AI Summit 2026 - The new knowledge world: Seven signals from the artificial intelligence seismograph

FIfty- quaternity people in a beantown research laboratory, wired to EEG monitors, written material essays o'er four months. Three groups: one using ChatGPT, one using a search engine, and one using nothing but their own minds. The MIT Media Lab study, published in June 2025, found that LLM users displayed the weakest brain connectivity of any group, up to 55% lower cognitive engagement than those working unaided. Over four months, the deficit compounded. When ChatGPT users were reassigned to work without AI, the reduced connectivity persisted, and they struggled to recall their own arguments. The researchers called it “cognitive debt”.India is now one of the leading geographies in terms of generative AI users. It plans an AI curriculum for young schoolchildren, has allocated hundreds of crores for centres of excellence, and, starting on Monday, hosts the first major AI summit convened by the Global South. Organised around seven “chakras” spanning inclusion, governance, science, and economic growth, the summit will deliberate on AI as a resource: who gets it, how much, how fast. It may be the most important near-term agenda any developing country can pursue.But two years into mass LLM adoption, early evidence is accumulating — from neuroscience labs, labour markets, election forensics, and cultural production — that AI does something beyond what you point it at. It begins to reshape the conditions under which thinking, creating, governing, and competing take place. These signals are preliminary. But they stem from enough independent domains, with enough internal coherence, to demand any deliberation’s attention.CognitionThe MIT finding does not stand 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 17- to 25-year-olds showing the steepest decline. Anthropic’s own internal data revealed nearly half of students asked its Claude model for direct answers, bypassing the productive struggle that builds understanding.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 said this past week that AI will be incorporated into education starting from the third grade — not only as a subject, but as a research theme and communication strategy. Cognitive science suggests a paradox the summit’s Human Capital chakra will need to address: the more eagerly a nation adopts AI in classrooms, the greater the risk of producing a generation fluent in finding answers but incapable of developing understanding — unless the pedagogy is designed to prevent exactly that.CultureWhat happens inside one's head is the cognition question. What happens when millions of heads are nudged in the same direction is the culture question.A Science Advances study by Doshi and Hauser found that generative AI makes individuals more creative while making the collective output of a group less diverse. Everyone converges on similar AI-suggested ideas. A CHI 2025 study sharpened this for the Global South: in a controlled experiment with 118 participants from India and the US, AI writing suggestions led Indian participants to adopt Western stylistic norms. Descriptions of Diwali lost their religious context; food descriptions became exoticised for a Western palate.The flattening of culture is compounded by the collapse of the open information ecosystem. Sixty-nine per cent of Google searches now end without a click to any website, up from 25% five years ago. Organic traffic to original content creators — publishers, bloggers, discussion forums — has plummeted.India’s counter-infrastructure may be its most original AI contribution globally. Bhashini hosts over 350 AI models in 22 languages, trained on texts that are culturally and contextually Indian. BharatGen, the IIT Bombay-led consortium with ₹980 crore in funding, is developing a multimodal LLM for all 22 scheduled languages. That linguistics could hold some promise, but how AI influences culture will still need to be closely understood.SovereigntyIndia’s IndiaAI Mission budget — ₹10,372 crore over five years — is roughly what OpenAI spends in six months. When the US convened the Pax Silica semiconductor summit in December 2025, India was initially absent, invited only in February 2026 after the omission drew attention. The episode revealed what the summit’s Democratising AI Resources chakra has not fully confronted: India is valued for talent and market access, not as a critical node – a weakness that has meant that India relies inordinately on foreign hardware and software (including models). The country has no domestic advanced GPU production, and most Indian professionals use AI tools built entirely on foreign models and foreign compute.A hopeful counter-signal lies in China: DeepSeek, which is a trained frontier-matching model for a claimed $5.6 million – a fraction of comparable Western models. UC Berkeley researchers have also replicated comparable reasoning for $50. India has efforts such as Sarvam AI, which is building a 70-billion-parameter sovereign model on government-backed NVIDIA H100 GPUs, but to achieve technological sovereignty — not merely declare it — India needs fabrication capacity, compute independence, and sustained investment at a scale its current budget does not approach.LabourIn 1988, AI