Overview
Project description submitted as part of application to Future of Life Institute - How to mitigate AI-driven power concentration Climate change intensifies agricultural challenges, requiring more and more advanced technological solutions. Small farmers increasingly rely on technical assistance, which is becoming centralized, dominated by large agricultural corporations and governments imposing sophisticated pre-designated solutions. As AI proliferates within these centralized solutions, diseases mitigation methods, climate credits, government subsidies, and regulations risk monopolizing farmers' activities. This tendency, amplified by AI development, threatens to undermine farmers' autonomy and limit their ability to make independent decisions, converting them into consumers of centralized technological solutions. We propose to develop FarmWorks — a platform for human-AI interaction in agriculture that enables personalized, farm-scale solutions while resisting power concentration associated with centralized AI systems. FarmWorks addresses the above challenges by providing an open source decentralized AI-powered agricultural platform that empowers individual farmers with cutting-edge technology while preserving their autonomy and promoting sustainable practices. By integrating real-time data collection, edge computing, and Active Inference models, FarmWorks enables farmers to make informed decisions tailored to their specific contexts (for example, integrating humidity data and epidemiological models to assist farmers with remediative and anticipatory treatments for mold).
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- Proposes FarmWorks, an open-source decentralized platform for human-AI interaction in agriculture intended to resist power concentration from centralized AI.
- Argues that Active Inference models in RxInfer.jl can be more computationally efficient and interpretable than deep or reinforcement learning for decentralized settings.
- Sets as a core design principle a coherent but flexible coupling between users' self-models and FarmWorks, and between users.
- Organizes the work into six work packages over a two-year timeline with stated evaluation metrics and risk mitigations.
Methods
- Planned on-site sensor network (Vin-Q/ROMI) over 4G, 5G and LoRa — WP1 proposes solar-powered sensors measuring soil and climate variables, with pilot installation in 20 Vin-Q fields in Catalonia.
- Farm-local data processing with optional sharing of model updates only — WP2 proposes per-farm instances where training data stays local and only model updates may be shared into a larger model.
- Active Inference generative models built with RxInfer.jl — WP4 plans variational/expected free energy models, including wine-grape epidemiological and intervention models, implemented in RxInfer.jl.
- Telegram-based virtual assistant optionally using local open-source LLMs — WP3 proposes a chat-bot interface to each farm's personalized knowledge database, with custom alerts and recommendations.
- QR/NFC embodied permission transfer for a data-sharing social network — WP5 proposes simple QR code or NFC exchanges to grant data access between farmers within existing networks of trust.
Summary sources: Paper metadata and evidence · Extracted source text.
Document downloads
PDFs are hosted here; supporting sources may use pinned archives.
- 2024_FarmWorks.pdf PDF · 0.64 MiB · extracted-text source
Citation
Citation metadata follows the unified bibliography.
Catalog details and resources
Related works
Ranked by recorded methods, topics, contributors and associated software, with contributor roles and domain-only matches labelled. Metadata overlap is not a citation or a scientific equivalence.