This study introduces a Probabilistic Graphical Model (PGM) framework utilizing Pearl's do-operator to causally audit LLM safety mechanisms, specifically isolating the effect of injecting cultural demographics into prompts. A large-scale empirical analysis was conducted across seven instruction-tuned models from diverse origins, including the UAE's Falcon3-7B, as well as models from the US, Europe, China, and India, using ToxiGen and BOLD datasets. The findings revealed a disparity between observational and interventional bias, demonstrating that standard fairness metrics can overestimate demographic bias. Western models exhibited higher causal refusal rates for specific demographic groups, while Eastern models showed low overall intervention rates with targeted sensitivities toward regional demographics. Why it matters: This research highlights the geopolitical nuances of LLM safety alignment and the potential for demographic-sensitive over-triggering to restrict benign discourse, which is particularly relevant for diverse regions like the Middle East in developing culturally-aware AI.
The paper introduces SectEval, a new benchmark to evaluate sectarian biases in LLMs concerning Sunni and Shia Islam, available in English and Hindi. Results show significant inconsistencies in LLM responses based on language, with some models favoring Shia responses in English but Sunni in Hindi. Location-based experiments further reveal that advanced models adapt their responses based on the user's claimed country, while smaller models exhibit a consistent Sunni-leaning bias.
A new methodology emulating fact-checker criteria assesses news outlet factuality and bias using LLMs. The approach uses prompts based on fact-checking criteria to elicit and aggregate LLM responses for predictions. Experiments demonstrate improvements over baselines, with error analysis on media popularity and region, and a released dataset/code at https://github.com/mbzuai-nlp/llm-media-profiling.
A new dataset called ArabCulture is introduced to address the lack of culturally relevant commonsense reasoning resources in Arabic AI. The dataset covers 13 countries across the Gulf, Levant, North Africa, and the Nile Valley, spanning 12 daily life domains with 54 fine-grained subtopics. It was built from scratch by native speakers writing and validating culturally relevant questions. Why it matters: The dataset highlights the need for more culturally aware models and benchmarks tailored to the Arabic-speaking world, moving beyond machine-translated resources.
Researchers compiled a 101 Billion Arabic Words Dataset by mining text from Common Crawl WET files and rigorously cleaning and deduplicating the extracted content. The dataset aims to address the scarcity of original, high-quality Arabic linguistic data, which often leads to bias in Arabic LLMs that rely on translated English data. This is the largest Arabic dataset available to date. Why it matters: The new dataset can significantly contribute to the development of authentic Arabic LLMs that are more linguistically and culturally accurate.
MBZUAI-based startup LibrAI has launched the Libra-Leaderboard, an evaluation framework for LLMs that assesses both capability and safety. The leaderboard evaluates 26 mainstream LLMs using 57 datasets, assigning scores based on bias, misinformation, and oversensitivity. LibrAI also launched the Interactive Safety Arena to engage the public and educate them on AI safety through adversarial prompt testing. Why it matters: The Libra-Leaderboard provides a benchmark for responsible AI development, emphasizing the importance of aligning AI capabilities with safety considerations in the rapidly evolving LLM landscape.
A study by MBZUAI's Preslav Nakov and Cornell co-authors examines how to develop systems that detect fake news in a landscape where text is generated by humans and machines. The research, presented at the 2024 Annual Conference of the North American Chapter of the Association for Computational Linguistics, analyzes fake news detectors' ability to identify human- and machine-written content. The study highlights biases in current detectors, which tend to classify machine-written news as fake and human-written news as true. Why it matters: Addressing these biases is crucial as machine-generated content becomes more prevalent in both real and fake news, requiring more nuanced detection methods.
MBZUAI and the University of Michigan Ann Arbor have announced a new collaboration in AI research, sponsored by the U.S. Mission to the UAE. The partnership focuses on projects addressing the cultural divide in AI, with research teams from both institutions collaborating throughout the 2023/2024 academic year. A workshop titled “Bridging the Cultural Divide in AI: Analyzing Fairness, Bias, and Transparency across Cultures” will be held on April 29-30. Why it matters: The collaboration strengthens ties between the UAE and the U.S. in AI, addressing critical issues of fairness and cultural sensitivity in AI development.