When the Uncertainty Is Bigger Than the Shock: Scenario Modelling for English Local Elections

# When AI Models Know to Say "I Don't Know" A data scientist studied election prediction models and found that sometimes the most honest thing a model can do is admit when there's too much uncertainty to make a reliable forecast. Rather than forcing a prediction, the better approach is showing multiple possible scenarios and their likelihood—which actually helps decision-makers plan better than a single confident-sounding guess that could be completely wrong. This matters beyond elections: whenever you're relying on AI predictions, you want to know not just what it thinks will happen, but how confident it actually is.