Why the shift from IoT to AIoT matters for food security in low-income countries

Global agriculture is under pressure due to climate change, rapid population growth and other factors. While Internet of Things (IoT) technologies help farmers monitor crops and soil, their impact has remained limited in many low-income settings. A new analysis published in the journal Future Internet argues that real transformation will come only with the integration of artificial intelligence directly into connected agricultural systems.

Published as From IoT to AIoT: Evolving Agricultural Systems Through Intelligent Connectivity in Low-Income Countries, the study traces how agricultural systems are shifting from basic IoT deployments toward Artificial Intelligence of Things (AIoT), and what this transition means for productivity, sustainability, and equity in resource-constrained regions.

From data collection to intelligent decision-making in agriculture

Over the past decade, IoT technologies have enabled farmers to collect real-time data on soil moisture, temperature, humidity, crop health, and livestock conditions. These systems have improved visibility across agricultural operations, but the authors argue that data collection alone does not translate into actionable insight, particularly in environments where technical support and continuous connectivity are unreliable.

AIoT represents a structural upgrade rather than a simple extension of IoT. By embedding artificial intelligence directly into connected systems, AIoT enables devices to analyze data, learn from patterns, and make decisions locally or collaboratively. This shift reduces dependence on centralized cloud infrastructure and allows agricultural systems to operate with greater autonomy, a critical advantage in rural and low-income contexts where bandwidth, energy supply, and technical maintenance are constrained.

Across the reviewed literature, the study identifies key application areas where AIoT has demonstrated measurable impact. In crop management, AIoT systems support early detection of pests, diseases, and nutrient deficiencies, enabling targeted interventions before yield losses occur. Smart irrigation systems equipped with edge-based intelligence adjust water delivery dynamically based on soil conditions, weather forecasts, and crop growth stages, reducing water use while maintaining productivity.

Livestock management emerges as another major application, with AIoT-enabled sensors monitoring animal health, movement, and feeding patterns. These systems allow early detection of illness and stress, improving animal welfare and reducing economic losses. In post-harvest management, AIoT supports storage monitoring and spoilage prevention, addressing a major source of food loss in low-income regions.

The authors note that reported yield improvements in AIoT pilot projects typically range between 10 and 25 percent, alongside reductions in water and agrochemical use. However, they caution that these gains are not uniform and depend heavily on system design, local adaptation, and farmer engagement.

Structural barriers limit AIoT adoption in low-income countries

While the potential of AIoT is significant, the study emphasizes that adoption in low-income countries faces persistent and interconnected barriers. The authors organize these challenges around three core dimensions: data, features, and models.

Data scarcity and fragmentation are identified as the most fundamental constraint. Many AI models require large, high-quality datasets to perform reliably, yet agricultural data in low-income settings are often sparse, inconsistent, and poorly standardized. Smallholder farms operate in highly heterogeneous environments, where soil conditions, microclimates, and farming practices vary widely even within small geographic areas. This variability limits the transferability of models trained in one context to another.

Feature engineering presents another challenge. Extracting meaningful signals from noisy, incomplete data requires domain knowledge and computational resources that are often lacking. The study highlights that many AI models developed for agriculture in high-income countries rely on features derived from advanced sensors or satellite imagery that may not be available or affordable in low-income regions.

Model deployment and maintenance represent the third barrier. Cloud-dependent AI systems are vulnerable to connectivity disruptions and ongoing operational costs. In contrast, AIoT systems must operate under tight energy budgets, limited processing power, and minimal technical support. These constraints demand lightweight models optimized for efficiency rather than raw predictive performance.

The study highlights socio-economic and institutional barriers too. Limited digital literacy among farmers, lack of training and extension services, and unclear data ownership frameworks reduce trust and uptake. Financing models often favor large commercial farms, leaving smallholders excluded from innovation pipelines. Policy environments in many low-income countries remain underdeveloped, with insufficient standards for data governance, interoperability, and long-term sustainability.

The authors argue that these challenges explain why many agricultural AI solutions fail to scale beyond pilot stages in low-income settings. Technology designed without contextual awareness risks becoming unusable or economically unviable once external funding or technical support ends.

A lightweight, context-aware architecture for inclusive AIoT

To address these limitations, the study proposes a modular, lightweight AIoT reference architecture specifically tailored to low-income agricultural environments. The architecture prioritizes simplicity, adaptability, and local autonomy over centralized complexity.

At the sensing layer, the framework emphasizes low-cost, low-power sensors capable of collecting essential environmental and crop data. Long-range, low-bandwidth communication technologies are favored to ensure connectivity across dispersed rural areas. Rather than relying on continuous cloud access, the architecture incorporates edge intelligence, allowing devices to process data locally and make decisions in near real time.

The study highlights the role of TinyML and edge-based machine learning, which enable models to run on constrained hardware with minimal energy consumption. These models support core tasks such as anomaly detection, classification, and basic prediction without requiring constant retraining or high computational overhead.

Decision-support mechanisms are designed to be interpretable and actionable for farmers, delivering recommendations that align with local practices and constraints. The authors stress that usability and trust are as important as technical accuracy. Systems that provide opaque or overly complex outputs are unlikely to be adopted, regardless of their theoretical performance.

Importantly, the proposed architecture embeds governance and capacity-building components alongside technical layers. Data ownership and stewardship are framed as critical issues, with farmers retaining control over their data and participating in decisions about how it is used. Training and extension services are identified as essential for long-term success, ensuring that farmers understand both the benefits and limitations of AIoT systems.

The study also calls for inclusive financing models and public-private partnerships that reduce upfront costs and spread risk. Without supportive economic structures, even well-designed AIoT systems may remain inaccessible to those who could benefit most.