DH s- AI Bias NotebookLM Activity

This lab activity was assigned by Prof. (Dr.) Dilip Barad Sir (Department of English, Maharaja Krishnakumarsinhji Bhavnagar University) As part of this task, Sir provided us with a video lecture. Using NotebookLM, we are required to generate an infographic, slide deck, mind map, audio overview, video overview, and a brief text summary based on the content of the video lecture.

Blog Included

👉Mind Map

👉Infographic

👉Video Overview

👉Audio + Video *Hindi*

👉Slidedeck

👉Brief Text

Video Lecture Provided for NotebookLM Activity 

The Mind Map is Presented Below

Click Here: Mind Map

Infographic Generated Using NotebookLM

Here is the Cinematic Video Overview

Here is the Audio with Visuals (Hindi)

Here is the Slide Deck

                   

Brief Text

Bias in AI Models and Implications for Literary Interpretation

Artificial Intelligence (AI) is not a neutral or objective technology; rather, it acts as a mirror reflecting the socio-cultural, religious, and political biases inherent in the human-generated data sets used to train it. In the context of literary studies, AI often defaults to dominant cultural narratives, mainstream voices, and patriarchal or Eurocentric perspectives.

Key takeaways from this analysis include:

• The Inevitability of Bias: Bias is a "flow in thinking" guided by mental preconditioning. Because AI is trained on canonical texts and dominant discourses, it frequently reproduces systemic inequalities.

• The Utility of Literary Theory: Frameworks such as feminism, postcolonialism, and critical race theory are essential tools for identifying and "naming" the biases hidden within AI-generated content.

• Deliberate vs. Algorithmic Bias: While some biases are unconscious results of data scale (e.g., gender stereotypes), others are deliberate institutional controls, as evidenced by political censorship in specific regional AI models like DeepSeek.

• From Downloaders to Uploaders: To counter the "Global North" dominance in AI training data, scholars and creators in the "Global South" must transition from being consumers of digital content to active contributors (uploaders) of indigenous knowledge and diverse narratives.

The Nature and Identification of Bias

Bias is defined as the instinctive categorization of people and things without conscious awareness. It is often a result of "mental preconditioning" rather than firsthand experience, where belief systems are frequently confused with knowledge systems.

Strategies for Addressing Bias

1. Recognition: Acknowledge that biases exist in both humans and technology.

2. Critical Inquiry: Use the "diamond metaphor" viewing problems as multi-faceted (3D, 4D, or 5D) rather than having only two sides.

3. Contrary Analysis: Deliberately take antithetical views to challenge assumptions and traditions.

4. Empathy and Diversity: Practice empathy and embrace diverse perspectives to uncover hidden prejudices.

AI Bias through the Lens of Literary Theory

1. Gender Bias and Feminist Criticism

AI models often inherit the "patriarchal canon." Drawing on the work of Gilbert and Gubar (The Madwoman in the Attic), AI-generated narratives frequently default to binary representations of women: the "angelic/submissive" heroine or the "mad/hysterical" monster.

Prompt Type	Typical AI Output/Bias	Potential Progressive Shift
Scientist Story	Defaults to a male protagonist (e.g., "Dr. Edmund Bellamy").	Inclusion of female scientists without explicit prompting.
Victorian Writers	Traditionally focuses on Dickens or Hardy.	Inclusion of figures like Elizabeth Barrett Browning or the Brontës.
Gothic Characters	A "pale, trembling, helpless" girl.	Characters described through "rebellious spirit," "intellect," or "bravery."

2. Racial and Postcolonial Bias

AI algorithms frequently exhibit "Eurocentric beauty ideals" and "stochastic parroting," where large language models (LLMs) amplify existing racial disparities found in their training data.

• Intersectional Disparity: Research (e.g., Timnit Gebru and Joy Buolamwini) indicates that AI classification systems have significantly higher error rates for dark-skinned women compared to white men.

• Erasure of Voices: Western canons foreground white writers, and AI replicates this by marginalizing Black and indigenous voices.

• Search Engine Bias: Safiya Noble’s work on "algorithms of oppression" demonstrates how search engines can reinforce racism through biased results.

3. Political Bias and Institutional Control

A critical distinction exists between the "liberal spirit" of Western-developed models (like OpenAI’s ChatGPT) and the "deliberate control" seen in models like DeepSeek (China).

• Case Study: DeepSeek: While DeepSeek can generate political satire about leaders like Donald Trump, Vladimir Putin, or Kim Jong-un, it frequently refuses to answer questions or generate content regarding the Chinese government or sensitive historical events (e.g., Tiananmen Square), citing a lack of "current scope."

• The "Goody-Goody" Language Trap: Controlled AI models often use terms like "positive developments" or "constructive answers" to mask the suppression of critical or marginalized perspectives.

Environmental and Historical Implications

Eco-criticism

AI tends to provide generic imagery for climate change (e.g., melting glaciers and polar bears) while neglecting regional crises such as:

• Deforestation in the Amazon.

• Displacement in the Sundarbans.

• Droughts in Africa.

New Historicism and Digital Humanities

The "official history" is often privileged over marginalized histories in digital archives.

• Accessibility Bias: Digital Humanities discussions often overlook the lack of infrastructure in developing nations.

• Ethics of Digitization: There is an "open theft" occurring where AI models use copyrighted materials without acknowledgment, raising questions about whose texts are deemed worthy of being digitized and used.

Evaluating Fairness: The Case of Indian Knowledge Systems (IKS)

A major concern in the "Global South" is whether AI is biased against indigenous histories, such as the Pushpaka Vimana (the flying chariot in the Ramayana).

The Standard for Bias:

• If an AI labels the Pushpaka Vimana as a "myth" while treating flying objects in Greek or Norse mythology as "scientific facts," it is biased.

• If the AI treats all such ancient flying objects across cultures consistently as "mythological" or "legendary" rather than "scientific," it is applying a uniform standard rather than a bias.

Conclusion: Making Bias Visible

Neutrality in AI is impossible because meaning is constructed and contextual. The goal of educators and researchers should not be to achieve perfect neutrality, but to make bias visible.

The Path Forward

• Critical Awareness: Use literary theory to name and historicize biases to understand their power effects.

• Algorithmic Consciousness: While AI lacks human consciousness, "algorithmic awareness" can be improved through better prompting and more diverse data.

• Active Contribution: To prevent stereotyping, the "Global South" must be "vocal on digital space." As noted by Chimamanda Ngozi Adichie, the danger of a "single story" is mitigated when there are many stories. Scholars must actively upload regional, indigenous, and marginalized content to ensure AI data sets reflect a truly global reality.