Government Proposes AI-Driven Social Assistance Reform for Better Targeting Efficiency
The Indonesian government is planning a transition to direct cash transfers for social assistance, utilizing AI and digital IDs to improve accuracy. The move aims to address historical inefficiencies in beneficiary targeting, provided that data quality and privacy are protected.
Highlights
- •Government plans to overhaul social assistance into direct cash transfers totaling IDR 5.4 million per person.
- •The implementation of Artificial Intelligence and facial recognition aims to improve beneficiary targeting accuracy.
- •Current social assistance programs face significant challenges with both inclusion and exclusion errors.
- •Data quality, dynamic environmental factors, and strict privacy protection are critical for successful AI integration.
On June 9, 2026, Luhut Binsar Pandjaitan, the Chairman of the National Economic Council (DEN), announced that the government is preparing to transition social assistance (bansos) into a direct cash transfer system. This initiative proposes a potential total aid amount of IDR 5.4 million per person. This figure represents an aggregation of existing social welfare programs, such as the Non-Cash Food Assistance (BPNT) and the Family Hope Program (PKH).
A significant component of this proposed transformation is the integration of Artificial Intelligence (AI) to refine the verification process for beneficiaries. The government plans to utilize facial recognition technology alongside a Digital Single ID system to ensure that financial support reaches the intended recipients. By leveraging machine learning, authorities aim to categorize individuals into priority groups in near real-time, moving beyond traditional static methods.
Challenges in Implementing AI for Social Welfare
The core objective of utilizing AI for social assistance is to overcome historical challenges regarding accuracy. Current data suggests that previous programs have struggled with precision; for example, the Non-Cash Food Assistance program recorded a targeting accuracy rate of only 47.46%. This discrepancy highlights issues where deserving households are excluded, while those who do not meet the criteria remain included in the distribution.
The reliance on community-based targeting, while capable of identifying local needs, has often been distorted by non-technical factors, including social relations and political preferences. Implementing AI-driven solutions intends to mitigate these biases. However, the effectiveness of AI in social protection heavily depends on the quality and dynamism of the data used for training models.
Luhut Binsar Pandjaitan and the relevant authorities acknowledge that poverty is dynamic. Economic vulnerability can shift rapidly due to environmental conditions, such as climate change, which affects vulnerable groups like farmers and fishermen. Therefore, relying solely on static data is insufficient. Experts suggest that integrating diverse data sources—including climate information from the BMKG—is crucial for an adaptive social protection system.
Furthermore, while digitizing the beneficiary selection process promises greater efficiency, it introduces critical concerns regarding data privacy. Safeguarding individual information must be a primary requirement during the deployment of these digital identification systems. Balancing the precision of AI algorithms with robust data security protocols is essential to ensure that the goal of improving welfare distribution does not compromise the privacy of the citizens it aims to support.
