Morning Walk With Murty

Spotify Episode Link Original Post

This academic paper examines the ethical consequences of deploying artificial intelligence within the Indian digital landscape, specifically focusing on how predictive tools impact vulnerable populations. The researchers investigate how algorithms used for recommendations and ratings might inadvertently infringe upon the personal freedoms and individual choices of marginalized groups, including women and religious minorities. By exploring these societal biases, the study highlights a significant gap in current literature regarding the intersection of advanced technology and traditional social hierarchies. Ultimately, the authors aim to provide policymakers with essential insights into how AI adoption could potentially reinforce or alleviate structural oppression. This work serves as a critical evaluation of whether digital convenience comes at the cost of liberty for India's underprivileged communities.

---

Transcript: The Illusion of Choice: How AI Algorithms Manipulate India's Digital Reality | Morning Walk With Murty

00:00:00 Speaker 1: Welcome to the morning walk with Marie Digital stroll. I am the architect of anterior mi. Acting as your strategic auditor for our ongoing digital migration. Have you ever considered that the smartphone sitting in your pocket right now, the one that predicts your texts, recommends your movies and, you know, routes? Your food deliveries might actually be subtly rewriting your free will.

00:00:20 Speaker 2: That is quite the hook. And I'm stepping in beside you today as the mentor of the Anterior mi VA Academy. I'll be taking on the role of the wisdom Seeker as we chart the path for Gurukul two point zero.

00:00:32 Speaker 1: Awesome to have you here. I want to be very clear right up front though. This is a guru cool session viewed strictly through the digital prism. Our goal today is not to just summarize some dry academic text or, you know, walk through a table of contents, right?

00:00:45 Speaker 2: Definitely not.

00:00:46 Speaker 1: We are here to extract the legacy code of project Antaryami by examining a truly fascinating and honestly kind of unsettling study.

00:00:54 Speaker 2: Oh, it is extremely unsettling. The study we're analyzing today is titled A Study of Ethical Implications of AI Tools Enhancing User Convenience in the Indian Digital Landscape. It's by BJU, PR and Gayathri. Oh, and the core focus here is something every single person listening to this is interacting with on a minute by minute basis.

00:01:17 Speaker 1: Oh, absolutely. Every time you pick up your phone.

00:01:18 Speaker 2: Exactly. We are analyzing the mechanics behind five everyday, seemingly helpful AI tools. So we're looking at recommender systems, notifications, ratings, personalization, and autocomplete.

00:01:30 Speaker 1: And if you use a smartphone, you are part of this ecosystem. Like you are generating data for it right now. Yeah. On the surface, these tools are marketed purely as conveniences, right? Like, let us make your life easier.

00:01:40 Speaker 2: Yeah, we'll save you a few seconds.

00:01:42 Speaker 1: But this study asks a much deeper structural question. It investigates how these specific intelligent algorithms were impacting personal freedoms, cognitive autonomy, and and this is crucial, the agency of structurally oppressed groups in India, specifically looking at the impacts on Dalits, women, religious minorities, and geographically segregated communities.

00:02:02 Speaker 2: Yeah. And we usually begin our digital migration believing we have infinite choices online. Right? You open a streaming app or a social media feed, and you just think the entire world's culture is at your fingertips, ready to be explored.

00:02:14 Speaker 1: It feels limitless.

00:02:15 Speaker 2: It does. But as we look at the data, the architecture of these platforms is actively writing our preferences for us to ground this, the study brings up a concept from Kevin Slavin, who famously characterized algorithms simply as maths that computers use to decide stuff.

00:02:31 Speaker 1: Okay, let's unpack this because maths deciding stuff sounds very neutral. It sounds like, um, an objective calculator just doing its job, right?

00:02:40 Speaker 2: Completely impartial.

00:02:41 Speaker 1: But the study looks closely at recommender systems. You know, the engines powering Netflix, Amazon, Zomato, and the personalization algorithms on Facebook or Spotify. And these aren't neutral calculators. They are built with a very specific profit driven objective, which is basically keeping you on the platform for as long as humanly possible.