researcher Hans Moravec observed that computers find cognitive tasks far easier than physical ones — adult-level chess is simple; a toddler’s motor skills are nearly impossible to replicate. That paradox is now asserting itself with unexpected economic force. Oracle delayed data centre builds not for lack of chips but for lack of physical labour. Ford’s CEO warned AI could halve white-collar employment while creating mass demand for the workers who build and maintain AI infrastructure. Columbia Business School research found people valued identical artworks 62% higher when labelled “human-made.”India’s story diverges from this template. With over 490 million informal workers — street vendors, domestic workers, small farmers — the real test is whether AI reaches the people furthest from the keyboard. But the cognitive side of Moravec’s paradox is visible too in India’s most globally prominent sector: the IT services. Last week, the Nifty IT index fell 20.5% from its September 2024 peak, while TCS reported a net reduction of 12,261 employees in nine months. The manifestation of the Moravec paradox is one of the strongest signals that needs attention.DemocracyThe 2024 super-election cycle — 3.7 billion eligible voters across 72 countries — delivered a surprise. The anticipated AI apocalypse did not materialise. The News Literacy Project found traditional “cheapfakes” were used seven times more often than AI-generated content in the US election. Meta reported AI content represented less than 1% of fact-checked misinformation.The deeper corrosion proved subtler. Law professors Bobby Chesney and Danielle Citron identified it years ago as what they wrote to be the “liar’s dividend”: as deepfake awareness grows, authentic evidence gets dismissed as synthetic. The democratic threat may not primarily be that fake things spread faster — it is that true things stop mattering.The scale of synthetic media is nonetheless staggering — 8 million deepfakes in circulation in 2025, up from 500,000 in 2023. Over the Christmas week of 2025, xAI’s Grok chatbot generated nonconsensual sexualised images at an estimated 6,700 per hour; an AI Forensics analysis found 81% of the subjects were women, some appearing to be minors. India faces acute vulnerability: over 800 million internet users, 22 languages where automated content moderation barely functions, and WhatsApp — opaque to external fact-checking by design — as the dominant channel for political communication.DiscoveryDeepMind’s AlphaFold has predicted over 200 million protein structures, been used by 3 million scientists across 190 countries, and won the 2024 Nobel Prize — a genuine democratisation of scientific capability, freely available. NOAA’s AI Global Forecast System produces 16-day weather forecasts using 0.3% of the computing resources of traditional methods. Over 100 AI-developed compounds are now in clinical pipelines, with early-stage success rates roughly double conventional approaches.But the distributional question looms. Ninety per cent of notable AI models now come from industry, up from 60% in 2023. US private AI investment — $109 billion in 2024 — is 12 times China’s and 24 times the UK’s. The tools that emerged from open research are genuinely open. The capacity to build new ones is concentrating. For India, whose scientific establishment relies heavily on open-access models, the trajectory matters more than the snapshot.GovernanceThe EU wrote the world’s first comprehensive AI law and within nine months began softening its own rules — capital was flowing to jurisdictions that did not regulate. The US scrapped safeguards entirely, and the costs landed on individuals: deepfake fraud caused over $3 billion in losses in nine months of 2025. No jurisdiction has demonstrated a governance model that works at the speed of deployment.India’s bet is deliberate: a voluntary, principle-based advisory that routes enforcement through existing sectoral regulators rather than new statute. Given the EU’s experience, there is a reasonable case that waiting is prudence, not inaction.But the recent evidence reveals a specific vulnerability, and it is not about industry, in India. The State is the country’s single most consequential AI deployer — across healthcare, agriculture, welfare, education, law enforcement, and content regulation. Yet the governance framework does not address this. The voluntary guidelines, the proposed institutions, the transparency reporting — all are designed for industry. There is no public registry of state AI use. No independent audit of facial recognition deployment. No accountability mechanism for algorithmic welfare decisions.To be sure, this asymmetry is not uniquely Indian, but India’s Digital Public Infrastructure stack means the state’s AI footprint on 1.4 billion citizens is larger than in any other democracy. A summit framed around democratising AI gains credibility precisely by reckoning with this duality.The summit’s seven chakras are the right agenda for the resource question — who gets AI, how much, how fast. But the early readings from the AI seismograph are registering something the resource framework does not yet capture: what AI does to those who get it. To cognition, to culture, to sovereignty, to the relationship between a state and its citizens.