Scholarlink

<p>Large Language Models (LLMs) are being explored as transformative tools for agricultural extension. Early studies suggest that LLM chatbots can democratize knowledge access by summarizing complex science and providing 24/7 multilingual advisories tailored to local contexts. For example, a pilot <i>Farmer.Chat</i> system has already served 15,000 smallholders with personalized Q&amp;A, demonstrating the potential reach of AI-driven extension. However, LLM output can inherit biases and omit traditional knowledge, sometimes yielding generic or impractical recommendations for smallholders. In Nigeria, an LLM advising on cassava weed control produced technically correct yet overly generic suggestions unsuitable for resource-poor farmers. If deployed without safeguards, LLMs risk worsening epistemic injustice and the digital divide by privileging wealthy, connected farmers over marginalized ones. We argue that under a framework of responsible innovation—emphasizing participatory design, farmer empowerment, and accountable AI—LLMs can become net benefits in AEAS. Hybrid human–AI advisory models, robust data governance (e.g., FAIR/CARE principles), and critical AI literacy are essential to ensure generative AI complements rather than undermines the trusted social foundations of rural extension.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sowing trust in the machine: ethical pathways for large language models in farmer-centered extension

  • 期刊论文

  • 发表日期：

    2025年12月11日

查看全文
  • Khondokar H. Kabir

摘要

Large Language Models (LLMs) are being explored as transformative tools for agricultural extension. Early studies suggest that LLM chatbots can democratize knowledge access by summarizing complex science and providing 24/7 multilingual advisories tailored to local contexts. For example, a pilot Farmer.Chat system has already served 15,000 smallholders with personalized Q&A, demonstrating the potential reach of AI-driven extension. However, LLM output can inherit biases and omit traditional knowledge, sometimes yielding generic or impractical recommendations for smallholders. In Nigeria, an LLM advising on cassava weed control produced technically correct yet overly generic suggestions unsuitable for resource-poor farmers. If deployed without safeguards, LLMs risk worsening epistemic injustice and the digital divide by privileging wealthy, connected farmers over marginalized ones. We argue that under a framework of responsible innovation—emphasizing participatory design, farmer empowerment, and accountable AI—LLMs can become net benefits in AEAS. Hybrid human–AI advisory models, robust data governance (e.g., FAIR/CARE principles), and critical AI literacy are essential to ensure generative AI complements rather than undermines the trusted social foundations of rural extension.