Cultural Bias and Cultural Alignment of Large Language Models

• URL: http://arxiv.org/abs/2311.14096v2
• Date: Wed, 26 Jun 2024 15:26:44 GMT
• Title: Cultural Bias and Cultural Alignment of Large Language Models
• Authors: Yan Tao, Olga Viberg, Ryan S. Baker, Rene F. Kizilcec,
• Abstract summary: We conduct a disaggregated evaluation of cultural bias for five widely used large language models. All models exhibit cultural values resembling English-speaking and Protestant European countries. We suggest using cultural prompting and ongoing evaluation to reduce cultural bias in the output of generative AI.
• Score: 0.9374652839580183
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Culture fundamentally shapes people's reasoning, behavior, and communication. As people increasingly use generative artificial intelligence (AI) to expedite and automate personal and professional tasks, cultural values embedded in AI models may bias people's authentic expression and contribute to the dominance of certain cultures. We conduct a disaggregated evaluation of cultural bias for five widely used large language models (OpenAI's GPT-4o/4-turbo/4/3.5-turbo/3) by comparing the models' responses to nationally representative survey data. All models exhibit cultural values resembling English-speaking and Protestant European countries. We test cultural prompting as a control strategy to increase cultural alignment for each country/territory. For recent models (GPT-4, 4-turbo, 4o), this improves the cultural alignment of the models' output for 71-81% of countries and territories. We suggest using cultural prompting and ongoing evaluation to reduce cultural bias in the output of generative AI.

Related papers

• CAReDiO: Cultural Alignment of LLM via Representativeness and Distinctiveness Guided Data Optimization [50.90288681622152]
  Large Language Models (LLMs) more deeply integrate into human life across various regions. Existing approaches develop culturally aligned LLMs through fine-tuning with culture-specific corpora. We introduce CAReDiO, a novel cultural data construction framework.
  arXiv  Detail & Related papers  (2025-04-09T13:40:13Z)
• CultureVLM: Characterizing and Improving Cultural Understanding of Vision-Language Models for over 100 Countries [63.00147630084146]
  Vision-language models (VLMs) have advanced human-AI interaction but struggle with cultural understanding. CultureVerse is a large-scale multimodal benchmark covering 19, 682 cultural concepts, 188 countries/regions, 15 cultural concepts, and 3 question types. We propose CultureVLM, a series of VLMs fine-tuned on our dataset to achieve significant performance improvement in cultural understanding.
  arXiv  Detail & Related papers  (2025-01-02T14:42:37Z)
• Risks of Cultural Erasure in Large Language Models [4.613949381428196]
  We argue for the need of metricizable evaluations of language technologies that interrogate and account for historical power inequities. We probe representations that a language model produces about different places around the world when asked to describe these contexts. We analyze the cultures represented in the travel recommendations produced by a set of language model applications.
  arXiv  Detail & Related papers  (2025-01-02T04:57:50Z)
• Attributing Culture-Conditioned Generations to Pretraining Corpora [26.992883552982335]
  We analyze how models associate entities with cultures based on pretraining data patterns. We find that high-frequency cultures in pretraining data yield more generations with memorized symbols, while some low-frequency cultures produce none.
  arXiv  Detail & Related papers  (2024-12-30T07:09:25Z)
• Navigating the Cultural Kaleidoscope: A Hitchhiker's Guide to Sensitivity in Large Language Models [4.771099208181585]
  LLMs are increasingly deployed in global applications, ensuring users from diverse backgrounds feel respected and understood. Cultural harm can arise when these models fail to align with specific cultural norms, resulting in misrepresentations or violations of cultural values. We present two key contributions: A cultural harm test dataset, created to assess model outputs across different cultural contexts through scenarios that expose potential cultural insensitivities, and a culturally aligned preference dataset, aimed at restoring cultural sensitivity through fine-tuning based on feedback from diverse annotators.
  arXiv  Detail & Related papers  (2024-10-15T18:13:10Z)
• Extrinsic Evaluation of Cultural Competence in Large Language Models [53.626808086522985]
  We focus on extrinsic evaluation of cultural competence in two text generation tasks. We evaluate model outputs when an explicit cue of culture, specifically nationality, is perturbed in the prompts. We find weak correlations between text similarity of outputs for different countries and the cultural values of these countries.
  arXiv  Detail & Related papers  (2024-06-17T14:03:27Z)
• CulturePark: Boosting Cross-cultural Understanding in Large Language Models [63.452948673344395]
  This paper introduces CulturePark, an LLM-powered multi-agent communication framework for cultural data collection. It generates high-quality cross-cultural dialogues encapsulating human beliefs, norms, and customs. We evaluate these models across three downstream tasks: content moderation, cultural alignment, and cultural education.
  arXiv  Detail & Related papers  (2024-05-24T01:49:02Z)
• CultureBank: An Online Community-Driven Knowledge Base Towards Culturally Aware Language Technologies [53.2331634010413]
  CultureBank is a knowledge base built upon users' self-narratives. It contains 12K cultural descriptors sourced from TikTok and 11K from Reddit. We offer recommendations for future culturally aware language technologies.
  arXiv  Detail & Related papers  (2024-04-23T17:16:08Z)
• CULTURE-GEN: Revealing Global Cultural Perception in Language Models through Natural Language Prompting [73.94059188347582]
  We uncover culture perceptions of three SOTA models on 110 countries and regions on 8 culture-related topics through culture-conditioned generations. We discover that culture-conditioned generation consist of linguistic "markers" that distinguish marginalized cultures apart from default cultures.
  arXiv  Detail & Related papers  (2024-04-16T00:50:43Z)
• Massively Multi-Cultural Knowledge Acquisition & LM Benchmarking [48.21982147529661]
  This paper introduces a novel approach for massively multicultural knowledge acquisition. Our method strategically navigates from densely informative Wikipedia documents on cultural topics to an extensive network of linked pages. Our work marks an important step towards deeper understanding and bridging the gaps of cultural disparities in AI.
  arXiv  Detail & Related papers  (2024-02-14T18:16:54Z)
• Not All Countries Celebrate Thanksgiving: On the Cultural Dominance in Large Language Models [89.94270049334479]
  This paper identifies a cultural dominance issue within large language models (LLMs) LLMs often provide inappropriate English-culture-related answers that are not relevant to the expected culture when users ask in non-English languages.
  arXiv  Detail & Related papers  (2023-10-19T05:38:23Z)
• Navigating Cultural Chasms: Exploring and Unlocking the Cultural POV of Text-To-Image Models [32.99865895211158]
  We explore the cultural perception embedded in Text-To-Image (TTI) models by characterizing culture across three tiers. We propose a comprehensive suite of evaluation techniques, including intrinsic evaluations using the CLIP space. To bolster our research, we introduce the CulText2I dataset, derived from six diverse TTI models and spanning ten languages.
  arXiv  Detail & Related papers  (2023-10-03T10:13:36Z)
• Cultural Alignment in Large Language Models: An Explanatory Analysis Based on Hofstede's Cultural Dimensions [10.415002561977655]
  This research proposes a Cultural Alignment Test (Hoftede's CAT) to quantify cultural alignment using Hofstede's cultural dimension framework. We quantitatively evaluate large language models (LLMs) against the cultural dimensions of regions like the United States, China, and Arab countries. Our results quantify the cultural alignment of LLMs and reveal the difference between LLMs in explanatory cultural dimensions.
  arXiv  Detail & Related papers  (2023-08-25T14:50:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.