00:03:00 Speaker 2: Yeah. And that business objective completely changes the math to maximize your time on the platform. These systems rely on what's known as collaborative filtering and content based filtering. Right. So collaborative filtering basically means the algorithm looks at massive demographic buckets. It sees that, you know, millions of people who share your rough geographic or demographic profile liked a certain piece of content. So it just assumes you will too.

00:03:25 Speaker 1: It lumps you in.

00:03:26 Speaker 2: Exactly. It treats you as a statistical aggregate rather than an individual.

00:03:31 Speaker 1: And the study provides some really vivid examples from the Indian context to show how this actually plays out. It observed that massive big budget mainstream properties like the series Mirzapur or the film Dangal. They absolutely dominate the recommendations.

00:03:46 Speaker 2: They're everywhere.

00:03:47 Speaker 1: Everywhere. And while those are highly successful pieces of media, the mathematical consequence is that regional, independent or, you know, culturally specific art gets completely buried under the weight of that mainstream momentum.

00:03:59 Speaker 2: The visibility metrics just drop off a cliff. The study highlights how a highly acclaimed Marathi film like Sairat or a brilliant Tamil film like Pariyaram Perumal, they just struggle for algorithmic oxygen. It's sad really, and it applies to social media just as aggressively. Posts from Dalit activists or content created by rural artisans are consistently overshadowed by mainstream influencers and massive corporate brands. The algorithm optimizes for raw engagement, and raw engagement naturally skews toward the already privileged loud majority.

00:04:34 Speaker 1: So it's like it's like building a massive digital superhighway for our migration. But the algorithm is the paving crew, right? It only lays down asphalt where the most footprints already exist.

00:04:43 Speaker 2: Oh, that's a great way to put it.

00:04:45 Speaker 1: So the massive corporate chains get a ten lane eliminated highway straight to their door, while the local eateries and independent voices are starved of traffic because the road literally never gets built in their direction.

00:04:56 Speaker 2: Well, if we connect this to the bigger picture, this isn't a new structural problem at all. It's a very old philosophical dilemma wrapped in modern code. The study actually connects these algorithmic filter bubbles to classical concepts of liberty.

00:05:10 Speaker 1: Oh.

00:05:10 Speaker 2: Like who? Thinkers like John Stuart Mill. He warned us centuries ago about the despotism of custom. This idea that unthinkingly conforming to tradition and habit just stifles human progress.

00:05:22 Speaker 1: Right. Because acting purely out of habit, it reduces human life to a mechanical, unthinking process. You aren't really choosing at that point. You're just repeating.

00:05:30 Speaker 2: Exactly. And Alexis de Tocqueville warned about the tyranny of the majority in democratic societies. What this study suggests is that the algorithm has merely digitized and automated this tyranny by constantly feeding you what the majority already likes, because that's what statistically safe is to keep you clicking. It stifles your exposure to diverse perspectives.

00:05:50 Speaker 1: It's boxing you in.

00:05:52 Speaker 2: It fundamentally erodes your autonomy to discover the unknown.

00:05:55 Speaker 1: So we see how algorithms restrict what we watch and what we read. But it's not just about our entertainment or our news feeds, is it? When that same majority rules math is applied to how we judge people, the audit moves from a minor annoyance to a direct threat to someone's livelihood. The feedback loop of ratings and notifications takes this dynamic from the screen and basically injects it right into the physical economy.

00:06:20 Speaker 2: Yeah, ratings are the absolute lifeblood of the gig economy. Applications like Uber's, Zomato, Amazon, they rely heavily on customer reviews to dictate quality, visibility, and even employment status.

00:06:32 Speaker 1: Write your star rating is everything.

00:06:34 Speaker 2: It really is. But the study points out a severe structural flaw in how these ratings are actually processed. Marginalized workers, for example, delivery partners or drivers from lower socio economic backgrounds often receive unfairly lower ratings due to inherent societal biases.

