ILLUSTRATION: RHODE ISLAND THEATRE MAKERS ROUNDTABLE
NEPAL’S EXPERT class has a certain texture you learn to recognise. The panel discussion starts at the five-star hotel. The moderator introduces someone as a “renowned economist and policy analyst”. The camera pans to a man who worked at the finance ministry for three years in the 1990s, since when he has appeared on television some 700 times, given his views on monetary policy, trade deficits, federal restructuring, agricultural development, cryptocurrency regulation and the geopolitics of the India-Nepal border. He has published no peer-reviewed research. Neither has he produced any original economic data. He has however produced opinions continuously, at high volume, for thirty years, and the opinions have solidified into a reputation that is indistinguishable from expertise to anyone who doesn’t know the difference.
That is not a petty problem. In fact it is the operating system of Nepal’s policy environment.
Start with the growth forecast. Every May the finance minister announces a budget targeting 7% GDP growth. Every year the economy delivers something in the range of 3-5%. Every year a parade of economists, consultants and “policy analysts” appears on television to endorse the 7% target as ambitious but achievable. The Asian Development Bank projects 3.9. The World Bank projects 2.3. Nepal’s National Statistics Office delivered 3.85% for the year just ended. None of the panel endorsers faced any professional consequence for having been wrong. The same people will be on the same panels next May endorsing next year’s target. The target functions as a ritual, and the experts are the priests.
The Pokhara International Airport is the ugliest example of what this costs. The feasibility study commissioned before construction projected 280,000 international passengers a year through Pokhara by 2025. As of July 2026 the airport has no scheduled regular international flights. Someone produced that projection. Someone reviewed it. Someone with expert credentials signed off on a $215.96m Chinese loan for a project whose stated demand did not exist. A parliamentary sub-committee’s investigation into the airport’s procurement found corruption and incompetence throughout the project’s design phase. The experts who produced the feasibility numbers were not prosecuted. They were probably commissioned for the next project.
Now add AI and the problem gets worse in a way that is not being discussed honestly.
Nepal’s newspapers — English-language outlets in particular — have been filling their analysis sections with articles generated by ChatGPT, Claude and Gemini, edited minimally, and published under bylines that imply original reporting and thought.
This is not speculation: the tell-signs are obvious to anyone who reads widely: the “on one hand… on the other hand” structure, the inability to name a specific person or cite an actual number that wasn’t already circulating online, the prose that sounds authoritative in each individual sentence and accumulates into nothing by the end of the paragraph. The pieces say things like “Nepal must develop a robust framework for sustainable development while ensuring that fiscal discipline remains at the forefront of budgetary considerations.” The sentence contains no information. It is a sentence-shaped arrangement of words that policy people have been trained, by years of donor-funded workshops, to produce and to receive as consequential.
Here is the real damage. AI is better at producing this genre of content than humans are: because the genre itself is content-free and AI is very good at generating fluent content-free prose. It knows what an expert opinion piece looks like. It knows the moves: state the problem, acknowledge the complexity, cite a general trend, recommend a robust framework, close with a call for stakeholder engagement. The output is indistinguishable from what Nepal’s self-described expert class has been producing manually. This is not a criticism of AI,meaning it is a criticism of a discourse that has been so hollowed out that a language model can reproduce it without noticing the difference.
The policy consequences are disastrous. Think of the health insurance scheme that was designed with input from people who called themselves healthcare financing experts and produced a programme that now owes hospitals billions of rupees it cannot pay. The scheme’s insolvency — spending nearly double what it collects every year — was foreseeable from the basic arithmetic of premiums as well as benefit levels and enrolled population. Anyone who had done the maths would have seen it. The experts who endorsed the design either didn’t do the maths, or did it and said nothing, or produced reports that said what the ministry wanted to read.
Think of the cement price-fixing that cost ordinary Nepali households billions in inflated construction costs. The Competition Promotion Board has been told about this by economists for years. It has never successfully prosecuted a cement company. The experts who serve on the relevant advisory committees are, in many cases, the same people who advise the companies they are supposed to scrutinise.
Real expertise has a quality: it is falsifiable. A real forecast can be wrong, and when it is, the forecaster explains why. A real policy recommendation has defined success criteria, and when the policy fails against those criteria, the analyst updates their model. Nepal’s expert discourse fails this test almost universally. The same person who endorsed last year’s wrong growth forecast will endorse next year’s wrong growth forecast, and describe both endorsements as analysis. The AI tools now assisting in the production of this discourse fail the test even more completely, because a language model has no professional skin in the game and no reputation that can be damaged by being consistently wrong.
The true version of expertise is rare everywhere and rarer in places where expert status confers access to consulting contracts, advisory fees, television appearances and project funding. Nepal’s policy environment has built a closed loop: donors commission reports, consultants produce them, experts validate them, ministries implement them and the same consultants produce the evaluation that confirms the implementation worked. The loop requires participants who look like experts and produce content that looks like analysis. The arrival of AI makes the appearance cheaper to maintain. It does nothing about the substance that was never there. ■
