What Could AI Do to the Isle of Man Economy?
Three scenarios for the economic consequences of AI between 2026 and 2030, adapted from the Anthropic Institute's “Economic Scenarios for Transformative AI” framework and calibrated with Isle of Man census, employment, and AI exposure data.
Anton Korinek, Charles I. Jones, Szymon Sacher, Tess Cotter, and Peter McCrory (September 2026). The Anthropic Institute, Working Paper No. 2026-02.
Why the Isle of Man is Different
The original paper models the US economy. The Isle of Man is structurally very different, and those differences amplify some effects of AI while dampening others.
Key structural differences
- 68% cognitive workforce — the IoM has a larger share of workers in occupations directly exposed to AI than the US (62.4%). Financial services, eGaming, and professional services dominate.
- Near-zero unemployment (0.6%) — displaced workers face a very different labor market. With 7.4 vacancies per unemployed person, reabsorption may be faster — but the economy has never had to absorb large-scale occupational transitions.
- Stagnant baseline GDP — real GDP has been roughly flat over 2016–2024 (vs US ~2% growth). AI could be restorative rather than merely additive.
- Small open economy — capital is more mobile (banking assets ~7x GDP), which means the return to capital rises less and more of the productivity gains flow to wages. We set the capital supply elasticity at 6 vs the paper's 3.
- Population: 84,000 — with a flat labor force and no net migration growth, the economy cannot grow through headcount. Every productivity gain from AI is amplified in per-capita terms.
The Three Scenarios
Following the paper, we model three scenarios — modest, substantial, and extreme change — that differ in how quickly AI capability and adoption advance, and how disruptive each increment of AI is for workers. These are not predictions; they trace what would follow from different assumptions about the pace and nature of AI.
Modest Change
Substantial Change
Extreme Change
Projections 2025–2030
GDP Above No-AI Path
Total Factor Productivity Above No-AI Path
Share of Tasks Performed with AI
Capital Stock Above No-AI Path
IoM Workforce AI Exposure
Our AI exposure assessment covers 100 SOC codes, 37,686 census workers across the IoM economy, using Frey-Osborne automation probabilities and Felten-Raj-Seamans AI occupational exposure indices, adapted for the island's occupation mix.
See the Workforce Resilience Index and Census AI Explorer for detailed occupation-level analysis.
Calibration: IoM vs US
The model parameters are adapted from the paper's Table 1, recalibrated for the Isle of Man using census data, World Bank indicators, and government statistics.
| Parameter | IoM | US (Paper) | Note |
|---|---|---|---|
| Labor share of income (sL) | 55% | 60% | Adjusted from nominal 35% |
| Cognitive workforce share | 68% | 62.4% | 2021 Census, SOC groups 1–4, 7 |
| Capital supply elasticity (ε) | 6 | 3 | Small open economy, high capital mobility |
| Baseline GDP growth (g + n) | 0.5% | 2% | IoM real GDP has stagnated since 2016 |
| Normal unemployment (Ū) | 0.6% | 3.8% | 260 unemployed / ~44,000 workforce |
| Services share of GVA | 95% | ~77% | World Bank, 2022 |
| Population | 84,160 | ~330M | Flat, no net growth |
Methodology
The model is a task-based framework in the tradition of Acemoglu and Restrepo (2018). Output is a CES aggregate over task instances, each performed by labor or capital. AI automates or augments a growing fraction of cognitive tasks, raising productivity but displacing workers who must search for jobs in other occupations.
An AI scenario is defined by five paths: the affected mass (what fraction of tasks AI can do), diffusion (how widely it's actually used), the gain per task (productivity improvement), the automation share (what fraction is full automation vs augmentation), and reinstatement (rate of new task creation). The affected mass and diffusion follow logistic curves; the gain grows linearly.
We use the paper's first-order approximation equations (Eq. 11) to compute the rental rate gap, wage gap, labor share change, and output per worker gap at each time step. Unemployment is estimated from the occupational reallocation required and the search friction parameters.
The key adaptation for the IoM is the higher cognitive workforce share (68% vs 62.4%), the higher capital supply elasticity (6 vs 3, reflecting a small open economy), and the much lower baseline unemployment and GDP growth. These produce scenarios where AI's GDP impact is amplified (more tasks can be affected) but unemployment rises less (tight labor market absorbs displaced workers faster, and higher capital elasticity means more gains flow to wages).
Data Gaps and Assumptions
These projections are illustrative, not predictive. Several important gaps affect the IoM calibration:
- Labor share (55%) — The nominal IoM labor share is ~35%, heavily distorted by corporate domicile profit booking. We adjust to 55% as a rough estimate of the domestic production labor share, but this is uncertain. The IoM National Income Report would provide a better figure but is not yet digitised in structured form.
- Capital supply elasticity (6) — No direct IoM measurement exists. We assume higher than the US because the IoM is a small open economy with a large banking sector, meaning capital can flow in more easily. This assumption is consequential: a lower value would mean more of the gains go to capital owners and less to wages.
- Job flow rates — The IoM does not publish quit rates, job finding rates, or occupational switching matrices. We proxy from the vacancy-to-unemployment ratio (7.4:1, extremely tight) and use the paper's US-calibrated matching function parameters.
- AI diffusion anchor — We set initial AI diffusion at 12% (slightly above the US 10%) given the IoM's eGaming, fintech, and professional services concentration, but no IoM-specific AI adoption survey exists.
- Cognitive/non-cognitive mapping — We map UK SOC 2020 major groups 1–4 and 7 to “cognitive” occupations. The paper uses US SOC groups 11–29, 41, 43. The mappings are comparable but not identical.
- Small-economy dynamics — The model assumes competitive factor markets. In a small economy with concentrated employers (the IoM has high employer concentration in some sectors), wage-setting may behave differently.
References
Korinek, A., Jones, C. I., Sacher, S., Cotter, T., & McCrory, P. (2026). “Economic Scenarios for Transformative AI.” The Anthropic Institute Working Paper No. 2026-02. anthropic.com/institute/econ-scenarios
Acemoglu, D. & Restrepo, P. (2018). “The Race Between Man and Machine.” American Economic Review, 108(6), 1488–1542.
Acemoglu, D. (2025). “The Simple Macroeconomics of AI.” NBER Working Paper No. 32487.
Manning, A. & Aguirre, J. (2026). “Adaptive Capacity Framework.” NBER Working Paper w34705. (IoM adaptation in Workforce Resilience Index)
IoM Census 2021, Isle of Man Government. World Bank Development Indicators (IMN). IoM Treasury Economic Briefing, April 2026. AI Job Guides: SmartIsland AI exposure assessment (100 SOC codes, 37,686 census workers).