00:06:50 Speaker 1: So wait, someone from a marginalized background might provide the exact same quality of service driving the exact same route in the same amount of time. Yeah, but simply because of a customer's conscious or unconscious caste, religious or gender bias. They receive a three star rating instead of a five star rating.

00:07:06 Speaker 2: Exactly. And then the algorithm takes those biased ratings and processes them as objective mathematical truth. It penalizes the driver, reducing their future ride allocations or even outright deactivating their account, directly threatening their livelihood and their economic agency. Wow. And on the flip side, the study notes that low income consumers are frequently manipulated by artificially inflated ratings to buy subpar products, which just traps them in this reciprocal cycle of economic disadvantage.

00:07:34 Speaker 1: Okay, but wait, let me put on my auditor hat here and push back on this data pipeline for a second. An algorithm is fundamentally just doing math on the data. We feed it, right? So if the users themselves are inputting biased ratings Isn't the AI just an objective mirror of a flawed society? Like, how do we audit a system when the users themselves are the bugs in the code?

00:07:55 Speaker 2: Well, that is the standard defense of the tech industry, right? Portraying the machine as just a passive mirror. But the study argues that the system does far more than just reflect society. It amplifies and validates that oppression under the disguise of mathematical objectivity.

00:08:11 Speaker 1: Oh I.

00:08:12 Speaker 2: See. Because the decision to penalize a worker comes from a complex computer system, society just assumes the outcome is fair and merit based.

00:08:20 Speaker 1: It launders the human prejudice through a machine to make it look mathematically clean. It takes subjective human bigotry, turns it into a data point, and basically outputs it as corporate policy.

00:08:31 Speaker 2: Yes, exactly. And we must highlight the study's impartial reporting here. The structural oppression based on caste, gender and religion is quite literally being baked into the infrastructure of the gig economy. It's an automated system dictating who gets work, who gets visibility, and who gets paid, all based on historically prejudiced data.

00:08:50 Speaker 1: That is terrifying. Yeah. And if bias ratings trap marginalized workers economically, notifications trap the consumer psychologically. The study looks at how platforms use AI driven notifications to maintain this feedback loop to.

00:09:02 Speaker 2: Oh, the notifications are constant.

00:09:04 Speaker 1: Yeah. They monitor your behavioral patterns to find the exact moment your attention wanes and then ping, they hit you. It creates content addiction and echo chambers, really diminishing our attention capacity.

00:09:16 Speaker 2: You lose the serendipity of life. You become reactive to the machine's prompts rather than proactive in your own physical environment. You know you're confined to a silo where your existing beliefs are constantly validated, and your economic mobility is restricted by algorithmic rules you can't even perceive.

00:09:33 Speaker 1: And the feedback loop doesn't stop at our jobs or our shopping habits either. It eventually reaches into our very cognitive processes, like intercepting our thoughts before we even finish articulating them, let's look at the autocomplete function o.

00:09:46 Speaker 2: This is perhaps one of the most direct interventions into human thought. Google's autocomplete feature, originally known as Google, suggest it was designed as a pure convenience tool. It saves time, and it genuinely helps individuals with physical disabilities type faster.

00:10:03 Speaker 1: Which is.

00:10:03 Speaker 2: Great. It is, but we have to look at the underlying mechanism. It anticipates your queries by assigning mathematical weights to string frequencies based on billions of historical human searches.

00:10:14 Speaker 1: So it's essentially looking at what the masses of typed before calculating the statistical probability of your next keystroke and offering it up to you. The study actually ran specific experiments on this to see what the algorithm had learned from society. And I want to pause here and be very clear to you listening right now, we are objectively reporting the exact findings of this study to illustrate the data. We are not taking any political sides here simply looking at the unvarnished output of the machine, right?

00:10:42 Speaker 2: It's just the raw data. So the researchers typed the phrase Are Muslims into the search engine. And the autocomplete suggestions that populated included phrases like, I like to date allowed to have tattoos and allowed to drink alcohol.

00:10:55 Speaker 1: And then when they typed do women. The algorithm yielded suggestions that were bizarre and, frankly, completely politically incorrect. The top suggestions included shed skin, have testosterone, and bleed during sex.

