> ## Content Index
> Fetch the complete content index at: https://www.slightlyacidic.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Why I disagree with Arundhati Roy’s views on AI
- URL: https://www.slightlyacidic.com/2026/10/31/disagree-arundhati-roy-views-artificial-intelligence-equator/
- Published: 2026-10-31T05:26:50.000Z
- Updated: 2026-10-31T05:26:50.000Z
- Author: VM
- Tags: power relations, Arundhati Roy, artificial intelligence, Bernie Sanders, Equator, Arvind Narayanan, Sayash Kapoor, Anil Dash, CorridorKey, Apocalypse Now, Cyclopes, Palantir, OpenAI, Anthropic, AI4Bharat, DeepMind, Massachusetts Institute of Technology, scientific research, capitalism, AI alarmism, IIT Madras, AlphaFold, US National Bureau of Economic Research, protein folding, human extinction, auto-targeting, democratic government, public health, proprietary model, Colonialism

Arundhati Roy recently published a sweeping essay on artificial intelligence (AI) in [*Equator*](https://www.equator.org/articles/arundhati-roy-artificial-intelligence?ref=slightlyacidic.com), arguing that the technology is drastically changing the way people work, write, and think as well as how power is distributed in society. In her accurate telling, the AI development industry is dominated by billionaires who appropriate human knowledge, exploit labour, and consume enormous amounts of natural resources while increasingly entrenching themselves within governments and militaries. Drawing parallels with colonialism, she portrays AI as the culmination of capitalism’s nihilistic drive towards extraction, accumulation, and control — and calls for resistance.

But what is it that we should resist?

Our ability to think, create, learn independently is worth protecting. Anyone who has been following the [upheaval in mathematics](https://www.slightlyacidic.com/2026/10/09/mulling-ai-and-maths/) will have discerned the multitude of voices, from undergraduate students to Fields Medallists, discussing what it will mean to be a mathematician in the near future, and the repeated highlighting and more explicit articulation of what it has meant to do mathematics so far, especially its human aspects, and deliberations on how mathematics education must change to host them and how academia must change to reward them. Writing AI off wholesale is not the answer.

In fact, what do we mean when we say “AI”: the technology, the specific corporate behemoths developing its most popular forms or, more particularly, its inalienable capitalist organisation?

The development of AI as we know it today, manifested most prominently through the products of OpenAI, Anthropic, and Google, exploits workers, privatises knowledge produced collectively, contributes to environmental degradation, and increases corporate influence over governments. All these problems capitalism has produced in multiple industries. That does not make them okay, obviously, but what makes these consequences inherent to AI, as Roy repeatedly argues — and could a publicly owned, democratically governed AI system avoid any of them? That question she leaves unanswered.

According to Roy, the current trajectory of corporate AI development will all but culminate in the absolute subordination of society to a handful of corporations and the extinction of human autonomy. But I refuse to believe the future of AI is an already settled question, a foregone conclusion. For one, Roy believes that as AI advances, human intelligence is set to decline. Researchers have found important problems that arise when people overrely on AI models. One Massachusetts Institute of Technology [study](https://arxiv.org/pdf/2506.08872?ref=slightlyacidic.com) that caught many people’s attention when it was uploaded in June last year, including Roy’s, reported weaker cognitive engagement and recall among people who used ChatGPT to write essays. But the study involved only 54 people in its three rounds and an even lower 18 in the fourth, and its own authors warned against generalising their findings. The example also does not address the difference between a person asking AI to perform a task instead of doing so themselves and AI helping a person understand and investigate a problem; the latter is key to the premise of AI expanding human capabilities.

For another example, Roy contends that powerful companies like OpenAI demand regulation in order to reinforce their own legitimacy and their relationship with the state. Regulatory capture is a known occurrence and perniciously so, but not every regulatory intervention strengthens corporate power. In 2023, for example, Italy temporarily restricted public access to ChatGPT until OpenAI introduced stronger privacy protections, including allowing users to opt out of having their conversations with the model to be used for training. Since 2025, through its AI Act, the European Union has prohibited AI companies from scraping data pertaining to users’ faces to build facial recognition databases. Governments can also hold companies legally liable for harms rendered by their products and enforce competition laws to prevent a few dominant firms from excluding many smaller competitors.

Taking this argument a step further: Roy says AI companies that warn of (or whose employees warn of) human extinction in any form do so in order to make their technologies appear incredibly powerful and worthy of enormous investments, if nothing else to allow the companies to ‘safely’ harness them for humankind’s benefit. Fine. But then she also describes AI as a technology that will destroy the world as we know it. If apocalyptic speculation helps to legitimise the AI industry’s claims about its own transformative power, her own apocalyptic account risks reinforcing the narrative of technological omnipotence she criticises the industry for promoting.

Discussing such claims solely on the basis that they could benefit corporations is to say nothing about where the dangers lie and what we can do about them. I refuse to discard a technology just because it is politically useful somewhere as long as I make sure that I’m aware of the social and technical underpinning and that I’m using it in a conscionable way — and this is what I would like for everyone. In fact, my colleagues at *The Hindu*, Srinivasan V.R., Pon Vasanth B.A., and Sambavi Parthasarathy, among other journalists, have used the same technology to analyse the Special Intensive Revision of the Election Commission of India with enough granularity to reveal the Commission’s and the Indian government’s [prejudice against Muslims](https://www.thehindu.com/data/muslims-disproportionately-flagged-in-delhi-sir/article71580772.ece?ref=slightlyacidic.com) hidden in several terabytes of data.

