This paper examines how AI-driven language models (chatbots) contribute to global conservation injustices. The authors interviewed ChatGPT 30,000 times on ecological restoration, finding that its responses predominantly relied on expertise from male academics in the US, neglecting contributions from low- and lower-middle-income countries and Indigenous communities. The chatbot showed a bias towards tree planting and reforestation, overlooking holistic approaches and non-forest ecosystems. The study highlights how biases in AI knowledge production can reinforce Western science and calls for safeguard mechanisms to ensure just principles are incorporated into chatbot development.

Danilo Urzedo, Zarrin Tasnim Sworna, Andrew J. Hoskins, Cathy J. Robinson