Existential Risk Forecasting: Superforecasters and Domain Experts Show Parity in Near-Term Predictions

In a comprehensive study released on September 2, 2025, researchers from the Forecasting Research Institute (FRI) evaluated the accuracy of near-term forecasts made by superforecasters and domain experts during the Existential Risk Persuasion Tournament (XPT). The XPT, held in June–October 2022, brought together 169 participants to predict potential global risks, including those from artificial intelligence (AI), climate change, nuclear war, and pandemics. This report provides a detailed analysis of the forecasting performance on 38 near-term questions that resolved by mid-2025.

Why it Matters

The accuracy of forecasts in areas like AI development and climate technology is crucial for policymakers and investors who need to make informed decisions about resource allocation and risk management. The XPT aimed to assess whether superforecasters, known for their proven track records, could outperform domain experts with specific subject-matter knowledge. The findings have significant implications for how we approach long-term existential risks and the value of different forecasting methodologies.

Key Findings

1. Performance Parity: Both superforecasters and domain experts demonstrated similar levels of accuracy in predicting near-term events. There was no statistically significant difference between the two groups, indicating that subject-matter expertise is as valuable as general forecasting skills in this context.
2. AI Progress Underestimation: Both groups underestimated the pace of AI advancements, highlighting a potential blind spot in current risk assessments. This underestimation could have serious implications for policy and investment decisions related to AI governance and safety.
3. Climate Technology Overestimation: Conversely, both superforecasters and domain experts overestimated the progress in climate technology. This suggests a need for more realistic expectations and possibly a reassessment of strategies for addressing climate change.
4. Outperformance of General Public: Both groups significantly outperformed a baseline of educated members of the general public, reinforcing the value of specialized expertise and forecasting skills.
5. Individual vs. Aggregate Performance: At the individual level, median superforecasters and domain experts performed similarly to simple extrapolation algorithms. However, at the aggregate level, both groups showed improved accuracy and some evidence of outperforming these algorithms.
6. No Correlation with Long-Term Forecasts: There was no statistically significant correlation between near-term forecasting accuracy and long-term existential risk forecasts. This suggests that different skills and methodologies may be required for short-term versus long-term predictions.

Key Risks

1. Blind Spots in AI Forecasting: The underestimation of AI progress could lead to inadequate preparation for the potential risks associated with advanced AI systems. Policymakers and stakeholders must remain vigilant and continuously update their risk assessments.
2. Overoptimistic Climate Projections: Overestimating climate technology advancements may result in complacency or misallocation of resources. Realistic timelines and achievable goals are essential for effective climate action.
3. Reliance on Aggregate Data: While aggregate data from both groups showed improved accuracy, individual forecasts were less reliable. This highlights the importance of using ensemble methods and diverse perspectives to enhance forecasting accuracy.

The Opportunity

1. Enhanced Decision-Making: The findings underscore the value of combining superforecasting skills with domain expertise. Policymakers and investors can leverage this approach to make more informed decisions about global risks.
2. Improved Forecasting Models: The study's insights can inform the development of more sophisticated forecasting models that integrate both general and specialized knowledge.
3. Public Awareness and Engagement: By demonstrating the importance of accurate forecasting, the XPT can help raise public awareness and engagement in critical issues like AI safety and climate change.

This research was supported by the Musk Foundation, Open Philanthropy, and the Long-Term Future Fund. The authors also acknowledge the assistance of Sam Glover, Rory Svarc, and Bridget Williams throughout the project.