From Predictive Power to Productive Practice: A Critical Synthesis of Artificial Intelligence Applications in Agricultural Systems

Agriculture faces mounting pressure from population growth, climate volatility, resource scarcity, labour shortages and stagnant farm profitability and artificial intelligence is widely promoted as a route toward more productive, efficient and resilient food systems. A large and rapidly expanding body of research reports impressive algorithmic performance across crop monitoring, disease detection, yield prediction, irrigation scheduling, weed management, robotics and livestock monitoring. However, the extent to which these algorithmic achievements translate into meaningful farm level and system level outcomes remains contested and, in many domains, poorly evidenced. The evidence indicates that classification and prediction models frequently achieve high accuracy under controlled or benchmark conditions, yet studies that validate these models across seasons, geographies and farm types remain comparatively scarce and rigorous farm level economic evaluations are even rarer. The literature further reveals persistent disagreement over whether artificial intelligence narrows or widens inequalities between large commercial operations and smallholder or resource constrained farmers. Data governance, algorithmic bias, limited explain ability and weak integration with agricultural extension systems constitute recurring and unresolved concerns. The review concludes that artificial intelligence has produced genuine advances in agronomic prediction and automation, but that the pathway from algorithmic accuracy to agricultural transformation is neither automatic nor uniform and that farmer centered design, longitudinal field validation and responsible data governance are essential preconditions for translating technical promise into equitable and sustainable agricultural outcomes.