Ramya Srinivasan

Ryosuke Sonoda, Ramya Srinivasan

Large language models (LLMs) such as ChatGPT, GPT-4, Claude-3, and Llama are being integrated across a variety of industries. Despite this rapid proliferation, experts are calling for caution in the interpretation and adoption of LLMs, owing to numerous associated ethical concerns. Research has also uncovered shortcomings in LLMs' reasoning and logical abilities, raising questions on the potential of LLMs as evaluation tools. In this paper, we investigate LLMs' self-evaluation capabilities on a novel proverb reasoning task. We introduce a novel proverb database consisting of 300 proverb pairs that are similar in intent but different in wordings, across topics spanning gender, wisdom, and society. We propose tests to evaluate textual consistencies as well as numerical consistencies across similar proverbs, and demonstrate the effectiveness of our method and dataset in identifying failures in LLMs' self-evaluation which in turn can highlight issues related to gender stereotypes and lack of cultural understanding in LLMs.

Akihito Yoshii, Ryosuke Sonoda, Ramya Srinivasan

Existing deepfake detection methods often exhibit bias, lack transparency, and fail to capture temporal information, leading to biased decisions and unreliable results across different demographic groups. In this paper, we propose a fairness-aware deepfake detection framework that integrates temporal feature learning and demographic-aware data augmentation to enhance fairness and interpretability. Our method leverages sequence-based clustering for temporal modeling of deepfake videos and concept extraction to improve detection reliability while also facilitating interpretable decisions for non-expert users. Additionally, we introduce a demography-aware data augmentation method that balances underrepresented groups and applies frequency-domain transformations to preserve deepfake artifacts, thereby mitigating bias and improving generalization. Extensive experiments on FaceForensics++, DFD, Celeb-DF, and DFDC datasets using state-of-the-art (SoTA) architectures (Xception, ResNet) demonstrate the efficacy of the proposed method in obtaining the best tradeoff between fairness and accuracy when compared to SoTA.

Ramya Srinivasan, Devi Parikh

In recent years, there has been an increased emphasis on understanding and mitigating adverse impacts of artificial intelligence (AI) technologies on society. Across academia, industry, and government bodies, a variety of endeavours are being pursued towards enhancing AI ethics. A significant challenge in the design of ethical AI systems is that there are multiple stakeholders in the AI pipeline, each with their own set of constraints and interests. These different perspectives are often not understood, due in part to communication gaps.For example, AI researchers who design and develop AI models are not necessarily aware of the instability induced in consumers' lives by the compounded effects of AI decisions. Educating different stakeholders about their roles and responsibilities in the broader context becomes necessary. In this position paper, we outline some potential ways in which generative artworks can play this role by serving as accessible and powerful educational tools for surfacing different perspectives. We hope to spark interdisciplinary discussions about computational creativity broadly as a tool for enhancing AI ethics.

Ramya Srinivasan, Ajay Chander

Crowdsourcing information constitutes an important aspect of human-in-the-loop learning for researchers across multiple disciplines such as AI, HCI, and social science. While using crowdsourced data for subjective tasks is not new, eliciting useful insights from such data remains challenging due to a variety of factors such as difficulty of the task, personal prejudices of the human evaluators, lack of question clarity, etc. In this paper, we consider one such subjective evaluation task, namely that of estimating experienced emotions of distressed individuals who are conversing with a human listener in an online coaching platform. We explore strategies to aggregate the evaluators choices, and show that a simple voting consensus is as effective as an optimum aggregation method for the task considered. Intrigued by how an objective assessment would compare to the subjective evaluation of evaluators, we also designed a machine learning algorithm to perform the same task. Interestingly, we observed a machine learning algorithm that is not explicitly modeled to characterize evaluators' subjectivity is as reliable as the human evaluation in terms of assessing the most dominant experienced emotions.