Algorithmic Summarization Bias: How Automated Summaries Reinforce or Disrupt Political Communication Norms

While much work (e.g., Shukla, 2025) critically unpacks algorithmic and data-driven biases in AI systems, there’s very little empirical research on how automated summarization tools actually shape political communication bias in real-world settings. This project would collect a dataset of political news articles and speeches, then generate summaries using both human editors and leading AI models (such as GPT-4 or custom summarization algorithms). By systematically comparing these summaries for bias (using frameworks from Bolan Tang, 2024, on linguistic mechanisms and agenda setting), we could assess where and how algorithms reinforce dominant narratives, disrupt established framing, or even create new, previously unobserved biases. This would extend the work on algorithmic bias by focusing specifically on summarization—a critical, underexplored process in the information pipeline. The findings could inform both media literacy and the ethical design of AI tools in journalism, potentially reshaping editorial workflows and transparency standards.

References:

1. Role of Language in Emotions and Bias: A Cognitive Linguistic Perspective of Affective Political Media in the US. Bolan Tang (2024). Lecture Notes in Education Psychology and Public Media.
2. Investigating AI systems: examining data and algorithmic bias through hermeneutic reverse engineering. Nishanshi Shukla (2025). Frontiers in Communication.