Issue Models trained on the past get worse as the climate stops resembling it

Summary

This issue identifies a limitation of AI that worsens as the climate problem deepens. AI models learn from historical data, so they struggle to anticipate events that have rarely happened before. As climate change accelerates, unusual weather is likely to become more common, pushing forecasts into conditions the models were never trained on. AI methods also rely on past data, are less transparent and handle probability less well, making their errors harder to audit. It may also be difficult to gather enough data for AI to understand complex ecosystems. The concern is that AI could be least reliable precisely when extreme and unfamiliar events matter most.