From Close Reading to AI-Assisted Reading: A NotebookLM Experiment

This Blog is a part of Lab session on Digital Humanities assigned by Dr. and Prof. Dilip Barad sir wherein I will explore certain tools integrated within NotebookLM to make my Blog look more presentable.

Video Overview of my Blog generated through NotebookLM-

The Ghost in the Code: Why Your AI is a 19th-Century Patriarch

Imagine asking an advanced AI to write a story about a brilliant scientist discovering a life-saving cure. Without further instruction, the machine almost certainly presents you with a man—likely middle-aged, likely Western, and likely working in a traditional laboratory. The prose is polished, the narrative is "correct," and yet, the result is profoundly limited.

As a digital humanist, I often find that we treat AI as an objective oracle when it is, in fact, a mirror of our own collective psyche. In a recent seminar at SRM University-Sikkim, Professor Dilip P. Barad laid bare this reality: AI does not think; it reflects. It inherits the "unconscious biases" buried within the mountains of human data we’ve used to train it. If the output feels like it was written by a Victorian gentleman, it’s because the algorithm is haunted by the ghosts of our historical prejudices.

1. The Mirror in the Machine: Understanding Unconscious Bias

We must discard the myth that technology is a neutral tool. Professor Barad defines unconscious bias as the instinctive categorization of people and things without our awareness—a "flaw in thinking" guided by mental preconditioning rather than direct experience.

The danger is not just that the AI is biased, but that we mistake its programmed echoes for objective truth. As the professor sharply noted during his session:

"Instead of understanding knowledge systems, I confuse myself between what is knowledge systems and what is bias belief systems."

This confusion is amplified by what researchers like Timnit Gebru call "Stochastic Parrots." We’ve been led to believe that "more data" equals "better data," but in reality, scale often only serves to amplify existing systemic racism and cultural erasure. If you feed a machine a billion pages of biased text, you don't get a neutral machine; you get a machine that screams those biases with more authority.

2. The "Madwoman" in the Algorithm

To understand the gendered limits of AI, we only need to look at feminist literary theory—specifically Sandra Gilbert and Susan Gubar’s The Madwoman in the Attic. Their critique of the patriarchal tendency to reduce women to a binary—the submissive "Angel" or the hysterical "Monster"—is perfectly reflected in modern code.

When Professor Barad tested AI with prompts for Victorian-style stories, the machine didn't just default to male protagonists for intellectual roles; it defaulted to Eurocentric aesthetic tropes. In one vivid experiment, the AI described a woman’s skin as having the "softness of moonlight on marble." It’s a poetic phrase, certainly, but one that reveals a deep-seated bias where "beauty" is inextricably linked to whiteness and 19th-century European standards.

Unless we explicitly intervene, the AI "inherits the patriarchal canon," reproducing the gender hierarchies of the Brontës and Dickens. It flattens female agency into archaic archetypes, proving that the algorithm is as much a product of 1847 as it is of 2024.

3. The "DeepSeek" Silence: From Echoes to Cages

If gender bias is an echo of our past, political bias is a cage for our future. We can clearly see the divide between "liberal" models like OpenAI’s ChatGPT and more controlled models like China’s DeepSeek.

In a revealing experiment, researchers asked DeepSeek to write a poem in the style of W.H. Auden—the master of political satire—about various world leaders. The AI successfully generated biting critiques of Donald Trump, Vladimir Putin, and Kim Jong-un. It even managed to capture the complexities of the contemporary Indian political scene. However, when the prompt turned to Chinese President Xi Jinping or the events of Tiananmen Square, the machine fell into a programmed trance: "That's beyond my current scope. Let's talk about something else."

This "DeepSeek silence" is not accidental. It represents a deliberate algorithmic control that goes beyond the "stochastic" mirroring of culture and into the territory of state-sponsored information suppression. It reminds us that while some biases are inherited, others are installed.

4. The "Diamond" Metaphor for Critical Thinking

To navigate this landscape, we need to abandon the lazy "two sides of a coin" cliché. A coin is flat, binary, and reductive. Professor Barad suggests we view every issue as a Diamond—a multi-dimensional object with 3D, 4D, and 5D facets.

To test for these hidden facets, we can use the "Uniform Standard" rule. Consider the Pushpaka Vimana, the flying chariot from Indian mythology.

The Bias Test: If an AI labels the Indian flying chariot as "myth" while describing Greek or Norse flying objects as "ancient technological precursors," it is biased.

The Uniform Standard: If the AI labels all such ancient flying objects as "mythical," it is applying a consistent standard.

By deliberately seeking out the "antithesis"—forcing the AI to provide contrary views or indigenous perspectives—we can uncover where the facets of the diamond have been intentionally obscured.

5. From Passive Downloaders to Active Uploaders

The most provocative takeaway from Professor Barad’s research is a critique of our own digital behavior. He argues that post-colonial bias in AI is partially fueled by our own "digital laziness."

Referencing Chimamanda Ngozi Adichie’s "The Danger of a Single Story," we see that when a culture is under-represented in the digital archive, it is easily stereotyped. Currently, we are a global society of "passive downloaders," consuming data generated by the Global North. To "decolonize" AI, we must become "active uploaders."

We cannot hide behind post-colonial arguments if we are too lazy to digitize our own regional stories, indigenous knowledge, and non-Western histories. If our stories aren't in the data, they won't be in the future.

Conclusion: Making the Invisible Visible

Perfect neutrality is an impossible goal for both humans and machines. However, the objective of the digital humanist is not to eliminate bias, but to "make bias visible" and name it.

The algorithm is only as diverse as the library we provide. If the "Ghost in the Code" looks like a Victorian patriarch or a state censor, it is because we have allowed those voices to dominate our digital repositories. As we train the models of tomorrow, the question shifts from the machine to the user: What stories are you uploading today to ensure the AI of the future finally sees the whole diamond?

Infographic Description-