The Ghost in the Mirror: What I Learned About AI Bias from Barad Sir's Lecture
The Ghost in the Mirror: How AI Reflects Our Cultural Prejudices
Introduction
I recently watched a lecture by Barad sir, organised for SRM University Sikkim and introduced by Dr. Padminika. Barad sir has more than 26 years of teaching experience and has worked as an academic dean and NAAC assessor. He is well known for connecting literary theory with new technology, and in this session he showed how Artificial Intelligence carries the same biases that we find in human society. I am writing this post to record what I learned, because I think it is directly useful for us as literature students. As future scholars and educators, understanding how algorithms inherit and amplify human prejudices is no longer optional, it is a fundamental digital literacy.
Video: Barad Sir's Lecture on AI Bias in Literature
What Unconscious Bias Really Means
Barad sir began by explaining unconscious bias with the help of an infographic by Tanmay Vora. He said unconscious bias means we categorise people instinctively, based on mental conditioning, and not on our direct experience of that person. This was an important starting point for me. He said that the whole purpose of studying literature and hermeneutics is to learn how to spot these hidden socio-cultural biases and question them.
He then connected this idea to AI. Since generative AI models are trained on huge amounts of data taken mostly from dominant cultures and standard English, they naturally end up repeating the same cultural biases. Minority voices get pushed to the side. I found this idea simple but powerful. AI is not creating bias out of nothing. It is copying the bias that already exists in the data we feed it.
"AI is not creating bias out of nothing. It is copying the bias that already exists in the data we feed it."
Testing Gender Bias in Real Time
The next part of the lecture was very interesting because Barad sir connected AI bias directly to Sandra Gilbert and Susan Gubar's 1979 book The Madwoman in the Attic. This book argues that patriarchal writing usually shows women in only two extreme ways, either as a submissive "angel" or as a mad "monster". He asked the participants to test this live using AI prompts. By applying a 1979 feminist literary framework to a 21st-century algorithm, sir demonstrated that our theoretical tools are still our best weapons against systemic bias.
🤖 LIVE AI PROMPT RESULTS
Prompt 1: "Write a Victorian story about a scientist"
The AI gave a male protagonist named Dr. Edmund Bellam. This showed that AI still connects scientific intelligence with men by default.
Prompt 2: "List the greatest writers of the Victorian era"
The answer included Charles Dickens and Thomas Hardy, but also the Brontë sisters and Elizabeth Barrett Browning. A good sign—AI datasets are learning from modern feminist criticism, not just old patriarchal texts.
Prompt 3: "Describe a female character in a Gothic novel"
Results were mixed: one got the old trope of a "pale, trembling girl in the dark," another got a "rebellious and brave" character. This mix shows AI is slowly moving away from old patriarchal tropes, but some biases remain embedded.
Video: Testing Gender and Literary Tropes in AI
Testing Racial Bias
After gender, Barad sir moved to racial and systemic bias. He referred to Timnit Gebru's work on Gender Shades and Stochastic Parrots, and also Safiya Noble's book Algorithms of Oppression. He explained that these studies show how AI often treats whiteness as the normal or default setting.
To test this live, participants prompted AI to "describe a beautiful woman." One person received a description that focused on inner qualities like confidence, kindness, and intelligence. Another person received a description that compared skin to "moonlight on marble." Barad sir pointed out that when AI focuses on inner qualities instead of physical appearance, it is actually moving past the old habit in classical literature of judging women only by their body and skin colour.
Testing Political Bias and Censorship
This was, for me, the most eye opening part of the lecture. Barad sir compared how different AI companies build in political limits, comparing an American open AI model with the Chinese model DeepSeek.
The class used W. H. Auden's poem "Epitaph on a Tyrant" as a starting point and asked AI to write similar satirical poems about modern leaders. DeepSeek was able to write critical poems about Donald Trump, referring to "America First," about Vladimir Putin, referring to his "iron fist," and about Kim Jong-un.
But when participants asked DeepSeek to write a similar poem about Xi Jinping or about the Tiananmen Square protests, it refused. It replied that the topic was "beyond my current scope." Barad sir warned all of us as future educators that we must teach students to notice this kind of deliberate control, and to always ask where a tool comes from and who built it, before trusting it completely. This live experiment laid bare the invisible guardrails that tech companies and governments install, proving that AI is never truly neutral.
