PAPER 3 — BOTH MODES IMPLEMENTED

From Reactive to Proactive: Anticipatory AI for Agricultural Sensor Networks

Frontiers in Plant ScienceAI in AgricultureSmart Agricultural Technology

Research Question

Does a proactive AI system — one that monitors sensor data and pushes alerts autonomously — improve farmer response time and crop outcomes compared to a reactive system where the farmer must ask?

Gap in Literature

Agricultural AI systems are overwhelmingly reactive: the farmer asks a question, the system answers. But farming problems don't wait for questions — soil moisture drops overnight, temperatures spike during storage root initiation, pH drifts below threshold over days.

The concept of "proactive AI" exists in HCI and conversational AI literature (arXiv:2410.12361, arXiv:2404.12670) but has never been formally evaluated in agriculture. No prior work compares reactive vs. proactive modes in a deployed agricultural system with real sensor data.

What's Already Built

Reactive mode (AgriRegion): Farmer asks → system retrieves → AI responds
Proactive mode (Farmerly AI): Sensor data triggers → rule engine evaluates → AI pushes alert
✅ Proactive rule engine with 9 condition-based rules (moisture, temp, pH, growth stage, drought)
✅ Priority classification (critical / high / medium / info)
✅ Floating tooltip delivery on web dashboard
✅ Both modes running on same platform — direct comparison possible

The Core Comparison

DimensionReactive (AgriRegion)Proactive (Farmerly AI)
TriggerFarmer asks a questionSensor threshold crossed
LatencyDepends on farmer awarenessImmediate (within sensor polling interval)
Knowledge sourceRAG retrieval from corpusRule engine + RAG context
DeliveryChat responseFloating tooltip on map / push notification
Farmer effortMust know what to askZero — system initiates
Risk of alert fatigueNone (no alerts)Possible — needs priority filtering
CoverageOnly what farmer asks aboutAll monitored conditions

Proposed Methodology

1. Simulation Study

Run 120-day sweet potato growing season simulation. Inject drought events, temperature spikes, pH drift. Measure: time-to-detection (how long until the farmer gets relevant information) in reactive vs. proactive mode.

2. Pilot Comparison

Split pilot farmers into two groups: reactive-only (must ask) vs. proactive (receives alerts). Measure: response time to critical conditions, number of missed events, farmer satisfaction.

3. Alert Fatigue Analysis

Track dismiss rates, insight relevance over time, priority distribution. Show that priority filtering (critical > high > medium > info) reduces fatigue while maintaining coverage.

4. Hybrid Mode Evaluation

Test a combined mode: proactive alerts for critical/high priority + reactive chat for everything else. Hypothesis: hybrid outperforms both pure modes.

Expected Results

What's Needed to Complete

Why This Paper Matters

This is the paper that connects AgriRegion to Farmerly AI — showing the evolution from reactive to proactive. It positions the entire research program: Paper 1 (AgriRegion) is the foundation, Paper 2 (deployment) is the engineering, and Paper 3 (this one) is the scientific contribution about how AI should interact with farmers.

It also directly addresses the "User Interaction Layer" future work item from the AgriRegion paper — showing you're executing on the roadmap.

Why This Venue

Frontiers in Plant Science (IF: 5.6) — publishes AI for agriculture, has a "Digital Agriculture" section. They love papers that show real impact on farming practice. The reactive-vs-proactive framing is novel enough for this venue. AI in Agriculture (IF: 8.2) is the backup — newer journal, hungry for novel AI paradigms applied to agriculture.