[Yang Myeong-gyun Column] Agricultural AI Should Be a Technology That Changes Perspectives Rather Than Just Producing Correct Answers.

[Expert Column] Yang Myeong-gyun, Department of Bio-Industrial Machinery Engineering, Jeonbuk National University

The Result of One Button and the Value of Agricultural AI
AI has already become a familiar technology. Creating sentences, analyzing images, and classifying data have all become much easier than before. We have entered an era in which pressing a single button produces a plausible result. This change is clearly positive. Greater accessibility to AI means that more people can use data and gain opportunities to view problems in new ways.

However, the value of agricultural AI does not end with producing quick results at the press of a button. Agriculture is a field where crops, varieties, environments, seasons, farm conditions, and work methods are intricately intertwined. Even with the same crop, responses differ depending on the cultivation period and environment, and even the same technology yields different results depending on the farm and facility conditions. Therefore, agricultural AI should not simply be a tool that generates answers, but a technology that reads agriculture more deeply and connects it to verifiable value.

AI That Changes Questions Rather Than AI That Produces Correct Answers
Until now, the important questions in agriculture have mostly been determined by people. Questions such as whether crops are growing well, whether pests and diseases have occurred, when harvest time is, and whether environmental conditions are appropriate. These questions are still important. However, in the AI era, there is a need to go one step further.
An important role of agricultural AI may be not only to answer more quickly the questions people already know, but also to discover questions that people had not yet thought to ask. The potential of agricultural AI lies not only in finding correct answers faster, but in changing the very questions and perspectives through which we view agriculture.

Technology That Reads Signals Humans Could Not See
Cases in which AI has posed new questions to existing perspectives have already appeared in various fields. The unfamiliar moves shown by AlphaGo posed new questions to the intuition humans had built up over a long time in the game of Go. AlphaFold created an opportunity to go beyond the limits of existing approaches in the long-standing scientific problem of protein structure prediction. These cases do not guarantee the success of agricultural AI as it is. However, they clearly show that AI can do more than simply imitate human judgment quickly; it can present possibilities that humans had not yet seen.

The same is true in agriculture. Features defined by humans, such as crop height, leaf area, disease symptoms, and cultivation recipes, are still important. However, as image, spectral, 3D, thermal, and environmental data accumulate repeatedly, AI can identify within the same data new features such as growth changes, stress signs, varietal response differences, and complex relationships between environment and growth that humans had not yet defined. The important value of agricultural AI does not end with calculating the features people already observe more quickly; it lies in opening new perspectives on crops.

Two Inertias That Slow Agricultural AI
From this perspective, there are two kinds of inertia to be cautious about when discussing agricultural AI. The first is the tendency to shut down new questions because of the difficulties of the field. Of course, agricultural sites are complex. Crops are not uniform, environments do not repeat themselves, and farmers’ conditions and judgments all differ. Therefore, we should be wary of the attitude that speaks as if AI were the answer to everything without understanding the field.

However, conversely, we must not push away attempts to apply AI or even the new perspectives AI offers simply because the field is difficult and complex. Deeply understanding the field should not be grounds for closing off future questions, but the starting point for creating better AI. The field should not be a reason to reject new technology, but a reason why new technology must become deeper.

The second is the tendency to chase the names of new technologies without sufficiently building the foundation. In national R&D as well, “new technology” is always an important criterion. However, in agricultural AI, novelty does not simply mean moving on to the next keyword or hastily changing direction. If previous technologies have not been sufficiently verified and the data, standards, and demonstration systems are weak, then even if the next technology is placed on top of them, changes in the field will inevitably remain limited.

Moreover, as AI technology changes rapidly, technologies once thought to have been developed in the past often need to be redefined, revalidated, and reconnected to fit the new AI environment. Agricultural AI is not a stepwise technology that is developed once and then moves on to the next technology; it is a technology that must continuously look back and renew itself on top of an accumulated foundation that includes even failures.

In the end, what slows agricultural AI is neither excessive expectations for new technology alone nor excessive caution toward new technology alone. The problem lies in speaking only of the field without an open perspective, or speaking only of novelty without building up the foundation. The field is the place where technology must be verified, and new technology should not be a justification for skipping the foundation, but an opportunity to rebuild that foundation anew.

Invisible Foundational Research Creates the Future of Agricultural AI
Then what is needed for the future of agricultural AI? One is an open mindset that accepts the unfamiliar signals presented by AI as new questions. Another is the invisible foundational research needed to turn those questions into value in the field. Data collection methods, sensor interpretation standards, phenotype definitions, modeling of crop-environment relationships, and long-term demonstration systems may not be clearly applicable right away and may not look like glamorous achievements.

However, without such foundations, new AI technologies will falter the moment they enter the field. Research whose immediate use is not clear is not necessarily distant research. Rather, in agricultural AI, processes that are not visible right now will later become the starting point that determines the success or failure of field application.

For agricultural AI to go beyond being a tool that produces results at the press of a button and become a technology that discovers features humans could not see and leads them to verified value, these foundations must be treated more seriously. When discussing agricultural AI, smart farming, physical AI, and agtech in the future, what matters is not quickly chasing the names of technologies. It is asking how that technology enables us to read agriculture in a new way, what data it leaves behind, and through what verification it returns as value to the field.

The future of agricultural AI begins not with faster correct answers, but with better questions. And when those questions pass through data and verification and lead to value in the field, AI can finally become not a buzzword, but the next foundational technology for Korean agriculture.