Artificial Intelligence (AI) is being applied across various fields and advancing rapidly with tremendous speed and scale. Transcending the domains of academia and industry, AI-based technologies are growing amidst the waves of competition, and technologies like ChatGPT, Bard, and DALLE-2 are bringing significant changes to our lives. This article introduces the efforts of the Data Science Group at the Institute for Basic Science (IBS) Pioneering Research Center for Mathematical and Computational Sciences, which utilizes AI to tackle international issues such as the Sustainable Development Goals (SDGs).

The current situation of global issues like poverty

According to the United Nations report, over 700 million people worldwide are living in extreme poverty, surviving on less than $2 a day. However, measuring poverty and identifying areas in need of assistance is a challenging task. 53 countries worldwide have not conducted agricultural production surveys in the past 15 years, and in 17 countries, population censuses have not even been conducted. Such gaps in data make it difficult for countries to formulate data-driven policies and can create disparities in data between regions, leading to unfairness.

Innovation in data science

The utilization of artificial satellite imagery has emerged as an innovative solution to address these data gaps. Satellite imagery can be captured from anywhere on Earth, making it useful for estimating the economic situation in areas where agricultural and population surveys are difficult. Thanks to advancements in optical equipment, high-resolution imagery can identify objects as small as 30 cm and capture vast areas spanning hundreds of square kilometers. This allows for the identification of objects such as buildings, roads, and vehicles, and the acquisition of extensive data. Factors such as building density, the level of development of transportation networks, and the scale of agricultural land are closely related to the economic status of an area.

The beginning of satellite image analysis using AI

Artificial intelligence is employed in the analysis of such visual data. By providing AI with the "answer key," or economic indicators corresponding to the area along with the images, AI can be trained to extract meaningful features from satellite imagery's visual patterns. Since 2016, many research teams starting with Stanford University have begun combining AI with satellite imagery and publishing predictions of economic indicators. However, training the AI models fundamentally requires the "answer key," or economic indicator data corresponding to the images. This makes it difficult to apply the AI model to some underdeveloped countries where actual measured data is scarce.

Low-cost indicator prediction through human-AI interaction

The IBS research team has introduced a novel method for effectively training AI models while efficiently reducing data collection costs. This model features a collaborative structure between humans and machines without relying on traditional statistical data, which allows it to exhibit high versatility and can be extended to include even the least developed countries like North Korea. Initially, satellite images are grouped based on similar visual patterns (e.g., forests, paddy fields), and humans compare the degree of economic activity by examining representative images of each group and arranging them in order. AI then learns from this information regarding economic scores and assigns scores to each image. This approach requires less data collection per image compared to gathering actual measured information for each image, making it a highly efficient method.

[Figure 1] Economic scale prediction using nighttime light intensity images (top left: background image provided by NASA Earth Observatory). South Korea appears bright due to extensive electric lighting, while North Korea, except for Pyongyang, appears dim due to limited electricity supply. However, the new model developed in this study provides more refined economic predictions for North Korea (top right) and five other countries in Asia (bottom: background image from Google Earth).

The research team applied the developed model to North Korea and five other least-developed countries in Asia (Nepal, Laos, Myanmar, Bangladesh, and Cambodia) and investigated the economic conditions of these regions using the economic indicator scores obtained from satellite images (Figure 1). The scores obtained from the model showed a high correlation with conventional socioeconomic indicators such as population density, employment, and number of businesses, proving the universality of the model applicable to least-developed countries like North Korea. It is expected that this model will be valuable for generating data that can be utilized for policymaking in the international community in the future.

[Figure 2] Comparison of economic scores within North Korea using satellite images from 2016 and 2019. Significant development progress was observed in the tourist development area of Wonsan Kalma District (top), while there was little change in the industrial development area of Wiwon Development Zone (bottom). (Background image: Provided by the European Space Agency (ESA))

Applying the research team's model to satellite images from various years allows for the measurement of changes in economic indicators over time. For example, by comparing actual economic activities such as the detection of new buildings and the development of tourist zones, significant trends can be identified (refer to Figure 2). The image provided in the example illustrates the changes in economic scale between 2016 and 2019 in North Korea's tourist development area, Wonsan Kalma District, and industrial development area, Wiwon Development Zone. This result was achieved through international collaborative research involving teams from KAIST, Sogang University, Hong Kong University of Science and Technology (HKUST), and the National University of Singapore (NUS).

Analysis of international issues using indicators

This research, which integrates knowledge from computer science, economics, and geography, holds significant importance in addressing poverty on a global scale. By leveraging indicators based on AI, various international issues can be more effectively analyzed. For example, by measuring changes in indicator scores in photos before and after disasters, the extent of damage can be detected, and areas in need of assistance can be quickly identified and supported. It is expected that the developed artificial intelligence algorithms will be expanded to address various international issues such as carbon dioxide emissions and the impacts of climate change. We look forward to more researchers showing interest in the field of data science for the betterment of humanity!