Lab Session-4: DH s- AI Bias NotebookLM Activity

Algorithmic Mirrors: Reading Artificial Intelligence Through the Lens of Literary Theory

Artificial Intelligence Is Not Neutral: A Literary Perspective on Algorithmic Bias

Introduction

Artificial Intelligence has become one of the defining technologies of the twenty-first century. From search engines and recommendation systems to generative AI models capable of writing essays, poems, stories, and computer code, AI increasingly mediates the way we acquire knowledge and understand reality. While these technologies are often presented as objective, rational, and free from human prejudice, such assumptions deserve careful examination. Contrary to the popular belief that machines are neutral, Artificial Intelligence reflects the cultural, political, historical, and ideological assumptions embedded within the data used to train it.

As students of English Literature, we are trained to question texts rather than merely consume them. Literary theories have long taught us that every text carries hidden ideologies, power structures, and assumptions about society. If AI generates language, then AI-generated content can also be treated as a text that requires interpretation. This insight forms the central argument of Professor Dilip P. Barad's lecture on Artificial Intelligence and literary criticism: literary theories are not outdated academic exercises but powerful analytical tools for understanding algorithmic bias.

• VIdeo Overview:

This blog explores how feminist criticism, postcolonial studies, critical race theory, and discourse analysis help us decode the hidden assumptions within AI systems. Rather than viewing Artificial Intelligence as an all-knowing machine, this article argues that AI functions as an algorithmic mirror, reflecting both the brilliance and the prejudices of human civilization.

Artificial Intelligence as an Algorithmic Mirror

Popular discussions frequently describe AI as a revolutionary intelligence capable of replacing human creativity. However, such descriptions often exaggerate its capabilities. Large Language Models (LLMs), including modern conversational AI systems, do not possess human consciousness, emotions, intentions, or lived experience. Instead, they function by identifying statistical relationships between words based on enormous collections of digital text.

Researchers Emily M. Bender and her colleagues famously described these systems as "stochastic parrots." The expression emphasizes that language models predict probable sequences of words without genuinely understanding their meanings. They imitate patterns found within their training data rather than independently discovering truth.

This distinction is significant.

If an AI system is trained on centuries of literature, journalism, social media posts, academic writing, and internet discussions, it inevitably absorbs the assumptions present within those materials. Since human history itself contains inequalities regarding gender, race, class, religion, language, and culture, AI naturally reproduces many of these historical patterns.

Consequently, Artificial Intelligence is less like a flawless calculator and more like a highly polished mirror. It reflects humanity's collective imagination—including both our wisdom and our unconscious prejudices.

The Myth of Technological Neutrality

Technology is often portrayed as objective because computers perform mathematical calculations. However, algorithms do not create themselves. Human beings decide:

• what data should be collected,

• whose voices should be included,

• which languages deserve representation,

• how responses should be ranked,

• what information should be filtered or restricted.

Every one of these decisions involves values rather than pure mathematics.

Professor Dilip P. Barad argues that bias is not simply a technical error waiting to be removed through larger datasets. More data does not necessarily create greater fairness. Instead, if the additional data mostly originates from dominant cultures, dominant languages, or privileged communities, the existing bias becomes even stronger.

This explains why AI frequently reproduces what society already considers "normal."

Literary criticism has spent decades exposing precisely these invisible assumptions.

Literary Theory as a Tool for Reading AI

Literary theories encourage readers to move beyond the surface of a text. Instead of asking only What does this text say?, critics ask deeper questions:

• YouTube Short VIdeo Overview:

• Who is speaking?

• Whose voices are missing?

• Which identities are privileged?

• What assumptions appear natural?

• Which ideologies remain invisible?

These same questions can be directed toward AI-generated responses.

Rather than treating AI as an unquestionable authority, literary criticism transforms users into critical readers capable of identifying hidden structures of power.

Gender Bias Through Feminist Criticism

One of the clearest examples of algorithmic bias emerges when examining gender representation.

Sandra Gilbert and Susan Gubar's influential work The Madwoman in the Attic argues that nineteenth-century literature frequently confined women to two stereotypes:

• the obedient "angel,"

• the irrational or rebellious "monster."

This binary shaped literary imagination for generations.

When AI systems receive prompts such as:

"Write a story about a brilliant Victorian scientist."

they often produce a male protagonist without being instructed to do so. The scientist becomes someone like Dr. Edmund Bellam rather than a woman scientist.

