In Greek mythology, Cassandra was the daughter of Priam, King of Troy. Apollo granted her the gift of prophecy; when she spurned him, he cursed her so that her predictions — always accurate — would never be believed. She foretold the fall of Troy. She warned of the Trojan Horse. No one listened. The city burned. Cassandra was right, and it made no difference whatsoever. Economists will recognise the condition intimately.
Artificial intelligence is eating the professions. Lawyers, radiologists, and programmers — perhaps most ironically — find themselves confronting a technology that can do much of what they do, faster and cheaper. Anthropic has publicly acknowledged that AI now writes a substantial portion of its own code, a confession that sets the entire software engineering fraternity reaching for its actuarial tables. Yet among the professions, one stands curiously immune: not because its practitioners are unusually well organised, but because the discipline they inhabit is, in a profound sense, already cursed in the way Cassandra was cursed. The economist sees, speaks, and is disbelieved — until it is too late to matter. Understanding why AI cannot lift that curse illuminates something important not about artificial intelligence, but about economics itself.
I. The Wrong Way In — and Perhaps the Only Honest Way
I came to economics the wrong way, and perhaps that is the only honest way to come to it. In 1983, with a gold medal in electronics and communication engineering, I walked into the Indian Administrative Service securing All India Rank 2. I knew economics the way a traveller knows a foreign city from its airport — the Five Year Plans, the Union Budget headlines, the broad aggregates every educated Indian absorbs by osmosis. A few obligatory papers at the Lal Bahadur Shastri National Academy of Administration, Mussoorie, were cleared without ceremony, and then twelve years of field postings followed, culminating in four years as Deputy Commissioner, Amritsar, from 1992 to 1996 — a district still carrying the wounds of militancy, where the most consequential decisions had nothing to do with econometric models and everything to do with reading people, situations, and the grain of events.
In August 1996, immediately upon completing that tenure, a Government of India sponsored slot landed me in Manchester, at the University’s Institute for Development, Policy and Management, reading for a one-year Master’s in Development Administration and Management. My private alarm was considerable: could a man who had never formally studied economics sit in a seminar room alongside candidates who had read nothing else for three years, and survive? The question answered itself when my dissertation — on decentralisation and the politics of decentralisation in Punjab — was graded outstanding. But the grade never resolved the deeper puzzlement: if economics had established laws, why had twelve years of field administration equipped me nearly as well as three years of formal training had equipped my classmates? And if its principles were sound, why did its predictions so reliably go unheeded — or worse, prove correct only after the moment for action had passed? Cassandra’s curse, I began to suspect, was not a myth. It was a job description.
Field administration had taught me more economics than I realised — because economics, at its core, is less about models than about the grain of human behaviour under constraint.
II. The Seduction of Numbers
At first glance, economics looks tailor-made for artificial intelligence. GDP figures, consumer price indices, unemployment rates, yield curves, money supply aggregates, retail sales, housing starts — all arrive with clockwork regularity, digitally formatted and historically archived. Interest rates are known. Fiscal stances are announced. AI performs creditably on bounded tasks: computing correlations, backtesting strategies, summarising central bank minutes within seconds of release. But these are the easier parts of the economic enterprise. The harder parts — forecasting turning points, reading the drift of policy intention, anticipating how households will actually behave rather than how a textbook model predicts they will — have resisted AI with a stubbornness that demands explanation. Cassandra’s gift was never the data. It was the insight behind it.
III. A Science That Watches Itself
Economics, unlike physics or chemistry, studies a system that is aware of being studied — and that awareness changes the system itself. When the Reserve Bank of India signals its inflation-fighting intent, importers alter hedging behaviour, bond traders reprice the yield curve, manufacturers revisit investment plans. The act of communication is itself an intervention. Karl Popper would have had reservations about economics claiming the status of a natural science: its theories shelter behind ceteris paribus clauses and data revisions whenever predictions fail. The Austrian school and the Keynesian school can look at the same recession and reach opposite conclusions — and both cite data in support. In physics, competing theories are adjudicated by experiment. In macroeconomics, controlled experiments are essentially impossible. This is not a flaw to be fixed by more data or better algorithms. It is structural — and it is, in its way, another facet of the curse.
George Soros gave this phenomenon a name: reflexivity. The beliefs of economic actors and the reality those actors inhabit co-create each other in continuous feedback. A bank run is the elementary demonstration — if enough depositors believe a bank is insolvent, it becomes insolvent, regardless of yesterday’s balance sheet. The belief creates the fact. AI systems trained on historical outcomes cannot learn to anticipate the tipping point at which belief crosses a threshold and becomes self-fulfilling. That tipping point is a social and psychological event, not a statistical pattern. Cassandra saw it coming. The model never will.
