New AI models cannot predict crop yields across African borders

Researchers tested advanced AI models to predict maize yields across different African countries. It turns out these models cannot predict yields across national borders, performing no better than older methods.

Many assumed advanced AI models could easily adapt to new regions, especially for critical tasks like predicting crop yields for food security. This paper shows that even the newest AI models struggle to generalize across national borders in Africa, failing to predict yields in one country after being trained in another. This means that efforts to use AI for continent-wide food planning will need much more localized data and training than previously thought.