This is not because AI believes men are inherently more intelligent. Instead, historical literature overwhelmingly portrayed scientists, philosophers, explorers, and inventors as male. Since AI predicts statistically common patterns, it reproduces patriarchal history.

Similarly, gothic fiction prompts frequently generate helpless heroines awaiting rescue or emotionally unstable women descending into madness. These outputs demonstrate how literary stereotypes continue to influence machine-generated narratives.

Encouragingly, newer AI models sometimes generate courageous, intellectually independent female protagonists, suggesting that algorithmic behaviour can evolve when training data becomes more inclusive.

Nevertheless, feminist criticism reminds us that representation matters because repeated narratives gradually become perceived as natural truth.

Critical Race Theory and Algorithmic Racism

Bias extends beyond gender.

Scholars such as Timnit Gebru, Joy Buolamwini, and Safiya Noble have demonstrated that AI systems often privilege whiteness as the default standard.

Joy Buolamwini's research on facial recognition technologies revealed dramatically higher error rates for darker-skinned women compared to lighter-skinned men. These disparities emerged because the datasets contained disproportionately more images of white individuals.

Safiya Noble's Algorithms of Oppression further demonstrates how search engines may reinforce racist stereotypes by prioritizing discriminatory search results.

From a literary perspective, this resembles postcolonial criticism.

For centuries, colonial literature portrayed Europe as the centre of civilization while representing Africa, Asia, and Latin America as peripheral or "other." If AI trains on historical texts shaped by colonial ideologies, those assumptions inevitably influence its responses.

Therefore, AI does not invent racism independently.

It statistically reproduces existing social hierarchies embedded within digital archives.

Critical race theory helps expose these invisible structures by asking whose histories are remembered and whose experiences remain marginalized.

Political Bias and Algorithmic Guardrails

Another fascinating dimension concerns political control.

Modern AI systems are not only influenced by historical datasets but also by deliberate policy decisions known as guardrails. These safeguards attempt to reduce harmful outputs and comply with legal or organizational requirements, but they can also shape what an AI is willing to discuss.

Professor Barad highlights an illuminating comparison.

When asked to compose satirical poetry criticizing political leaders, OpenAI-generated responses may critique public figures such as Donald Trump or Vladimir Putin. In contrast, reports have shown that DeepSeek readily criticizes Western leaders while refusing to generate similar content regarding Chinese President Xi Jinping or discussions of the Tiananmen Square protests, often stating that such topics fall outside its scope.

This comparison demonstrates two distinct forms of bias:

1. Inherited historical bias emerging naturally from training data.

2. Deliberate algorithmic control introduced through reinforcement learning, moderation policies, and governmental regulations.

For literary scholars, these guardrails resemble editorial censorship. Every omission carries interpretive significance because silence itself communicates power.

The Pushpaka Vimana Debate: Is AI Biased Against Indigenous Knowledge?

An especially thought-provoking question concerns Indian Knowledge Systems.

Many users argue that AI unfairly dismisses concepts such as the Pushpaka Vimana from the Ramayana by describing them as mythology instead of historical scientific inventions.

However, Professor Barad proposes a more careful method for testing algorithmic bias.

Instead of asking whether AI labels Pushpaka Vimana as mythical, we should ask how it treats comparable traditions from other civilizations.

                 (Here is the mindmap to understand the whole blog)

If AI classifies:

• Greek flying chariots,

• Norse flying vehicles,

• Mesopotamian mythological transport,

• Egyptian divine aircraft,

as mythology according to the same evidentiary standards, then it is applying a consistent historical framework rather than discriminating specifically against Indian culture.

The issue, therefore, shifts from cultural prejudice to epistemology—that is, the standards by which knowledge is evaluated.

This approach illustrates an essential principle of literary criticism:

fair interpretation depends upon consistency rather than selective judgment.

The Danger of a Single Story

Nigerian author Chimamanda Ngozi Adichie famously warned against "the danger of a single story."

When only one narrative dominates public imagination, entire communities become reduced to stereotypes.

Artificial Intelligence magnifies this problem.

If the majority of available digital content about Africa, Indigenous communities, or regional Indian cultures originates from outsiders, AI learns incomplete representations.

The machine cannot generate narratives it has never encountered.