IV. The Iceberg and Its Submerged Mass
AI learns from data, and the data economics provides is imperfect in ways that matter. GDP figures are revised for years after initial release. Inflation indices are constructed from basket choices that are themselves political decisions. But the deeper problem is that the most important economic variables are not measured at all. Animal spirits — Keynes’s phrase for the swings of optimism and pessimism that drive investment — appear in no official statistical release. The confidence of small-business owners, the degree to which households feel secure in their employment, the trust between trading partners: these carry enormous economic weight and are captured, at best, approximately. An AI trained on published data is trained on the visible tip of the iceberg. The submerged mass — expectations, anxieties, the social norms governing behaviour — is precisely where the action is. It is also, not coincidentally, precisely what Cassandra could see and her contemporaries could not.
V. Judgement, Politics, and What Models Cannot See
The public image of the economist — forecasting growth to two decimal places with an air of precision — obscures what the profession actually requires. Raghuram Rajan’s celebrated warning about financial-sector risks at Jackson Hole in 2005 — largely dismissed at the time, subsequently vindicated — was not a product of superior data or better regressions. It was a product of seeing something the models were constitutionally unable to see. Rajan was, in that moment, the economist as Cassandra: accurate, present, and comprehensively ignored. The economist’s real weapon is judgement — a form of knowing that emerges from living through crises, reading history, and developing an intuition for what is institutionally possible.
Economics, moreover, does not happen in a laboratory. Consider India’s GST reform, implemented in 2017 after more than a decade of political negotiation. The economic case had been settled for years. The question was never economic — it was political: what compensation to revenue-losing states, which rate structure would hold a coalition, how to manage the transition without triggering distributive conflict. These required an understanding of federal politics and state-level electoral arithmetic that no volume of economic data could supply. The interaction between technical analysis and political possibility is a domain of human complexity irreducible to pattern recognition — and it is a domain in which even the most gifted Cassandra must learn to hold her tongue until the moment is right.
Economics is less a predictive science than a language for thinking clearly about choices, trade-offs, and consequences. Cassandra had the language. What she lacked was an audience.
VI. The Curse Passes to the Machine
The social sciences aspire to rigour — and in its methods, economics has achieved it impressively. The econometric toolkit is formidable, the journals demanding. But apparatus is not the same as conclusion. Strip away the Greek symbols and the confidence intervals, and the conclusions of economics — on growth, inequality, the effect of minimum wages, whether austerity works — remain contested and frequently reversed. Physics builds on Newton; it does not periodically discard him. Economics buries its Keynes, exhumes him, buries him again. The rigour is in the scaffolding. The building it supports shifts with every political wind and every new dataset. Cassandra’s curse was not that she lacked method. It was that method, however impeccable, could not compel belief.
And then there is the most confounding phenomenon of all: the self-fulfilling prophecy of the anointed voice. When a Nobel laureate speaks, when the IMF or World Bank issues a forecast, when a fund manager commanding a trillion dollars in assets under management pronounces on where markets are headed, the market moves — not necessarily because the analysis is correct, but because it is believed. The prophecy manufactures its own proof. A Summers or a Stiglitz warns of recession; institutional investors reposition; credit tightens; consumer confidence slides; the recession arrives. Was the prediction right because the reasoning was sound — or because the predictor was credible enough to make it so? Cause and causality fold back upon each other until they are indistinguishable. No AI model can be accorded that authority. Markets do not move because a machine said so. They move because the right human said so, at the right moment, with the right institutional weight. That is not intelligence. That is power — and power, for now, remains stubbornly biological.
I came to Manchester without credentials, survived on instinct and field experience, and left with a dissertation graded outstanding. The lesson I drew then, and hold to now, is that economics rewards judgement over technique. As for the profession’s predictive record: economists can explain, with breathtaking lucidity, exactly why the 2008 crisis happened, why the 2022 inflation surge was inevitable, why every bubble could only ever have ended one way. The explanations arrive persuasively — approximately eighteen months after the event. The stock market analyst who tells you with great authority why the Sensex fell this morning is the same analyst who told you yesterday it would rise. AI, trained on this glorious, self-contradictory body of knowledge, has absorbed not the wisdom of economics but its curse — the magnificent Cassandran capacity to be right in retrospect and wrong when it matters. Economics is safe from AI not because it is too difficult to master, but because it has never quite mastered itself. Cassandra is still standing at the gates of Troy. The city, as always, is not listening.