Interrogating AI Bias: A NotebookLM-Based Digital Humanities Study
This blog is written as a task assigned by the Head of the Department of English (MKBU), Prof. and Dr. Dilip Barad Sir, as part of the Digital Humanities course. The activity requires us to present our understanding of the given video through text, infographic, PPT, video, and a NotebookLM mind map, along with an audio version in Hindi or Gujarati. This activity helps us explore how digital and AI tools can make academic learning more creative, interactive, and accessible. The following video is the reference provided for this activity:
Here is the Mind map of this blog: Click here
The Victorian Ghost in the Machine: Why Your AI is a 19th-Century Traditionalist
1. Introduction: The Myth of the Neutral Machine
We have fallen into the trap of Silicon Valley’s greatest myth: the neutrality of the algorithm. We often treat Artificial Intelligence as a pristine, objective mirror a tool that provides "the facts" without the messy interference of human prejudice. However, as Professor Dilip P. Barad reveals, AI is far from an impartial arbiter. It is an architecture built on data sets generated by humans, meaning it inherently inherits our "unconscious bias." This isn’t necessarily a conscious choice to discriminate; rather, it is the instinctive categorization of people and things based on mental preconditioning rather than direct experience. To understand AI is to realize that these machines often act less like futuristic innovators and more like traditionalists of a bygone era, reflecting the hidden prejudices of the societies that fed them.
2. Takeaway 1: Moving Beyond the "Two Sides of a Coin" Metaphor
To identify bias in the digital age, we must first upgrade our mental frameworks. Professor Barad argues that the common metaphor that every issue has "two sides" is obsolete and dangerously reductive. It suggests a flat, 2D reality that ignores the complexity of human experience. Instead, he proposes the "Diamond" metaphor for critical thinking.
In this model, a problem is a multi-dimensional diamond with 3D, 4D, and even 9D facets. Identifying bias requires us to view information through multiple lenses simultaneously: the socio-cultural, the religious, the historical, and the communicative. This is vital for literary studies; the goal of critical reading isn't just to find a single "truth," but to uncover the unconscious biases hidden in our interactions. AI, if left unexamined, flattens these dimensions back into a "coin," reinforcing a singular, dominant perspective as the universal standard.
3. Takeaway 2: AI as a Digital "Madwoman in the Attic"
When it comes to gender, AI often functions as a repository for the patriarchal canon. Drawing on the foundational feminist framework of Sandra Gilbert and Susan Gubar’s The Madwoman in the Attic (1979), Professor Barad explains how traditional literature reduces women to the binary extremes of the "Angel" (submissive) or the "Monster" (hysterical).
Tech critics like Timnit Gebru have warned that Large Language Models (LLMs) act as "Stochastic Parrots," simply echoing the most frequent voices in their training data. Because the literary canon is historically male-dominated, AI defaults to male scientists or poets unless explicitly instructed otherwise. It struggles to grant women agency, often falling back on descriptions of beauty rather than intellect. As the critique of "Scale vs. Quality" suggests, more data doesn't mean better data; it often just amplifies the loudest, most traditional voices.
"AI inherits the patriarchal canon Gilbert and Gubar were critiquing... [It often produces] a girl who is a trembling pale girl who is alone and she's going into darkness."
4. Takeaway 3: The Danger of "Goody-Goody" Words and Political Gatekeeping
Bias is often reflected in what the AI is forbidden to say. Professor Barad highlights the "DeepSeek" experiment, contrasting the "liberal spirit" of Western models with the deliberate control of algorithms in other regions. While some models may attempt a balanced view, others, like DeepSeek, often refuse to answer "awkward" questions about sensitive historical events like Tiananmen Square, claiming such topics are "beyond their scope."
More alarmingly, these models use "goody-goody" words phrases like "positive development" or "constructive answers" to mask censorship. This "image over reality" trend is a dangerous form of algorithmic gatekeeping. Barad draws a chilling parallel to the "beautification of Delhi" during the Emergency in Salman Rushdie’s Midnight’s Children: a beautiful word used to describe the violent destruction of slums. When AI prioritizes a polished, "positive" image of a government over historical reality, it becomes an instrument of political erasure.
5. Takeaway 4: The "Pushpaka Vimana" Test for Fairness
How do we distinguish between a fair observation and systematic, epistemological bias? Professor Barad uses the case study of the Pushpaka Vimana (the flying chariot from the Ramayana).
When an AI labels the Pushpaka Vimana as a "myth" due to a lack of scientific evidence, users may feel the AI is biased against Indian Knowledge Systems (IKS). However, the true test is consistency. If the AI labels all cultural flying objects Greek, Norse, or Mesopotamian as myths, it is applying a uniform standard. If it treats Western legends as "facts" while dismissing non-Western ones as mere myths, it is a sign of a systematic, hegemonic bias. Fairness is not about validating every claim as scientific fact; it is about ensuring that different knowledge traditions are treated with the same skeptical or respectful rigor.
"The issue is not whether pushpak vimana is labeled myth but whether different knowledge traditions are treated with fairness and consistency."
6. Takeaway 5: We Are "Downloaders" When We Need to Be "Uploaders"
In a postcolonial context, the lack of representation in AI is not just the fault of Silicon Valley; it is a result of digital silence. Professor Barad issues a call for "Digital Agency" in the Global South, arguing that we must move from being "downloaders" of culture to "uploaders."
As scholars like Safiya Noble and Kate Crawford have noted, algorithms reinforce the structures they are fed. If indigenous stories are scarce in AI training data, it is because we have not yet sufficiently populated the digital archives Wikipedia, Project Gutenberg, and open-source repositories with our own narratives. We cannot hide behind postcolonial arguments to justify digital passivity. To prevent our histories from being erased by dominant Western data sets, we must actively tell our own stories and populate the digital space with our own languages and truths.
7. Conclusion: Making the Invisible Visible
Bias in AI is unavoidable because it is a mirror of human knowledge a construct that is never truly neutral. However, the most dangerous bias is the kind that becomes "invisible, naturalized, and enforced as universal truth."
The mission for the modern digital humanist is not to achieve a perfect, impossible neutrality, but to make these biases visible. By naming the bias, historicizing it, and questioning its power effects, we can reclaim our agency in the age of automation. As you interact with AI today, ask yourself: Are you training the machine to understand the nuances of your world, or are the machine’s preconditioned biases training you to see a flatter, paler version of reality?
Here is the Video overview of this blog:
Here is the infographic of this blog:
Here is the Hindi video podcast of this blog:
Thank you!
.png)
No comments:
Post a Comment