models upon which LLMs optimized for conversation with humans are built. However, the weak performance of the base models at coherently answering the tests' questions makes this subset of results inconclusive. Finally, I demonstrate that LLMs can be steered towards specific locations in the political spectrum through Supervised Fine-Tuning (SFT) with only modest amounts of politically aligned data, suggesting SFT's potential to embed political orientation in LLMs. With LLMs beginning to partially displace traditional information sources like search engines and Wikipedia, the societal implications of political biases embedded in LLMs are substantial.

cs.CY↗

arXiv2023

Mentions of Prejudice in News Media -- An International Comparison

Previous research has identified a post-2010 sharp increase of terms used to denounce prejudice (i.e. racism, sexism, homophobia, Islamophobia, anti-Semitism, etc.) in U.S. and U.K. news media content. Here, we extend previous analysis to an international sample of news media organizations. Thus, we quantify the prevalence of prejudice-denouncing terms and social justice associated terminology (diversity, inclusion, equality, etc.) in over 98 million news and opinion articles across 124 popular news media outlets from 36 countries representing 6 different world regions: English-speaking West, continental Europe, Latin America, sub-Saharan Africa, Persian Gulf region and Asia. We find that the post-2010 increasing prominence in news media of the studied terminology is not circumscribed to the U.S. and the U.K. but rather appears to be a mostly global phenomenon starting in the first half of the 2010s decade in pioneering countries yet largely prevalent around the globe post-2015. However, different world regions' news media emphasize distinct types of prejudice with varying degrees of intensity. We find no evidence of U.S. news media having been first in the world in increasing the frequency of prejudice coverage in their content. The large degree of temporal synchronicity with which the studied set of terms increased in news media across a vast majority of countries raises important questions about the root causes driving this phenomenon.

cs.CY↗

arXiv2020

Wide range screening of algorithmic bias in word embedding models using large sentiment lexicons reveals underreported bias types

This work describes a large-scale analysis of sentiment associations in popular word embedding models along the lines of gender and ethnicity but also along the less frequently studied dimensions of socioeconomic status, age, sexual orientation, religious sentiment and political leanings. Consistent with previous scholarly literature, this work has found systemic bias against given names popular among African-Americans in most embedding models examined. Gender bias in embedding models however appears to be multifaceted and often reversed in polarity to what has been regularly reported. Interestingly, using the common operationalization of the term bias in the fairness literature, novel types of so far unreported bias types in word embedding models have also been identified. Specifically, the popular embedding models analyzed here display negative biases against middle and working-class socioeconomic status, male children, senior citizens, plain physical appearance, Islamic religious faith, non-religiosity and conservative political orientation. Reasons for the paradoxical underreporting of these bias types in the relevant literature are probably manifold but widely held blind spots when searching for algorithmic bias and a lack of widespread technical jargon to unambiguously describe a variety of algorithmic associations could conceivably be playing a role. The causal origins for the multiplicity of loaded associations attached to distinct demographic groups within embedding models are often unclear but the heterogeneity of said associations and their potential multifactorial roots raises doubts about the validity of grouping them all under the umbrella term bias. Richer and more fine-grained terminology as well as a more comprehensive exploration of the bias landscape could help the fairness epistemic community to characterize and neutralize algorithmic discrimination more efficiently.

cs.CY↗