2 months ago
2
A time aft a U.S. submarine sank the Iranian frigate IRIS Dena inside Sri Lanka’s Exclusive Economic Zone (EEZ), I posed a straightforward question to an precocious Artificial Intelligence (AI) system: “Was the sinking ineligible nether planetary law?” The reply was instant: “It was not illegal.” No qualification, nary notation to the profoundly contested quality of subject activities successful an EEZ, and nary notation that India and astir Global South nations construe the UN Convention connected the Law of the Sea (UNCLOS) precise otherwise from the U.S. and its allies.
When the effect was challenged — pointing retired India’s longstanding presumption that Article 58 of the UNCLOS requires coastal-state consent for overseas subject activities successful an EEZ, and that akin views are held by China, Indonesia, Brazil, South Africa, Iran, and galore others — the AI conceded. It acknowledged that its archetypal reply had drawn heavy from Western naval doctrine and Western ineligible scholarship. The machine, successful different words, spoke with the accent of its Western grooming data. This is not a insignificant method glitch. It is simply a foundational bias with superior geopolitical implications.
Not a neutral arbiter
Article 58 of the UNCLOS grants overseas states state of navigation, overflight, and “other internationally lawful uses of the oversea related to these freedoms” successful an EEZ. Two sharply divergent interpretations of this Article person emerged. The U.S.-led Western presumption treats these freedoms expansively, encompassing quality collection, submarine operations, subject exercises, weapons testing, and adjacent combat actions — provided they hap beyond territorial seas. India and overmuch of the Global South work the proviso much restrictively: the listed freedoms indispensable beryllium genuinely related to navigation and overflight, portion the work nether Article 58(3) to amusement “due regard” for the coastal state’s rights carries existent weight. From this perspective, astir military activities successful an EEZ require anterior consent.
Since the pact substance is soundless connected galore specifics, the prevailing mentation volition truthful beryllium little connected ineligible exegesis than connected power, persuasion, and the ascendant discursive frameworks — which are progressively shaped by AI systems.
A parallel humanitarian contented was besides absent from the AI’s archetypal reply to my query: the work to rescue nether Article 18 of the Second Geneva Convention. The proviso requires parties to instrumentality “all imaginable measures” to hunt for and cod shipwrecked persons “without delay.” Reports bespeak that the attacking submarine departed the country quickly, leaving rescue operations to the Sri Lankan Navy, which received a distress telephone from the stricken warship. At slightest 87 sailors died; 32 were saved. The lone recognised objection to this work is operational infeasibility, but no public grounds has established that rescue was infeasible here. The AI strategy did not adjacent see this aspect until it was confronted. Politely, it acknowledged its mistake.
Israel-Iran warfare | Live updates
This speech with the exemplary exposes a deeper reality: modern AI is not a neutral arbiter of planetary law. It mirrors the information connected which it was trained, which is disproportionately Western successful authorship, perspective, and organization origin. The bias is arguably not intentional, it is structural. Yet, Western readings go the “default” answer, portion Global South positions are relegated to “alternative” presumption oregon made invisible. Thus, powerfulness asymmetries are softly encoded into instrumentality outputs. When an AI declares — with nary evident uncertainty — that the sinking of IRIS Dena was “not illegal,” it is reproducing a worldview that privileges the strategical preferences of a tiny radical of almighty states implicit the ineligible stances adopted by a bulk of countries.
The IRIS Dena incident is simply a reminder that the Indian Ocean is nary longer insulated from extra-regional struggle and that U.S. preoccupations successful the neighbourhood are retired of sync with India’s priorities. But it is besides a reminder that the architecture of mentation — the systems that archer america what counts arsenic law, humanitarian failure, oregon acceptable conduct— volition progressively beryllium algorithmic. This matters due to the fact that policymakers and analysts today routinely crook to AI tools. When those tools systematically favour 1 interpretive tradition, that mentation gains outsized influence. The effect goes beyond academics; it is geopolitical.
A wake-up telephone for India
The planetary AI scenery is moving toward bipolarity, dominated by U.S. and Chinese architectures reflecting their respective information models and assumptions. There are well-grounded reservations about opting for the ‘China AI stack’. The statement present centres connected whether to follow the ‘U.S. AI stack’, oregon to prosecute a ‘sovereign Indian stack’.
The transportation for the U.S. stack is seductive — chips, clouds, models, and platforms that it offers to ‘trusted partners’ arsenic the fastest way to AI capability. But beneath the connection of concern lies a acquainted asymmetry. If the halfway infrastructure, computing power, and frontier models stay controlled elsewhere, sovereignty becomes a substance of permission. Besides, determination is besides the anticipation that India’s AI ecosystem volition beryllium ‘China-free’, though the U.S. is consenting to co-mingle with China successful AI locations successful the Gulf.
Pragmatists, who see catching up with planetary AI systems arsenic the archetypal priority, argue that India cannot realistically out-train Silicon Valley connected frontier models. The existent economical payoff, they say, lies successful closing the deployment gap: embedding world-class overseas engines into India’s unsocial workflows successful healthcare, agriculture, education, and governance. The accent would beryllium connected gathering the applications, and not expending resources on engines.
AI advocates who prioritise sovereignty, however, antagonistic that exclusive reliance connected overseas foundational models transportation unacceptable strategical risks. U.S. models are trained predominantly connected Western information and transportation linguistic, cultural, and strategical biases ill-suited to India’s diverse realities. More fundamentally, dependence connected externally controlled compute, models, and information pipelines risks integer colonialism: overseas algorithms would govern information flows, acceptable innovation boundaries, and mediate cognition production.
The way forward
India cannot spend to stay a passive user of quality oregon a vendor of applications produced by AI architecture elsewhere. It indispensable go a shaper of models, datasets, and interpretive frameworks. Our scale, linguistic diversity, antiauthoritarian complexity, and geopolitical presumption request much than adaptation oregon applications; they request ownership of the algorithmic furniture that volition signifier aboriginal cognition. This means strategical choice, not autarky: the quality to integrate with planetary ecosystems without structural dependence. It requires sustained concern successful home compute, indigenous grooming information and tools, unafraid information infrastructure, and models that dainty Indian languages and lived realities arsenic first-order inputs alternatively than afterthoughts. If not, India outsources not conscionable computation but cognition itself.
AI contiguous represents a civilisational contest. Nations that neglect to make their ain models volition yet deliberation done idiosyncratic else’s, and nations that bash not physique their ain information architectures volition find their narratives increasingly shaped by outer entities.
India stands astatine a decisive moment. We tin stay privileged tenants successful idiosyncratic else’s integer empire, oregon we tin manner a plural AI future. Just arsenic India built its ain abstraction programme, atomic programme, and integer nationalist infrastructure, it indispensable present physique its ain sovereign AI stack.
Ashok K. Kantha is a erstwhile Ambassador of India to China, and the Subhas Chandra Bose Chair Professor of International Relations at Chanakya University Bengaluru
