In a series of interviews, we're meeting some of the AAAI/SIGAI Doctoral Consortium participants to find out more about their research. In this latest interview, we hear from Mahammed Kamruzzaman, who is looking into biases in large language models. We find out about his research so far during the PhD, what he is planning to investigate next, and what inspired him to focus on this aspect of the field. I am currently pursuing my PhD at the University of South Florida in the Department of Computer Science and Engineering. My research focuses on understanding and mitigating biases in Large Language Models (LLMs), particularly how these biases manifest across various sociodemographic and cultural dimensions.

Measurement systems (e.g., currencies) differ across cultures, but the conversions between them are well defined so that humans can state facts using any measurement system of their choice. Being available to users from diverse cultural backgrounds, large language models (LLMs) should also be able to provide accurate information irrespective of the measurement system at hand. Using newly compiled datasets we test if this is the case for seven open-source LLMs, addressing three key research questions: (RQ1) What is the default system used by LLMs for each type of measurement? (RQ2) Do LLMs' answers and their accuracy vary across different measurement systems? (RQ3) Can LLMs mitigate potential challenges w.r.t. underrepresented systems via reasoning? Our findings show that LLMs default to the measurement system predominantly used in the data. Additionally, we observe considerable instability and variance in performance across different measurement systems. While this instability can in part be mitigated by employing reasoning methods such as chain-of-thought (CoT), this implies longer responses and thereby significantly increases test-time compute (and inference costs), marginalizing users from cultural backgrounds that use underrepresented measurement systems.

This study challenges the widely-reported tradeoff between generative AI's (GenAI) contribution to creative outcomes and decreased diversity of these outcomes. We modified the design of such a study, by Doshi and Hauser (2024), in which participants wrote short stories either aided or unaided by GenAI plot ideas[1]. In the modified study, plot ideas were generated through ten unique GenAI "personas" with diverse traits (e.g. cultural backgrounds, thinking styles, genre preferences), creating a pool of 300 story plots. While plot ideas from any individual persona showed high similarity (average cosine similarity of 0.92), ideas across different personas exhibited substantial variation (average similarity of 0.20). When human participants wrote stories based on these diverse plot ideas, their collective outputs maintained the same level of diversity as stories written without GenAI assistance, effectively eliminating the diversity reduction observed in [1]. Traditional text analytics further revealed that GenAI-assisted stories featured greater diversity in descriptive and emotional language compared to purely human-generated stories without GenAI assistance. Our findings demonstrate that introducing diversity at the AI input stage through distinct personas can preserve and potentially enhance the collective diversity of human creative outputs when collaborating with GenAI.

Laughter serves as a multifaceted communicative signal in human interaction, yet its identification within dialogue presents a significant challenge for conversational AI systems. This study addresses this challenge by annotating laughable contexts in Japanese spontaneous text conversation data and developing a taxonomy to classify the underlying reasons for such contexts. Initially, multiple annotators manually labeled laughable contexts using a binary decision (laughable or non-laughable). Subsequently, an LLM was used to generate explanations for the binary annotations of laughable contexts, which were then categorized into a taxonomy comprising ten categories, including "Empathy and Affinity" and "Humor and Surprise," highlighting the diverse range of laughter-inducing scenarios. The study also evaluated GPT-4's performance in recognizing the majority labels of laughable contexts, achieving an F1 score of 43.14%. These findings contribute to the advancement of conversational AI by establishing a foundation for more nuanced recognition and generation of laughter, ultimately fostering more natural and engaging human-AI interactions.

Large vision-language models (VLMs) can assist visually impaired people by describing images from their daily lives. Current evaluation datasets may not reflect diverse cultural user backgrounds or the situational context of this use case. To address this problem, we create a survey to determine caption preferences and propose a culture-centric evaluation benchmark by filtering VizWiz, an existing dataset with images taken by people who are blind. We then evaluate several VLMs, investigating their reliability as visual assistants in a culturally diverse setting. While our results for state-of-the-art models are promising, we identify challenges such as hallucination and misalignment of automatic evaluation metrics with human judgment. We make our survey, data, code, and model outputs publicly available.

