Try the policy before you pass it

Four researchers at Nanyang Technological University are building an AI tool that estimates how people would react to measures like cash handouts, ahead of any rollout.

Spend it, or stash it

The system is meant to predict whether people put extra money to work or save it, what they buy with it, and how a different policy design would shift the answer.

Fourteen picked from 400

OpenAI chose the project from more than 400 proposals submitted since April. The team, led by Professor Hyeokkoo Eric Kwon, gets US$100,000 and six months to build and test it.

A million real wallets

Training data comes from over a million users of a South Korean budgeting app, covering spending, retailers and monthly income from 2023 to 2025, scraped from bank alerts, texts and emails.

No profile of one

Names and account numbers are stripped, and amounts, income and purchase times are grouped into ranges. A profile only counts as anonymised if at least three, five or ten users share it.

“There should never be a situation where the data can only be matched to one profile.”

Agents playing households

Give every citizen S$100 in vouchers and the system simulates the result: how much extra spending it sets off, which shops see it, and what a rival policy design would have done instead.

People are bad witnesses

Standard economic simulations run on rules drawn from interviews and surveys. Kwon's point is that what people say they would do with free money rarely matches what they actually do.

“there is often a gap between how people actually behave and how they respond in surveys”

Graded against Korea

South Korea handed out cash in July and September 2025, with a November deadline to spend it at eligible shops. That gives the team a real before-and-after record to score predictions against.

Open to everyone by 2027

By February 2027 the team wants the tool open-source and usable on other populations, reading interests and political views too, and explaining how it reached each answer.