A sophisticated artificial intelligence (AI) model has been developed to predict the risk of Highly Pathogenic Avian Influenza (HPAI) outbreaks, potentially reducing the need for widespread culling of poultry and cutting associated costs by up to 20%. This innovative system, a collaboration between the Animal and Plant Quarantine and Inspection Agency (APQIA) and data technology firm Big Value, utilizes approximately 60 key data points to forecast potential HPAI hotspots.
The initiative comes in response to the significant economic and agricultural impact of HPAI. The 2020-2021 winter season alone saw the culling of 30 million chickens due to HPAI, leading to substantial compensation payouts, increased prices for chicken and eggs, and ripple effects throughout the food processing and service industries. In the recent 2025-2026 winter period, 62 HPAI cases were reported domestically. Climate change is further exacerbating the problem, altering migratory patterns of wild birds, which are primary carriers of the virus. Outbreaks are no longer confined to coastal regions but are spreading inland, and the virus itself is evolving.
Developing an AI-Driven HPAI Risk Prediction System
Hong Sung-gil, former head of the APQIA’s Animal Disease Big Data Team, and Gu Reum, CEO of Big Value, spearheaded the development of this AI-based risk assessment system. Their work, detailed in a publication titled ‘National Intelligence,’ chronicles the challenges and breakthroughs in creating an AI system capable of transforming government operations, moving beyond simple automation to fundamentally change decision-making processes.
During an interview, Gu emphasized the critical need for AI-customized data generation and management. “South Korea ranks highly among OECD nations in terms of public data openness, with tens of thousands of datasets registered on data portals,” Gu explained. “However, the proportion of truly usable data is very low.” He further elaborated that only a fraction of this data is comprehensible and actionable for AI, highlighting a lack of interoperability between different government agencies, which prevents data from being linked and utilized effectively.
Hong stressed the importance of viewing AI not merely as a tool but as a collaborator. “We need to meticulously design the process structure, defining where AI’s responsibilities end and where human oversight and direct intervention begin,” he stated. Hong recently concluded his tenure at APQIA on the 30th of last month.
Why Focus on HPAI for Chickens?
When asked why HPAI risk prediction was prioritized over other animal diseases, Hong explained the unique characteristics of the egg industry. Unlike pork or beef, where production can be scaled up or down and exports are common, egg production is largely demand-driven. Countries typically produce only what they consume. If domestic production drops sharply due to factors like HPAI, the country must import eggs, a critical staple. “HPAI has become frequent in South Korea since the 2020s, causing significant difficulties in the domestic egg supply,” Hong noted. “This created a strong need for a system that could predict and respond to these risks rapidly.”
Project Objectives and Impact
The primary goal of the project was to move beyond reactive, manual-driven responses to HPAI. “Previously, responses to HPAI outbreaks followed a set manual,” Hong said. “We aimed to break that mold by creating a framework that alerts us to risks before they escalate, allowing for proactive preparation.”
The AI risk prediction model fundamentally alters culling strategies. Instead of culling solely based on proximity to an infected farm, the system simulates virus transmission pathways. “The model analyzes where the virus is heading after an infection is detected, identifies critical nodes in its potential path, and determines the optimal combination of interventions for maximum containment with minimal intervention,” Hong explained. This data-driven approach allows for more targeted and efficient responses.
Economic Benefits of Targeted Culling
Gu highlighted the economic advantages. “Farms are less likely to be culled simply because they are geographically close to an outbreak,” he stated. “This is positive from a national finance perspective, as it reduces unnecessary preventive culling and, consequently, compensation payouts.” Big Value’s analysis projects a potential reduction of over 20% in culling compensation costs.
Addressing Climate Change and Evolving Threats
Both experts acknowledged the increasing influence of climate change on disease patterns. Gu pointed out that HPAI is now appearing in regions previously considered safe havens for poultry in the U.S., indicating significant shifts. Hong added that climate change is particularly impactful in plant quarantine, where insect vectors carrying viruses are rapidly migrating across borders via air currents. “Insects that previously didn’t appear in Korea are now being introduced, posing a serious problem,” he warned.
The dynamic nature of these threats raises concerns about the relevance of historical data. Gu, however, believes that modeling techniques can overcome this. “By examining the causal relationships in models, we can delve into the root causes of variables,” he said. “For instance, studying migratory bird data, a root cause of HPAI, allows us to analyze the conditions affecting bird migration, which can then be incorporated as a variable in the model.”
Model Accuracy and Data Challenges
The developed HPAI risk model demonstrates a notable level of accuracy. When predicting the top 10% of farms most at risk, the model correctly identified 56.8% of actual outbreak farms. If the prediction scope is widened to the top 20% of farms, the accuracy increases to 75%. On a regional basis, identifying the top 20% of at-risk areas captures 95.5% of outbreaks. For comparison, a European Food Safety Authority (EFSA) model identifies 73% of outbreaks when selecting the top 33% of regions.
Despite these advancements, data acquisition remains a significant hurdle. Hong noted that incorporating data from live poultry trading in traditional markets could further enhance the model’s accuracy. However, the lack of a standardized data collection system across markets—some using manual logs, others Excel files, and some no data at all—creates inconsistencies. “We considered using this data, but the lack of sustainability and standardization meant we couldn’t rely on it as a core variable,” Hong explained. “If a key variable disappears, the entire model could become unstable.”
Ultimately, the team narrowed down the vast amount of potential data to approximately 60 key variables. These variables were categorized into four main axes: environmental factors, quarantine factors, transmission factors, and seasonal factors, derived from historical analysis of HPAI’s causal relationships.
The Challenge of Public AI (AX) Integration
Gu addressed the inherent limitations of Public AI (AX), where objectives can diverge based on which government department takes the lead. “While wild birds are subjects of control in disease prevention, they are also protected species from an environmental perspective,” Gu stated. “Restricting birds or destroying nests can be seen as ecological destruction.” He warned that if each department develops AI tailored to its specific goals, it could lead to conflicting decisions and a chaotic overall system.
To prevent this, Gu proposed that a top decision-maker must establish clear priorities. “Depending on the situation, the priority between disease control and ecological protection must be decided,” he said. “This is ultimately a political process.”
Tips for Successful National AI Implementation
Hong reiterated his earlier point: AI should be viewed as a partner, not just a tool. “We need to design the process structure, clearly defining the scope of AI’s role and where human expertise takes over,” he advised.
Exporting a National AI System
Gu suggested that once a national AI system is established using public funds, it could be exported to create new value. If South Korea’s system becomes a global standard for livestock disease prevention AI, it could significantly enhance the nation’s image. “Currently, there is no international standard for AI systems in livestock disease prevention,” Gu noted. “If Korea presents a proven system on the global stage, it could effectively set that standard.”
The book ‘National Intelligence’ focuses on the HPAI risk prediction model developed by Big Value and APQIA. It explores the conditions necessary for South Korea to advance from AI adoption to AI transformation (AX), using this case study. The book details the methodology of AX, which aims to change the fundamental decision-making and operational structures of government, moving beyond simple robotic process automation (RPA).
Insights from the research and experiences of Hong and Gu will be presented at the ‘Government AX Forum’ hosted by 조선비즈 on the 9th.
