Paper models social cost of shared fertilizer-recommendation models
Original title: Algorithmic Monoculture in Machine Learning-based Site-Specific Fertilizer Recommendation
Under a 10% regret cap, diversification lowered modeled social loss by 10.89–41.53% in five Moroccan cereal systems.
The paper argues that correlated errors in machine-learning NPK fertilizer recommendations can add up when many farms adopt the same models. It links expected social loss to adoption shares, uses the Herfindahl-Hirschman Index for concentration, and extends the model to correlated model families. Nine model families across five Moroccan cereal systems calibrate the benchmark. Under illustrative parameters and a 10% regret cap, accuracy-constrained diversification reduced modeled social loss in all five systems; the authors say these are not agronomic or monetary welfare estimates.
Unconfirmed: Results are counterfactual model outcomes under illustrative parameters, not measured farm or monetary welfare.
- Modeled loss fell 10.89–41.53%, with a mean of 19.41%.
- Uniform allocation reduced modeled loss but breached the regret cap in every crop.
- Highly dependent model errors weaken the modeled benefit of diversification.
- The calibration uses nine model families and five Moroccan cereal systems.
- Tags
- fertilizer-recommendationalgorithmic-monoculturemodel-diversificationagricultureerror-correlation
- Countries
- Morocco
From the original article
1 min read at arxiv.orgMachine learning recommendations are often evaluated one model at a time, despite deployment across interconnected decision makers. For continuous NPK fertilizer prescriptions, correlated errors can aggregate across farms and generate systemic costs.
By Abdelghani Belgaid · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.