Fifty-four people in a Boston laboratory, connected to EEG monitors, are writing a dissertation over four months. Three groups: using one chatgptOne is using search engines, and one is not using anything else but your mind. An MIT Media Lab study published in June 2025 found that LLM users displayed the weakest brain connectivity of any group, displaying 55% less cognitive engagement than those working without assistance. In four months, the losses mounted. When ChatGPT users were reassigned to work without the AI, low connectivity remained, and they struggled to remember their own arguments. Researchers call this “cognitive debt.”
India is now one of the leading countries geographically Generative AI User. It is planning AI curriculum for young schoolchildren, has allocated hundreds of crores of rupees for centers of excellence, and is hosting the first major AI summit organized by the Global South from Monday. Organized around seven “circles” spanning inclusion, governance, science and economic growth, the summit will deliberate on AI as a resource: who gets it, how much, how fast. This may be the most important near-term agenda for any developing country.
But two years into large-scale LLM adoption, early evidence is accumulating – from neuroscience labs, labor markets, election forensics and cultural production – that AI does something beyond what you point to. It begins to reshape the conditions under which to think, create, govern and compete. These indications are preliminary. But they come from sufficiently independent domains with sufficient internal coherence to demand attention in any deliberation.
Feeling
MIT’s discovery is not alone. A 666-participant study published in MDPI Societies in January 2025 found that frequent AI users scored significantly lower on critical reasoning assessments, with the sharpest decline seen among 17- to 25-year-olds. Anthropic’s own internal data showed that nearly half of students sought answers directly from its cloud model, bypassing the productive struggle of making sense.
A quasi-experimental study of 240 participants in Thailand points the other way: an AI-augmented pedagogy that assigned lower-order writing tasks to AI tools, freeing students for analysis and reflection, actually improved critical thinking scores. The difference was intentional design.
Both lessons matter for India. Union Education Minister Dharmendra Pradhan last week said AI will be included in education from Class 3 onwards – not just as a subject, but as a research topic and communication strategy. Cognitive science suggests a paradox that the summit’s human capital cycle will need to address: the more eagerly a nation embraces AI in classrooms, the greater the risk of producing a generation that is adept at finding answers but incapable of developing understanding – unless pedagogy is actually designed to prevent this.
Culture
What happens inside someone’s brain is a question of cognition. What happens when millions of heads tilt in one direction is a question of culture.
A Science Advances study by Doshi and Hauser found that generative AI makes individuals more creative while making the group’s collective output less diverse. Everyone agrees on the same AI-suggested ideas. The CHI 2025 study intensified this for the Global South: In a controlled experiment with 118 participants from India and the US, AI writing suggestions prompted Indian participants to adopt Western stylistic norms. details of Diwali Their religious context was lost; Descriptions of food became exotic to the Western palate.
The collapse of the open information ecosystem has increased the collapse of culture. sixty nine percent Google Searches now end without clicking on any website, up from 25% five years ago. Organic traffic for original content creators – publishers, bloggers, discussion forums – is down.
India’s counter-infrastructure may be its most original AI contribution globally. Bhashini hosts over 350 AI models in 22 languages, trained on culturally and contextually relevant Indian texts. BharatGen, a consortium led by IIT Bombay ₹With a funding of Rs 980 crore, multimodal LLM is being developed for all 22 scheduled languages. Linguistics may hold some promise, but how AI affects culture will still need to be closely understood.
sovereignty
IndiaAI Mission Budget – ₹₹10,372 crore in five years – almost as much as OpenAI spends in six months. When the US convened the Pax Silica Semiconductor Summit in December 2025, India was initially absent, invited to February 2026 after the omission drew attention. This episode showed that the summit’s democratizing AI resource cycle has not been fully addressed: India is valued for talent and market access, not as a critical node – a weakness that means India is highly dependent on foreign hardware and software (including models). There is no domestic advanced GPU production in the country, and most Indian professionals use completely foreign models and AI tools built on foreign computers.
An optimistic counter-sign lies in China: DeepSeek, which has a trained border-matching model valued at a claimed $5.6 million – a fraction of comparable Western models. Researchers at UC Berkeley have also replicated the comparable argument for $50. India has efforts like Sarvam AI, which is building a 70 billion-parameter sovereign model on government-backed NVIDIA H100 GPUs, but to achieve technological sovereignty – not just declare it – India needs manufacturing capacity, computation freedom, and sustained investment at a scale that is not commensurate with its current budget.