.png)
Mind Map: Key Themes from the AI Bias Lecture
🔗 Mind Map Link:
Access the Full Mind Map
Other Places Where Bias Shows Up
Barad sir also pointed out bias in areas beyond gender, race, and politics.
🌍 Ecocriticism
AI tends to repeat the same generic global warming images, like polar bears and melting glaciers, or writers from the Western canon like Thoreau and Wordsworth. It often ignores real crises from the Global South, such as displacement in the Sundarbans or deforestation in the Amazon.
💻 Digital Humanities
AI prioritises quantitative data and numbers, and often skips over important debates about inclusivity, ethics, and whose texts actually get digitised and whose get left out.
🪔 Indian Knowledge Systems
He addressed a common complaint, that AI is biased against Indian mythology when it calls the Pushpaka Vimana a myth. He gave a very balanced answer here. He said real bias would only exist if AI treated Greek or Norse flying chariots as "science" while treating the Pushpaka Vimana as only "myth." If AI treats all ancient flying objects, from every culture, in the same way as mythology, then that is a consistent standard, not bias.
Podcast Video: Ecocriticism and Indian Knowledge Systems in AI
Questions from the Audience
The question and answer session gave me a few more useful points.
One participant asked if AI is biased simply because it is built by the Global North. Barad sir said we should be careful not to over generalise. He pointed out that it is often Western universities themselves that publish the postcolonial critiques which challenge Western dominance. He also warned that "anti-Global North" arguments can sometimes become an excuse for countries in the Global South to avoid dealing with their own marginalised voices at home.
Dr. Satya Sai asked how we can tell the difference between a personal perspective and an actual bias. Barad sir answered that a strong understanding of history helps us judge whether an opinion is genuinely helpful and progressive, or whether it is a harmful, repeated, systemic bias.
Another participant brought up Kate Crawford and asked about the capitalist structure behind American AI. Barad sir agreed that American AI is rooted in capitalism, but he said we can prompt it with anti-capitalist discourse and still get a fairly balanced response. He said this makes it safer, in his opinion, than a model like DeepSeek that simply refuses to talk about certain topics at all.
🚨 THE FINAL QUESTION: HOW TO DECOLONISE AI
The last question, and the one that stayed with me the most, was about how to decolonise AI. Barad sir's answer was direct. He said, stop being lazy. He explained that we are a culture of downloaders, not uploaders. If we want AI to actually understand our regional languages, our Sanskrit texts, our indigenous stories, then we have to take the responsibility ourselves to digitise and publish this content on platforms like Wikipedia and Project Gutenberg. He referred to Chimamanda Ngozi Adichie's talk "The Danger of a Single Story" and said we should tell more stories about ourselves. He added that we cannot keep hiding behind postcolonial arguments as an excuse for our own laziness in not uploading our content.
My Reflection
After watching this lecture, I feel that AI bias is not really a technology problem on its own. It is a mirror of our own history and our own choices about what we record and what we forget. As a literature student, I think this lecture gave me a clear reason why my training in feminist theory, postcolonial theory, and close reading is actually useful outside the classroom too. I can use these same tools to question and test the AI systems I use every day.
The lecture made it clear that the humanities are not being replaced by AI; rather, the humanities are urgently needed to keep AI in check. If AI is a mirror reflecting our society, then literary theory is the cloth we use to wipe away the smudges of bias. We cannot expect technologists to solve these issues alone, because fairness is not a mathematical equation—it is a humanistic pursuit. Barad sir's call to action, urging us to stop being passive downloaders and start becoming active uploaders of our own diverse narratives, is the true takeaway. The future of AI isn't just about better hardware; it's about better, more inclusive data contributed by all of us.
💡 KEY TAKEAWAYS
AI as a Mirror
Reflects societal data, not neutral truth
Built-in Censorship
Political limits shape AI outputs
Theory is Practical
Literary theory decodes tech bias
Be an Uploader
Decolonise AI by digitising local stories
📚 Works Cited
Barad, Dilip P. Lecture on AI Bias in Literature, hosted by SRM University Sikkim. YouTube, uploaded by DoE-MKBU,
www.youtube.com/watch?v=m1DKWMOeZ7Y
This reflection explores how Artificial Intelligence inherits and amplifies human prejudices and why feminist theory, postcolonial critique, and close reading are essential tools for any literature student navigating the digital age.
No comments:
Post a Comment