A new paper co-authored by Dr Fanqi Zeng uses AI modelling to predict patterns of societal instability in countries around the world.
It finds that societies may face an increasing inability to cope with the combination of fiscal constraints that limit state revenue in the face of rising social spending.
Published in Futures, the study focuses on the United States, Sweden, India, and China, using data on population, government expenditures and revenue.
In particular, it uses data from the World Bank to show the working-age population (ages 15-64) as a percentage of the total population for the period between 1960 and 2050.
Using AI tools, the paper finds that the size of the working-age population is expected to fall in all four countries, with the most dramatic decline occurring in China, predicted to drop in size by 15%.
This trend presents challenges for societies in determining how to support a growing non-working population.
To care for an ageing population, governments usually need to spend more on pensions and healthcare. However, with fewer workers, tax revenues decline, creating budget deficits that force government borrowing.
Over time, persistent deficits drive up debt levels, raise interest payments, and lead investors to demand higher interest rates.
In the most likely scenario, this shrinking working-age population would lead to a spending crisis in China and to social tensions in the US, Sweden and India.
This prediction contrasts with previous theories that suggest a pattern of political crises in the US and beyond, driven by elite overproduction – a condition where scarce resources relative to the growing elite class block some groups' paths to wealth, fostering the rise of counter-elites.
The study is one of the first to extend predictions for societal instability beyond wealthy Western countries.
Dr Fanqi Zeng concludes:
AI tools show great potential for enhancing social science forecasting, but many 'black-box' models lack transparency in their decision-making.
To improve accuracy, we need more transparent and reproducible models, and we must refine prompts, compare multiple models, and integrate iterative learning to enhance the precision and utility of analyses.
