Research is underway to predict the expected effects of policies using artificial intelligence (AI) and to find designs that enhance effectiveness within the same budget. The aim is to provide numerical evidence for decision-making based on policy conditions.
On November 11, OpenAI held a media session titled 'OpenAI Insights: Utilizing AI in Policy Research' at WeWork in Gangnam, Seoul, where Professor Kwon Hyuk-koo from Nanyang Technological University (NTU) presented his research on GPT-based policy simulations.
This study uses the domestic consumer recovery subsidy coupons as a case to verify whether AI-predicted household consumption responses can be utilized in policy design by comparing them with actual consumption records.
The research team employs the OpenAI API to simulate consumer responses using the GPT model. They provide a consumer profile summarizing income and spending habits along with policy conditions such as payment amounts, usage locations, and expiration dates. GPT interprets this information to predict changes in spending and coupon usage, generating responses on whether new purchases will occur or if existing spending will be replaced.
A key difference from previous studies is that GPT assigns a sort of character to consumers to analyze policy conditions. Professor Kwon explained, 'This has reduced the initial prediction error from about 25% to approximately 10%.'
He emphasized that the goal of this research is to provide the necessary evidence for policy decision-making. He stated, 'For example, we can predict how much consumption would increase if a 10 trillion won subsidy were distributed to all citizens versus focusing on low-income households, helping to prevent policy delays and facilitate discussions.'
However, he noted that final policy judgments cannot be left to AI. Professor Kwon remarked, 'We must discuss what effects are deemed important and what level of costs society can bear. The purpose of the research is to provide thorough and accurate evidence for decision-making.'
This research is one of 14 projects selected under OpenAI's 'Industry Policy Research Support Program.' OpenAI is providing a total of $1 million in research funding and up to $1 million in API credits for all projects. In South Korea, Yonsei University has been selected to conduct a project on 'Democratic Accountability Using AI.'
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.