Why the domain fits
Farmers and agricultural systems continuously assimilate noisy data, such as soil moisture, imagery, and weather, and act through irrigation, fertilization, and pest control, a pattern the report frames directly through variational and expected free energy minimization. Expected free energy's split into epistemic value (e.g., pest scouting, sensor placement) and instrumental value (e.g., harvesting at predicted optimal times) maps naturally onto agricultural exploration-versus-exploitation trade-offs. Agriculture's nested scales, from hourly irrigation decisions to multi-year soil restoration and decadal climate adaptation, also suit active inference's hierarchical generative models, and the report extends this to biospheric Markov blankets partitioning farms into internal, sensory, and active states.
State of the literature
The report finds a spectrum of engagement with the term, from strict free-energy-principle work in sustainability theory (the 'Sustainability under Active Inference' article in Systems, 2024, and a related resource-management preprint) to information-theoretic uses in robotics and sensor networks. Applied examples include an EnhancedTinyCNN pest-detection framework whose energy budget is dominated by inference cycles, an ET0 edge-forecasting benchmarking study for irrigation, and wireless sensor network battery optimization via active inference and dynamic Gaussian Bayesian networks. The report is explicit that empirical, real-world evidence for performance gains remains sparse and that much of this literature is conceptual, cross-domain, or terminologically loose.
Open problems
The report identifies generative-model specification and validation against field data as a core technical gap, since most agronomic models are mechanistic rather than probabilistic and evidence comes mainly from simulations and small prototypes. Data ecosystems are fragmented across proprietary platforms, raising governance questions about data ownership and how sustainability priors are chosen and encoded. The report also flags interpretability for farmer trust, social/ethical risks of embedding normative priors, and the need for randomized, clinical-grade evaluation before active inference decision support can be considered proven in practice.
Reference Backbone
Karl J. Friston (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience. DOI: 10.1038/nrn2787. Maxwell J. D. Ramstead, Karl J. Friston, Axel Constant, Lancelot Da Costa, Casper Hesp, Beren Millidge, Alexander Tschantz (2023). On Bayesian Mechanics: A Physics of and by Beliefs. arXiv. Giovanni Pezzulo, Francesco Rigoli, Karl J. Friston (2017). Active Inference, homeostatic regulation and adaptive behavioural control. Progress in Neurobiology. DOI: 10.1016/j.pneurobio.2017.08.001. Daniel Ari Friedman, Alec Tschantz, Maxwell J. D. Ramstead, Karl J. Friston, Axel Constant (2021). An Active Inference Framework for Ant Colony Behavior. Frontiers in Behavioral Neuroscience. DOI: 10.3389/fnbeh.2021.647732.