Weathering the Artificial Intelligence revolution: public attitudes to potential adoption of Machine Learning-based forecasts.

Abstract We present the results from a survey of over 6,000 members of the UK public to understand how they view the potential accuracy of Machine Learning Weather Prediction (MLWP) forecasts, how confident they would feel using this information to make decisions, and how these views compare to their opinions of physics-based Numerical Weather Prediction (NWP) forecasts. Most respondents (∼78%) perceive current NWP forecasts as accurate, with only a very small proportion (∼1%) indicating that they “Don’t know” the accuracy of these forecasts. This perception is consistent across age, gender, social grade, and geographic regions across the UK. In contrast, there is much less agreement across groups about the potential accuracy of MLWP forecasts: only around 47% of respondents believe these forecasts would be accurate, with roughly a fifth of respondents indicating they “Don’t know” what the accuracy of ML-based forecasts would be. These findings highlight a contrast between current perceptions of NWP, shaped through direct experience, and current perceptions of MLWP, shaped largely by prior beliefs. However, the high agreement for perceived accuracy of physics-based NWP forecasts suggests that users may be willing to update their beliefs about MLWP capabilities, and that direct experience of using the information is likely to be important. The diverse perceptions across demographic, socioeconomic, and regional groups, together with emerging research on AI’s role in weather forecasting and risk communication, underscore the need for broader public engagement. This engagement should not only identify potential concerns, but also inform user-centred solutions that can be embedded within forecast systems to strengthen trust and confidence.