There is also not inconsiderable development happening outside the dominant corporate model. As part of the AI4Bharat initiative at IIT Madras, for example, scientists are developing open-source language datasets and tools to translate between, and beyond, India’s scheduled languages. In a [working paper](https://www.nber.org/papers/w35143?ref=slightlyacidic.com) published in April, researchers at the U.S.-based National Bureau of Economic Research reported that “basic research on proteins that had no structure information prior to AlphaFold” — the AI tool developed by DeepMind to [understand how proteins fold](https://www.thehindu.com/sci-tech/science/a-new-era-the-2024-nobel-prizes-and-the-power-of-machine-learning/article68739836.ece?ref=slightlyacidic.com) — “increases by 15 to 40% relative to proteins that already had a structure, shifting the direction of research toward less-studied proteins.” Yet even when Roy acknowledges AI’s considerable applications in medical research, she immediately reminds readers that poor people continue to die of preventable diseases. One is an argument about the utility of medical AI; the other is an indictment of public healthcare.

Even if AI4Bharat and DeepMind are neither necessarily free from having to depend on corporate infrastructure nor being able to evade the environmental degradation and ethical issues attached to AI development, they also establish that we are developing some AI-based capabilities and sharing them outside the proprietary systems Roy focuses on. Training a proprietary model on some work neither necessarily transfers ownership of the work nor renders the knowledge nonrivalrous. There is admittedly a gap between scientific benefits and their subsequent social benefits just as much as there is a conversation to be had over whether we should reject a technology capable of advancing scientific understanding because it was developed within an exploitative economic system or whether that presents an argument to transform the system.

As U.S. entrepreneur and writer Anil Dash put it in an October 2025 [blog post](https://www.anildash.com/2025/10/17/the-majority-ai-view/?ref=slightlyacidic.com): “Technologies like \[large language models\] have utility, but the absurd way they’ve been over-hyped, the fact they’re being forced on everyone, and the insistence on ignoring the many valid critiques about them make it very difficult to focus on legitimate uses where they might add value.”

Roy’s own characterisation of AI use by armies as constituting the ‘delivery’ of guilt-free murders is tantamount to conceding that humans and their decisions to deploy auto-targeting systems are irrelevant — and that is disagreeable: we ought to challenge those decisions and the people who make them.

“To view AI as normal is not to understate its impact—even transformative, general-purpose technologies such as electricity and the internet are ‘normal’ in our conception,” to quote computer scientists [Arvind Narayanan and Sayash Kapoor](https://www.normaltech.ai/p/ai-as-normal-technology?ref=slightlyacidic.com). “But it is in contrast to both utopian and dystopian visions of the future of AI which have a common tendency to treat it akin to a separate species, a highly autonomous, potentially superintelligent entity. … We view AI as a tool that we can and should remain in control of, and we argue that this goal does not require drastic policy interventions or technical breakthroughs.”

Even the way Roy frames the environmental cost of data centres allows little room to distinguish between the consequences of local authorities allowing companies to set these facilities up in water-stressed areas and those when authorities declined permission. It is her prerogative to conclude the costs outweigh the benefits, as she does, but her chain of reasoning is missing.

Her tendency to treat the whole industry or enterprise as a monolith is of a piece with her ultimate call to “terminate” AI development “with extreme prejudice”. Her Cyclopes metaphor, buttressed by the image of “steel threads of societal control” being drawn “together into one clenched, mechanical fist”, combines Palantir, OpenAI, Anthropic, other corporations, investors, and contractors into a single abomination with a unified purpose. But while they are all interested in protecting capital, even they do not have the same commercial incentives or relationships with the state, not to speak of the numerous smaller companies.

In fact, in [his discussion](https://www.anildash.com/2026/04/28/one-good-ai-is-here/?ref=slightlyacidic.com) around what he calls an example of “good AI” — a tool called CorridorKey that helps CGI folks simplify their work with green screens — Mr. Dash writes that the big AI companies’ “bad behaviours are a choice”, from using people’s data without their consent to beating people over their heads about how they really need AI models. To dismiss this reality is to dismiss the divergent effects regulations could bring to bear on them, including through competition policy, restrictions on military contracts, and conditions on funding research.

Finally, Roy describes Bernie Sanders’s proposal as “getting shares in the apocalypse”. The proposal is to redistribute corporate profits but, crucially, also to establish public ownership in a way that could change who has the power to decide how companies develop AI models. “Getting shares in the apocalypse” is a reductive view of this notion. Similarly, her objection to Sanders on the grounds that his proposal will redistribute wealth only within the Global North while allowing the companies to continue to exploit knowledge and labour from the Global South should first raise the possibility of internationalising the arrangement, not dismissing it altogether.

In all, Roy excels at exposing the power relations that govern AI development today, but their mere existence or diagnosis does not amount to an inevitable destiny.