Anne Fehres and Luke Conroy & AI4Media / Better Images of AI / Data is a Mirror of Us / Licenced by CC-BY 4.0 Humans are increasingly using artificial intelligence (AI) to inform decisions about our lives. AI is, for instance, helping to make hiring choices and offer medical diagnoses. If you were affected, you might want an explanation of why an AI system produced the decision it did. Yet AI systems are often so computationally complex that not even their designers fully know how the decisions were produced. That's why the development of "explainable AI" (or XAI) is booming.

For synergistic interactions between humans and artificial intelligence (AI) systems, AI outputs often need to be explainable to people. Explainable AI (XAI) systems are commonly tested in human user studies. However, whether XAI researchers consider potential cultural differences in human explanatory needs remains unexplored. We highlight psychological research that found significant differences in human explanations between many people from Western, commonly individualist countries and people from non-Western, often collectivist countries. We argue that XAI research currently overlooks these variations and that many popular XAI designs implicitly and problematically assume that Western explanatory needs are shared cross-culturally. Additionally, we systematically reviewed over 200 XAI user studies and found that most studies did not consider relevant cultural variations, sampled only Western populations, but drew conclusions about human-XAI interactions more generally. We also analyzed over 30 literature reviews of XAI studies. Most reviews did not mention cultural differences in explanatory needs or flag overly broad cross-cultural extrapolations of XAI user study results. Combined, our analyses provide evidence of a cultural bias toward Western populations in XAI research, highlighting an important knowledge gap regarding how culturally diverse users may respond to widely used XAI systems that future work can and should address.

Open-source large language models (LLMs) have gained significant strength across diverse fields. Nevertheless, the majority of studies primarily concentrate on English, with only limited exploration into the realm of multilingual supervised fine-tuning. In this work, we therefore construct an open-source multilingual supervised fine-tuning dataset. Different from previous works that simply translate English instructions, we consider both the language-specific and language-agnostic abilities of LLMs. For language-specific abilities, we introduce a knowledge-grounded data augmentation approach to elicit more culture-specific knowledge of LLMs, improving their ability to serve users from different countries. For language-agnostic abilities, we find through experiments that modern LLMs exhibit strong cross-lingual transfer capabilities, thus repeatedly learning identical content in various languages is not necessary. Consequently, we can substantially prune the language-agnostic SFT data without any performance degradation, making the SFT process more efficient. The resulting UltraLink dataset comprises approximately 1 million samples across five languages, and the proposed data construction method can also be easily extended to other languages. UltraLink-LM, which is trained on UltraLink, outperforms several representative baselines across many tasks.

Online reviews have become essential for users to make informed decisions in everyday tasks ranging from planning summer vacations to purchasing groceries and making financial investments. A key problem in using online reviews is the overabundance of online that overwhelms the users. As a result, recommendation systems for providing helpfulness of reviews are being developed. This paper argues that cultural background is an important feature that impacts the nature of a review written by the user, and must be considered as a feature in assessing the helpfulness of online reviews. The paper provides an in-depth study of differences in online reviews written by users from different cultural backgrounds and how incorporating culture as a feature can lead to better review helpfulness recommendations. In particular, we analyze online reviews originating from two distinct cultural spheres, namely Arabic and Western cultures, for two different products, hotels and books. Our analysis demonstrates that the nature of reviews written by users differs based on their cultural backgrounds and that this difference varies based on the specific product being reviewed. Finally, we have developed six different review helpfulness recommendation models that demonstrate that taking culture into account leads to better recommendations.

Detecting offensive language is a challenging task. Generalizing across different cultures and languages becomes even more challenging: besides lexical, syntactic and semantic differences, pragmatic aspects such as cultural norms and sensitivities, which are particularly relevant in this context, vary greatly. In this paper, we target Chinese offensive language detection and aim to investigate the impact of transfer learning using offensive language detection data from different cultural backgrounds, specifically Korean and English. We find that culture-specific biases in what is considered offensive negatively impact the transferability of language models (LMs) and that LMs trained on diverse cultural data are sensitive to different features in Chinese offensive language detection. In a few-shot learning scenario, however, our study shows promising prospects for non-English offensive language detection with limited resources. Our findings highlight the importance of cross-cultural transfer learning in improving offensive language detection and promoting inclusive digital spaces.