Why I Wrote an AI Ethics Book From India, Not Stanford
Why I Wrote an AI Ethics Book From India, Not Stanford
Go to your nearest bookstore. Walk to the technology section. Pick up every book on AI ethics you can find. Open each one and look at where the author is based, whose case studies they use, whose regulatory frameworks they analyse, whose problems they centre.
I will save you the trip. Nearly every major AI ethics book published in the last five years was written from a university in the United States or the United Kingdom. The case studies are American and European. The regulatory analysis centres on the EU AI Act and US executive orders. The fairness metrics were developed by researchers at MIT, Stanford, and Berkeley. The moral philosophy is Western liberal tradition, occasionally citing Rawls, occasionally citing Kant, rarely citing anything else.
This is not a complaint about those books. Some of them are excellent. I have read them, learned from them, and cited them in my own work. The researchers at these institutions are doing important, rigorous work, and the world is better for it.
But there is a problem. A large one. And if you are reading this from Bengaluru or Pune or Hyderabad or Delhi, you have probably already felt it.
Those books are not about you. Those frameworks were not built for your reality. And the AI systems that are increasingly making decisions about your credit, your job application, your healthcare, and your government benefits are operating in a context that Western AI ethics has barely begun to grapple with.
Let me give you the numbers.
India has over eight hundred million internet users. That is more than the United States and Europe combined. By next year, it will be the world’s most connected country by sheer scale. India jumped from seventh to third place in Stanford’s AI Vibrancy Index in 2025. Investment commitments in Indian AI crossed two hundred billion US dollars at the AI Impact Summit in February 2026. India’s GenAI startup ecosystem has grown nearly four fold, crossing 890 startups by mid 2025 according to NASSCOM, with thousands more building AI powered products across sectors.
This is not a country on the margins of the AI conversation. This is a country at the centre of it.
And yet, when you read the global AI ethics literature, India appears as a footnote. An occasional case study. A “developing country context” paragraph near the end of a chapter. A reminder that “these issues are especially acute in the Global South,” as if acknowledging the problem counts as addressing it.
Here is what the Western AI ethics literature does not know about India. Or rather, here is what it knows in the abstract but does not understand in the concrete.
India has twenty two officially recognised languages. Hundreds more are spoken across the country. When an AI language model is trained overwhelmingly on English data, with Hindi as a distant second and Tamil, Telugu, Marathi, Bengali, and the rest as afterthoughts, the bias is not just linguistic. It is civilisational. The model learns to think in English. It learns to reason from English language sources. It absorbs the cultural assumptions, the reference points, the value systems embedded in English language data. And then it is deployed to serve eight hundred million people, most of whom do not think in English.
India has caste. This is the thing that Western AI ethics frameworks are not equipped to handle. Caste does not map neatly onto race, the category that American fairness research centres. It does not map onto class, although it intersects with class. It is its own axis of inequality, deeply embedded in social structures, economic patterns, and yes, in data.
When a lending algorithm uses PIN codes as input features, it is using caste as a variable without ever naming it. When a hiring tool is trained on resumes from companies where upper caste representation has historically been disproportionately high, it learns what a “good” candidate looks like, and that picture carries caste in its bones. When a language model is asked about different communities, research has shown it reproduces caste stereotypes with alarming consistency.
The fairness metrics developed at Berkeley, demographic parity, equalized odds, calibration, do not include caste as a category. The benchmark datasets used to test AI systems for bias do not include caste as a dimension. The BBQ benchmark, one of the most widely used tools for measuring bias in language models, tests for nine dimensions of bias. Caste is not one of them.
This is not an oversight. It is a structural blind spot. And you cannot fix a problem that your measurement tools are not designed to detect.
Then there is the digital divide. Roughly 630 million Indians remain offline, 500 million of them in rural areas. And even among those who are connected, the data these systems are trained on reflects a very specific slice of online India: urban, English comfortable, smartphone owning, app using India. The farmer in Vidarbha, the daily wage worker in Bihar, the elderly woman in rural Rajasthan who has never used a smartphone — they are not in the training data. But the AI systems trained on that data are increasingly making decisions about their lives. About their ration cards. About their loan eligibility. About their children’s school admissions.
When Jharkhand’s biometric system denied rations to families because their fingerprints were too worn from manual labour, that was not a technology failure. It was a failure of imagination. A failure to ask: who is this system designed for, and who will it exclude?
India’s own Aadhaar ecosystem, the world’s largest biometric identity system, is simultaneously one of the most ambitious digital public infrastructure projects in history and one of the most consequential experiments in algorithmic exclusion. How you feel about it depends on where you stand, which is exactly the point.
The regulatory landscape is different too. Not just different in the way that every country’s laws are slightly different. Fundamentally different.
India chose a deliberately light touch approach to AI governance. The India AI Governance Guidelines, released in November 2025, offer seven principles, which they call “sutras,” and not a single binding obligation. There has not been one AI specific enforcement action taken by any Indian regulatory body as of this writing. Zero. In a country with eight hundred million internet users and the third most vibrant AI ecosystem on the planet.
The European Union, by contrast, built a four tiered risk framework with penalties up to thirty five million euros or seven percent of global turnover. The US has a patchwork of state laws and sector specific rules. China has binding regulations on deepfakes, recommendation algorithms, and generative content that predate anything the West has done.
India’s approach is none of these. It is its own thing. And you cannot understand it through the lens of Western regulatory theory. You have to understand it through the lens of Indian governance: the trust in industry self regulation, the preference for innovation over precaution, the experience of building digital public infrastructure at a scale no other country has attempted. Aadhaar. UPI. ONDC. India has built things at billion person scale that other countries study from a distance. That experience shapes how India thinks about AI governance, for better and for worse.
A book written from Stanford cannot capture this. Not because Stanford lacks intelligence, but because it lacks proximity. The choices India is making about AI governance right now will affect more people than the choices any other country makes. Those choices deserve to be examined on their own terms, by someone who builds in this ecosystem, who understands the constraints, who feels the consequences.
This is why I wrote Beyond Bias from India.
Not as a claim that the Indian perspective is the only one that matters. It is not. But as a corrective to a body of literature that has, almost entirely, ignored it.
The book covers the global landscape. The EU AI Act. US regulation. China’s approach. Singapore’s agentic AI framework. But it starts from India. The case studies are Indian. The regulatory analysis centres on Indian laws and guidelines. The fairness discussion grapples with caste, not as an exotic aside but as a central challenge that any serious AI ethics framework must confront. The privacy chapter examines the Digital Personal Data Protection Act in detail, not as a comparison point to GDPR but as a law that eight hundred million people will live under.
The framework at the heart of the book, the Four Way Test, is not Indian in the sense that it applies only to India. It is universal. But it was built by someone who understands that “universal” does not mean “designed in the West and applied everywhere else.” It means designed to work in Delhi and Detroit, in Pune and Paris, in a tier three town in Uttar Pradesh and a startup in San Francisco.
Most AI ethics books are written in the West, for the West. The rest of the world is invited to read along and adapt.
This one starts from our reality.
I’m have written a book about exactly this how AI and automated systems make decisions about your life, where accountability disappears, and what we can do about it. If you want to know morea about this book or order a copy, you can do it here: https://akshaywalimbe.com/beyond-bias/