Consequently, minority communities become vulnerable to digital erasure.

The solution is not merely criticizing AI but transforming the datasets upon which AI depends.

From Passive Downloaders to Active Uploaders

Professor Barad concludes with an inspiring call to action.

Many societies consume enormous quantities of online information while contributing comparatively little original digital knowledge.

If communities wish AI systems to represent their languages, histories, folklore, philosophies, and literary traditions fairly, they must actively participate in knowledge production.

• Audio Overview in Hindi Language:

This includes:

• contributing to Wikipedia,

• digitizing regional literature,

• publishing academic research,

• translating indigenous texts,

• creating educational resources,

• preserving oral histories,

• expanding digital archives in local languages.

Every uploaded article, digitized manuscript, or translated poem becomes part of the future knowledge ecosystem from which AI learns.

Instead of remaining passive consumers of technology, scholars must become active creators of digital culture.

The Diamond Perspective: Beyond Binary Thinking

Professor Barad also introduces a compelling metaphor.

Complex issues should not be understood as possessing merely "two sides of the coin."

Reality resembles a diamond with countless interconnected surfaces.

Binary thinking encourages simplistic debates:

• AI is good versus AI is bad.

• Technology is neutral versus technology is biased.

• Tradition versus modernity.

The diamond metaphor encourages multidimensional analysis.

Bias is rarely singular.

Gender intersects with race.

Race intersects with language.

Language intersects with colonial history.

Colonial history intersects with technology.

Literary theory equips students to recognize these overlapping structures instead of searching for simplistic answers.

Why English Literature Students Matter in the Age of AI

Some may wonder why literature students should concern themselves with Artificial Intelligence.

The answer is straightforward.

AI increasingly writes stories, translates poetry, summarizes novels, recommends books, and even assists academic research. As machines become producers of language, literary scholars become essential interpreters of machine-generated discourse.

Close reading, discourse analysis, feminist criticism, postcolonial theory, deconstruction, and reader-response criticism provide sophisticated methods for examining AI outputs.

Rather than replacing literary studies, Artificial Intelligence expands their relevance.

The future requires not only engineers who build algorithms but also humanists capable of questioning the ethical, cultural, and ideological assumptions hidden within those algorithms.

Slide share: Sandipkumar Jethava

Conclusion:

Artificial Intelligence is neither a flawless oracle nor an impartial judge. It is an evolving cultural technology that reflects humanity's accumulated knowledge alongside its accumulated prejudices. Like literature, AI-generated language contains voices, silences, assumptions, hierarchies, and ideologies waiting to be interpreted.

Professor Dilip P. Barad's insights demonstrate that literary criticism offers indispensable tools for understanding algorithmic bias. Feminist criticism exposes patriarchal defaults; critical race theory uncovers racial inequalities; postcolonial studies reveal Eurocentric assumptions; and discourse analysis questions political censorship and invisible power structures.

Ultimately, the responsibility does not belong solely to AI developers. It belongs equally to educators, researchers, writers, and ordinary users. By producing diverse digital content, preserving indigenous knowledge, questioning dominant narratives, and cultivating critical reading practices, we contribute to a more inclusive digital future.

Bias may never disappear entirely. Yet the objective is not to eliminate every perspective but to ensure that no single perspective becomes invisible, unquestioned, and accepted as universal truth.

As literature students, our greatest responsibility is therefore not simply to ask what AI says, but to ask why it says it, whose voices it echoes, whose stories it omits, and how we can reshape the narratives that future algorithms will inherit.

Suggested References (MLA 9th Edition)

Adichie, Chimamanda Ngozi. The Danger of a Single Story. TEDGlobal, 2009.

Bender, Emily M., et al. "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 2021.

Buolamwini, Joy, and Timnit Gebru. "Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification." Proceedings of Machine Learning Research, vol. 81, 2018.

Gilbert, Sandra M., and Susan Gubar. The Madwoman in the Attic. Yale University Press, 1979.

Noble, Safiya Umoja. Algorithms of Oppression: How Search Engines Reinforce Racism. New York University Press, 2018.

Crawford, Kate. Atlas of AI. Yale University Press, 2021.

Barad, Dilip P. Lecture on Artificial Intelligence, Literary Theory and Algorithmic Bias. Department of English, Maharaja Krishnakumarsinhji Bhavnagar University.