Document is current

Chapter History

First Online: 28 June 2025

Ethical Considerations

: We caution against relying exclusively on AI-generated predictions of hate speech for high-stakes or real-world applications. Hate speech is inherently subjective and highly context-dependent; even accurate models at the aggregate level may fail to account for cultural nuance, sarcasm, or reclaimed language at the individual level. These risks are amplified when models trained in one sociocultural context are applied to another, where norms and linguistic cues differ []. Furthermore, the opacity of large language models and the lack of transparency in their pretraining data make it difficult to assess how sociocultural priors are encoded or reinforced. To mitigate these risks, we emphasize the need for continual in-context evaluation, supported by human oversight and participatory annotation practices that reflect the values and expectations of affected communities. Transparency, context-awareness, and sustained monitoring are essential for the responsible deployment of LLMs in hate speech detection.

Conference Information

Conference Acronym: ICCSA

Conference Name: International Conference on Computational Science and Its Applications

Conference City: Istanbul

Conference Country: Türkiye

Conference Year: 2025

Conference Start Date: 30 June 2025

Conference End Date: 3 July 2025

Conference Number: 25

Conference ID: iccsa2025

Conference URL: https://iccsa.org/