What Raja Mohan's AI Optimism Misses: The Diplomatic Cost of Hallucinated Authority

C. Raja Mohan's recent argument that artificial intelligence represents a "force multiplier" for diplomacy deserves serious engagement. The distinguished strategic analyst correctly identifies the transformative potential of AI tools and Singapore Foreign Minister Vivian Balakrishnan's early adoption of AI agents signals genuine innovation in diplomatic practice. Yet the core premise — that foreign offices face an "adapt or fall behind" binary — misunderstands what diplomacy actually is.

Diplomacy is not primarily about producing analytical drafts or managing information flows. It is about reading cultural signals, building trust across decades, and interpreting the unspoken dynamics that shape international relations. These capabilities emerge from embodied experience, institutional memory, and cultural literacy — precisely the domains where current AI systems reveal fundamental limitations.

The Memory Problem: Filing Is Not Remembering

Current large language models suffer from what cognitive scientists call anterograde amnesia — they cannot form continuous memories across sessions. Balakrishnan's "NanoClaw" agent addresses this through graph database workarounds — sophisticated filing systems that create the appearance of memory. But diplomatic memory is interpretive, not archival.

An IFS officer who served in Tehran during the nuclear negotiations understands Iranian diplomatic culture in ways that cannot be reduced to searchable records. The officer knows when a particular phrasing signals flexibility, when silence indicates internal consultation, when a cultural reference carries strategic weight. This knowledge emerges from pattern recognition across thousands of interactions, filtered through cultural context and institutional wisdom.

The distinction matters because diplomacy operates in the space between the written and the understood. When External Affairs Minister S. Jaishankar references the Mahabharata in multilateral settings, he draws on civilisational knowledge that shapes how counterparts interpret India's positions. An AI system trained on text cannot replicate this cultural fluency because it lacks the embodied experience that gives meaning to such references.

Cultural Literacy: Beyond Pattern Matching

Raja Mohan's "second brain" framing reflects a fundamental misunderstanding of how cultural knowledge operates in diplomacy. LLMs are trained on text with no embodied cultural experience. They can identify patterns in how Gulf monarchies phrase official statements, but they cannot read the subtle variations that signal policy shifts to practitioners who have spent years in the region.

Indian diplomacy particularly relies on this cultural sophistication. The same IFS officer adjusts register when addressing Tamil Nadu political leaders versus Telugu counterparts, understands the different weights carried by Sanskrit versus Persian metaphors in Central Asian contexts, knows when British parliamentary procedure applies in Commonwealth settings and when it becomes counterproductive.

These capabilities develop through sustained cultural immersion. An AI system can parse the textual patterns of diplomatic language, but it cannot interpret the embodied signals — gesture, timing, context — that often carry more meaning than the formal communiqué.

The Irreducible Shape of Experience

Raja Mohan's vision of "five-person delegations with sovereign AI matching fifty-person missions" misunderstands what diplomatic missions actually do. A senior IFS officer shaped by postings across Latvia, Tehran, and Geneva carries irreducible knowledge — understanding of how different institutions work, relationships built across career trajectories, instincts honed through crisis management.

This experiential knowledge cannot be summarised or systematised. When India navigates complex trilateral relationships — managing simultaneous partnerships with Russia, the United States, and Iran — the decisions rest on diplomatic judgment accumulated across decades of similar balancing acts. The institutional memory held by India's diplomatic corps represents strategic depth that no AI system can replicate.

The fifty-person mission holds relationships, not just analytical capacity. These relationships — built through shared experiences, mutual respect, and demonstrated competence — create the trust necessary for sensitive negotiations. An AI agent cannot build such relationships because it cannot offer the human recognition that underpins diplomatic confidence.

Hallucination as Architecture: The Professional Evidence

The most concerning aspect of Raja Mohan's argument is the casual dismissal of AI hallucination risks. Current LLMs generate responses through statistical likelihood with no internal mechanism for distinguishing known from unknown information. In professional contexts that have stress-tested AI output against verifiable standards, the results are sobering.

The legal profession provides the clearest evidence. In Mata v. Avianca, lawyers were sanctioned $5,000 each for submitting ChatGPT-generated briefs containing six fabricated citations, including the nonexistent "Varghese v. China Southern Airlines." In Johnson v. Dunn, attorneys were disqualified from representing their client, with bar regulators in every state notified of the misconduct.

India's courts have encountered identical problems. The Supreme Court issued notice to the Attorney General, Solicitor General, and Bar Council of India after trial courts relied on AI-generated judgments, calling it professional misconduct. Justice B.V. Nagarathna encountered the fictitious "Mercy v. Mankind" case in a public interest litigation. Delhi High Court rejected a ChatGPT-crafted petition that cited nonexistent paragraph 73 of Raj Narain v. Indira Gandhi — a judgment with only 27 paragraphs.

The pattern extends across jurisdictions. ITAT Bengaluru recalled its own order after discovering AI-fabricated case laws in the Buckeye Trust matter. Punjab & Haryana High Court reprimanded lawyers for using AI during live hearings, stating that "AI cannot replace actual intelligence."

These cases matter for diplomacy because they demonstrate systematic failure in the one profession that routinely verifies AI output against ground truth. Legal practice has opposing counsel, appellate review, and citation checking as quality controls. Diplomacy operates without equivalent safeguards. A hallucinated treaty clause or fabricated precedent in a diplomatic demarche has no Judge Castel to catch it before it shapes policy decisions.

The Real Asymmetry

Raja Mohan's analysis inverts the actual competitive dynamic. The relevant gap is not between AI-adopting and AI-avoiding diplomatic corps, but between institutions with process discipline to verify AI output and those without systematic quality controls.

India's Supreme Court response to AI hallucination demonstrates institutional strength — senior justices immediately recognised fabricated citations and established verification protocols. This reflects the kind of institutional depth that creates competitive advantage. Countries that should worry are those whose junior officers will paste LLM drafts into sensitive communications without the cultural knowledge and institutional memory necessary to identify errors.

India's diplomatic tradition is precisely the kind of institutional asset that AI cannot replicate. The patient relationship-building that characterises Indian statecraft, the cultural literacy that enables effective engagement across diverse bilateral partnerships, the institutional memory that guides complex multilateral negotiations — these capabilities emerge from human experience and cultural knowledge that no current AI system can duplicate.

The question is not whether India's foreign ministry should adopt AI tools. Selective automation of routine tasks makes obvious sense. The question is whether Indian diplomacy should subordinate its cultural sophistication and institutional wisdom to systems that cannot distinguish between authoritative knowledge and statistical plausibility. The evidence from legal practice suggests that would be a profound strategic error, trading India's competitive advantages for tools that cannot deliver what they promise.