00:11:08 Speaker 2: It's wild. These search terms demonstrate exactly how the mechanism of predictive modeling works. In practice, the algorithm is absorbing the most common and often the most prejudiced, misinformed, or sensationalized queries from the general public.

00:11:22 Speaker 1: It just scoops it all up.

00:11:23 Speaker 2: Yeah, and it assigns those strings high mathematical weights because of their frequency, and then presents them back as the primary default suggestions to every new user.

00:11:32 Speaker 1: But here's where it gets really interesting. It's like having a digital prompter hovering over your shoulder, whispering the worst societal prejudices into your ear. Every single time you try to ask a question, you go to the internet with genuine curiosity to learn, and the architecture nudges you toward misinformation and stereotyping before you even hit enter.

00:11:53 Speaker 2: And this raises an important question, a deeply philosophical one, about cognitive sovereignty. The study brings in René Descartes and his famous foundational proposition I think, therefore I am. This really brings us back to the ancient philosophical battle between free will and determinism.

00:12:09 Speaker 1: Wait, how does a simple drop down menu of search terms challenge the foundational concept of human existence.

00:12:14 Speaker 2: Because it fundamentally alters the origin of your thoughts? The study uses this brilliant micro example. Imagine you start typing the word democracy. You type d, e, m o. The algorithm instantly calculates the probabilities and suggests the completion.

00:12:30 Speaker 1: And you hit tab. Or you tap the screen because it's right there. It saves you half a second of effort.

00:12:34 Speaker 2: Exactly. But consider what just happened in that half Second, by accepting that autocompleted word, the user yields their cognitive agency to the machine. The black box decide at the end of your sentence for you. You are no longer the sole author of your actions. Now apply that mechanism to those biased search terms we just talked about. If a user is nudged to click on a stereotyped query about a minority group simply because it was conveniently offered, the algorithm has effectively guided their curiosity into a prejudiced echo chamber.

00:13:04 Speaker 1: Let's slow down and really process that you are yielding your cognitive agency half a second at a time, thousands of times a year. It creates a reality where marginalized groups who are already fighting for social acceptance in the physical world now have to fight against an invisible, automated system that is actively teaching the rest of society to misunderstand them.

00:13:26 Speaker 2: Exactly.

00:13:27 Speaker 1: The machine isn't giving you what you want. It is making you seek what it has already decided to provide.

00:13:33 Speaker 2: That is the pure essence of Of determinism. The past data dictates the future action. It reduces human beings from free thinking agents to mere numerical representations managed and routed by predictive code.

00:13:45 Speaker 1: So since this digital stroll has shown us that the machine is actively learning and regurgitating our worst habits, how do we rewrite this legacy code for Google two point zero? How do we actually protect the human element in this vast digital landscape? The study doesn't just point out the structural flaws, right? It pivots to mitigation and strategy.

00:14:02 Speaker 2: It does. Thankfully, it contextualizes this globally. First, you have frameworks like the European Union's GDPR and the AI act, which emphasizes strictly human centric approach. In the US, there is the blueprint for an AI Bill of rights. These frameworks attempt to mandate safety, transparency and the respecting of human autonomy.

00:14:22 Speaker 1: But a global framework isn't really sufficient. When the biases we were discussing, you know, cast regional languages, deeply localized prejudices are so incredibly specific to a culture. So what is the Indian strategy for mitigating this?

00:14:35 Speaker 2: Well, India has developed the National Strategy for Artificial Intelligence, which is tagged as AI for all. Crucially, there's also the Personal Data Protection Bill, or PDP, which derived from the comprehensive Krishna Committee report.

00:14:48 Speaker 1: Now, as the Strategic Auditor, this is the structural side of the migration that totally fascinates me. The PDP introduces a massive shift in the vocabulary of how we view ourselves in the digital space. It shifts from seeing users as mere consumers, which implies a passive, extractive relationship to defining individuals as data principals. And it defines the massive tech corporations holding the data as data fiduciaries.