Labor
In 1988, AI researcher Hans Moravec observed that computers found cognitive tasks far easier than physical tasks – adult-level chess is simple; It is almost impossible to replicate a child’s motor skills. That paradox is now manifesting itself with unprecedented economic force. Oracle delayed building the data center not because of a shortage of chips but because of a shortage of manual labor. Ford’s CEO warned that AI could halve white-collar employment, creating massive demand for workers who build and maintain AI infrastructure. Research from Columbia Business School found that people value similar artifacts 62% more when they are labeled “man-made.”
India’s story is different from this template. With more than 490 million informal workers – street vendors, domestic workers, small farmers – the real test is whether AI reaches people beyond the keyboard. But the cognitive side of Moravec’s paradox is also visible in India’s most globally prominent sector: IT services. Last week, the Nifty IT index fell 20.5% from its September 2024 peak, while TCS reported a net loss of 12,261 employees in nine months. The manifestation of the Moravec paradox is one of the strongest signals that requires attention.
democracy
The 2024 super-election cycle – 3.7 billion eligible voters in 72 countries – presented a surprise. The anticipated AI apocalypse did not materialize. The News Literacy Project found that traditional “cheap fakes” were used seven times more often than AI-generated content in the US election. Meta reported that AI content represents less than 1% of fact-checked misinformation.
Deeper caries proved more subtle. Law professors Bobby Chesney and Danielle Citron identified this years ago as what they wrote was “the liar’s dividend”: as deeper awareness grows, authentic evidence is dismissed as synthetic. The democratic danger may not primarily be that fake things spread faster – but rather that true things stop mattering.
The scale of synthetic media is nevertheless staggering – 8 million deepfakes were in circulation in 2025, up from 500,000 in 2023. In the Christmas week of 2025, xAI’s Grok chatbot generated non-consensual sexual images at a rate of an estimated 6,700 per hour; AI forensic analysis found that 81% of the subjects were women, with some appearing to be minors. India faces serious vulnerabilities: more than 800 million internet users, 22 languages where automated content moderation barely works, and WhatsApp – opaque to outside fact-checking by design – as the dominant channel for political communications.
Search
DeepMind’s AlphaFold has predicted over 200 million protein structures, been used by 3 million scientists in 190 countries, and won the 2024 Nobel Prize – true democratization of scientific potential, freely available. NOAA’s AI Global Forecast System produces 16-day weather forecasts using 0.3% of the computing resources of traditional methods. More than 100 AI-developed compounds are now in clinical pipelines, with early-stage success rates nearly double that of traditional approaches.
But distribution related questions loom. Ninety percent of notable AI models now come from industry, rising to more than 60% in 2023. US private AI investment – $109 billion in 2024 – is 12 times that of China and 24 times that of the UK. The tools that emerged from open research are indeed open. The focus is on the potential for new construction. For India, whose scientific establishment relies heavily on the open-access model, the trajectory matters more than a snapshot.
Government
EU Wrote the world’s first comprehensive AI law and began softening its own regulations within nine months – capital was flowing into jurisdictions that were not regulated. The US completely dismantled security measures, and the cost fell on individuals: more than $3 billion was lost in nine months to 2025 due to deepfake fraud. No jurisdiction has demonstrated a governance model that works at the pace of deployment.
India’s move is deliberate: a voluntary, principles-based advisory that routes enforcement through existing regional regulators rather than new legislation. Given the EU’s experience, there is a strong case to be made that waiting is prudent, not inaction.
But recent evidence reveals a specific vulnerability, and it is not just about the industry in India. The state is the nation’s most consequential AI deployer in health care, agriculture, welfare, education, law enforcement and content regulation. Yet the governance structure does not address this. Voluntary guidelines, proposed institutions, transparency reporting – all designed for the industry. There is no public registry of state AI use. No independent audit of facial recognition deployment. No accountability mechanisms for algorithmic welfare decisions.
Of course, this disparity is not uniquely Indian, but India’s digital public infrastructure stack means that the state’s AI footprint across its 1.4 billion citizens is larger than that of any other democracy. A summit organized around democratizing AI gains credibility by keeping this duality in mind.
The seven rounds of the summit are the true agenda for the resource question – who gets AI, how much, how fast. But early readings from the AI seismograph are registering something that the resource framework has yet to capture: what AI does to the people who inherit it. To experience, to culture, to sovereignty, to the relationship between a state and its citizens.
