Tetlock Institute Forecasts 0.48% Catastrophic AI Risk by 2030

The Tetlock Institute's Forecasting Research Project has published an estimated 0.48% probability of an AI-related catastrophic incident by 2030. This assessment contrasts with internal warnings from some AI industry figures.

The project defines a catastrophic incident as a direct loss of ten million or more human lives. It also includes events causing severe, irreversible losses in human capabilities. These could encompass existential threats to human civilisation, or the permanent loss of human control over societal development. The definition deliberately excludes broader risks such as job displacement or societal inequality.

Forecasters involved in the project did not include scenarios where advanced AI capabilities are misused by state or non-state actors. Instead, their models focused on unaligned AI systems acting autonomously. The 0.48% probability reflects the aggregate median forecast from these experts.

This forecasting initiative aims to provide empirically grounded predictions. It applies methodologies developed for geopolitical and economic forecasting to AI risk assessment. The project’s lead, Dr. Tetlock, emphasises the value of structured expert judgement. He notes that such an approach can offer a more precise risk quantification. The findings represent a formalised, external perspective on AI safety timelines.

The institute intends this quantitative estimate to inform policy discussions. It provides a specific metric for potential high-impact, low-probability events. How will these quantitative forecasts shape ongoing debates about AI governance and deployment?