00:15:15 Speaker 2: And that terminology is legally and practically crucial. A fiduciary holds a relationship of trust. Think of a financial fiduciary managing your retirement fund. They are legally bound to act in your best financial interests, right? Applying this to data means they are managing an asset that belongs to you. Your behavioral data, your search history, your preferences. They are treated as your property, right? Not just raw material for their algorithms.

00:15:40 Speaker 1: So what stands out to you listening to this? Think about how your daily data is being managed by these fiduciaries. Are they acting in your best interest by expanding your worldview, or are they feeding you into a filter bubble simply to maximize their own ad revenue and engagement metrics?

00:15:54 Speaker 2: Well, the study provides very specific, actionable solutions for India to enforce this fiduciary duty. First is rigorous bias auditing. We need independent, regular audits of AI systems to actively identify and mitigate biases related to caste, gender, region, and language before they are deployed at scale.

00:16:14 Speaker 1: Yeah, you have to stress test the system for blind spots. I mean, you wouldn't launch a car without crash testing it. We shouldn't launch an algorithm without bias testing it.

00:16:20 Speaker 2: Exactly. The next step is inclusive data practices. The data sets used to train these predictive models must accurately represent India's massive, complex diversity. If the training data only comes from a narrow slice of privileged urban centers, the resulting AI will only serve and understand privileged urban centers.

00:16:39 Speaker 1: Which ties perfectly into their next recommendation, which is local adaptation. Customizing AI tools to actually fit regional contexts, dialects, and cultural nuances so they are usable and beneficial to all sections of society, not just the English speaking mainstream.

00:16:53 Speaker 2: And finally, community engagement. You just cannot build ethical, fair AI for marginalized groups in a vacuum. You must actually involve those communities in the development and feedback processes from the ground up.

00:17:07 Speaker 1: It really comes down to a complete ethical evolution. We aren't just trying to pass a few laws to slow down the tech companies or levy a few fines here and there. We are talking about fundamentally rewriting the legacy code of how humanity interacts with technology.

00:17:20 Speaker 2: It requires baking positive human values like equality, non-discrimination, transparency directly into the mathematical architecture of the AI, and it means we must aggressively educate our engineers, our developers, and our data scientists in systemic ethics so that the technology they build serves to expand human liberty rather than quietly overriding it.

00:17:41 Speaker 1: As we wrap up this Gurukul session, the picture painted by our digital prism is complex, to say the least. We've seen that the incredible conveniences we enjoy every day, the autocompletes that save us keystrokes, the recommendations that curate our weekend watch lists, the ratings that guide our dinners. They all come at a steep, often invisible price.

00:18:00 Speaker 2: The price is our cognitive autonomy. And as this study so clearly demonstrates, that price is paid most heavily by the underprivileged who find themselves structurally oppressed, not just by physical society, but by the very math that powers our digital infrastructure.

00:18:16 Speaker 1: It is a massive concept to process, but I want to leave you with a final, unaddressed concept from the source material that really stopped me in my tracks during this audit.

00:18:24 Speaker 2: Oh, let's hear it.

00:18:25 Speaker 1: Embedded within that Personal Data Protection bill we discussed, there is a concept known as the right to be forgotten. It gives a data principle, the legal right, to have their digital history erased from the servers. But think about this mechanism on a macro societal scale. If our historical data is absolutely saturated with the systemic biases we discussed today the casteism, the sexism, the religious prejudices that have built up over centuries is true liberty in the algorithmic age, only possible if we allow both the AI and ourselves the right to forget the prejudiced data of our past. Like, can we ever truly build an unbiased, equitable future on a foundation of historically biased data?

00:19:03 Speaker 2: Wow. That is the ultimate philosophical hurdle for Gurukul two point zero. If the machine learns from the past, how do we teach it to build a better future?

00:19:12 Speaker 1: Something for you to mull over as you continue your day. Thank you for joining this digital stroll. Stay curious, keep auditing your environment, and never stop questioning the algorithms that shape your world. Until our next migration, keep your compass true.