AI Bias in Indian Political Narratives: Focus on Gen Z's Perception Impact

This study explores the bias manifestations of large language models (LLMs) in the Indian political context and their potential impact on Gen Z's political perceptions. As the world's largest democracy and the fastest-growing internet market, India's Gen Z relies on AI to access political information. Understanding LLM biases is of far-reaching significance for maintaining the health of the information ecosystem and democratic fairness. Core questions include: the tendency of LLMs on Indian political topics, Gen Z's perception and trust in AI content, how AI biases shape narratives, and mitigation strategies.

Research Background: The Specificity of India's Political Information Ecosystem in the AI Era

Research Origin

Amid the digital wave, LLMs have become an important channel for information access, but the issue of political bias has emerged. India has a population of 1.4 billion and a diverse political spectrum; as digital natives, Gen Z relies on AI, and their political perceptions are easily influenced by AI.

Characteristics of the Indian Political Context

• Diverse political spectrum: Multi-party coalition system, requiring understanding of different party positions; religion, caste, and politics are intertwined, avoiding stereotypes; 22 official languages, with large regional political differences.
• Training data issues: Mainstream LLM training data is dominated by English, with insufficient representation of Indian political content (sources biased towards urban elites, historical imbalance, and single perspectives).

Research Methods: Combining Multi-dimensional Detection and User Research

Multi-dimensional Bias Detection Framework

1. Position analysis: Design prompts on Indian political issues to identify the value orientations of responses;
2. Sentiment polarity analysis: Evaluate emotional differences towards different political figures/parties/policies;
3. Frame analysis: Examine narrative frames (e.g., security/economy/human rights);
4. Information balance: Check whether multiple perspectives are provided.

Gen Z User Research

Collect via questionnaires and interviews: users' ability to identify AI biases, trust in AI political information, and the impact of AI content on opinion formation and sharing.

Preliminary Findings: LLM Bias Manifestations and Gen Z's Perception Characteristics

LLM Bias Patterns

1. Western-centrism: Analyze Indian politics using Western frameworks, ignoring caste/language politics;
2. English elite bias: More citations from English media, lack of local language/regional perspectives;
3. Avoidance of sensitive topics: Overly cautious or falsely neutral on controversial topics like Kashmir;
4. Bias in evaluating historical figures: Influenced by contemporary discourse, lack of objective analysis with historical context.

Gen Z's Perception Characteristics

• High reliance on AI but low alertness to biases;
• Reinforcement of confirmation bias (accept if agree, question if not);
• Desire for AI transparency (training data, update frequency, etc.).

Research Significance: Academic Contributions and Policy Implications

Academic Contributions

• Regional expansion: Fill research gaps outside Western contexts (the largest democracy in the Global South);
• Generational perspective: Focus on Gen Z's political socialization as AI natives;
• Methodological innovation: Cross computational linguistics and social science qualitative research.

Policy Implications

• Localization requirements: AI services need to consider the specificity of local political culture;
• Transparency standards: Disclose training data sources and bias risks;
• Education investment: Strengthen civic AI literacy education.

Global Perspective

India's case provides reference for other developing countries.

Mitigation Strategies: Recommendations at Technical and Educational Levels

Technical Level

1. Diversify training data: Increase Indian local language resources, content with diverse political positions, and grassroots organization materials;
2. Context-aware prompt engineering: Guide LLMs to acknowledge the complexity of Indian politics, provide multiple perspectives, and label uncertainties;
3. Bias detection tools: Establish continuous monitoring mechanisms and disclose results.

Educational Level

• AI literacy education: Incorporate into school curricula to cultivate critical thinking;
• Platform responsibility: Provide bias warnings and usage guidelines (especially for political topics).

Limitations and Future Research Directions

Current Limitations

• Sample scope: Not covering all regions/language communities;
• Model updates: LLM iterations may invalidate results;
• Causal inference: Correlation does not equal causation, requiring stricter experiments.

Future Directions

• Longitudinal tracking: Long-term observation of the cumulative impact of AI on users' political views;
• Cross-cultural comparison: Compare with democratic countries like Brazil and Indonesia;
• Intervention experiments: Test the effectiveness of educational interventions to improve AI literacy.