AIA Forecaster: Expert-Level LLM-Based Judgmental Forecasting
paperAuthors
Credibility Rating
Good quality. Reputable source with community review or editorial standards, but less rigorous than peer-reviewed venues.
Rating inherited from publication venue: arXiv
Relevant to AI safety discussions around increasingly capable AI systems matching or exceeding human expert judgment in forecasting tasks, with implications for AI-assisted decision-making and risk assessment.
Paper Details
Metadata
Abstract
This technical report describes the AIA Forecaster, a Large Language Model (LLM)-based system for judgmental forecasting using unstructured data. The AIA Forecaster approach combines three core elements: agentic search over high-quality news sources, a supervisor agent that reconciles disparate forecasts for the same event, and a set of statistical calibration techniques to counter behavioral biases in large language models. On the ForecastBench benchmark (Karger et al., 2024), the AIA Forecaster achieves performance equal to human superforecasters, surpassing prior LLM baselines. In addition to reporting on ForecastBench, we also introduce a more challenging forecasting benchmark sourced from liquid prediction markets. While the AIA Forecaster underperforms market consensus on this benchmark, an ensemble combining AIA Forecaster with market consensus outperforms consensus alone, demonstrating that our forecaster provides additive information. Our work establishes a new state of the art in AI forecasting and provides practical, transferable recommendations for future research. To the best of our knowledge, this is the first work that verifiably achieves expert-level forecasting at scale.
Summary
The AIA Forecaster is an LLM-based forecasting system combining agentic news search, a supervisor agent for forecast reconciliation, and statistical calibration to reduce LLM biases. It matches human superforecaster performance on the ForecastBench benchmark, representing the first verified expert-level AI forecasting at scale. An ensemble with market consensus further outperforms consensus alone, showing complementary predictive value.
Key Points
- •Achieves performance equal to human superforecasters on ForecastBench, surpassing prior LLM baselines and establishing a new state of the art in AI forecasting.
- •Combines three core elements: agentic search over news sources, a supervisor agent reconciling disparate forecasts, and statistical calibration to mitigate LLM behavioral biases.
- •Introduces a new, harder forecasting benchmark sourced from liquid prediction markets, where it underperforms market consensus alone.
- •Ensemble of AIA Forecaster + market consensus outperforms consensus alone, demonstrating the system provides additive, complementary predictive signal.
- •Claims to be the first work verifiably achieving expert-level forecasting at scale, with practical recommendations transferable to future research.
Cached Content Preview
AIA Forecaster: Technical Report
AIA Forecaster: Technical Report
Rohan Alur ∗ ,
Bradly C. Stadie ∗ ,
Daniel Kang, Ryan Chen, Matt McManus, Michael Rickert, Tyler Lee, Michael Federici, Richard Zhu, Dennis Fogerty, Hayley Williamson, Nina Lozinski, Aaron Linsky, Jasjeet S. Sekhon
\addr Bridgewater AIA Labs
New York, NY
Abstract
This technical report describes the AIA Forecaster , a Large Language Model (LLM)-based system for judgmental forecasting using unstructured data. The AIA Forecaster approach combines three core elements: agentic search over high-quality news sources, a supervisor agent that reconciles disparate forecasts for the same event, and a set of statistical calibration techniques to counter behavioral biases in large language models. On the ForecastBench benchmark ( karger2024forecastbench ) , the AIA Forecaster achieves performance equal to human superforecasters, surpassing prior LLM baselines. In addition to reporting on ForecastBench , we also introduce a more challenging forecasting benchmark sourced from liquid prediction markets. While the AIA Forecaster underperforms market consensus on this benchmark, an ensemble combining AIA Forecaster with market consensus outperforms consensus alone, demonstrating that our forecaster provides additive information. Our work establishes a new state of the art in AI forecasting and provides practical, transferable recommendations for future research. To the best of our knowledge, this is the first work that verifiably achieves expert-level forecasting at scale.
0 0 footnotetext: These authors contributed equally to this work. Correspondence to aialabs@bwater.com.
Contents
1 Introduction
Forecasting is a universal problem. Any system that makes informed judgments about what may happen in the future relies on at least some degree of forecasting. In agriculture, this takes the form of predicting future crop yields and food system resilience ( bueechi2023crop ; tanaka2023satellite ; paudel2021machine ) . In biotechnology and medicine, clinical trials are carefully designed to help researchers predict the impact of newly developed treatments ( qian2025enhancing ; curth2024using ) . Elsewhere in biology, AlphaFold can be viewed as a form of forecasting protein folding structure under uncertainty ( jumper2021highly ) . In academia, grant funding is a form of forecasting which scientific discoveries will be made conditional on funding ( tohalino2022predicting ) . Climate and environmental science models local and long-term weather trends, which are of universal importance ( soliman2024deepmind ; price2025probabilistic ; deepmind2024weathernext ) . In political science, forecasting election outcomes and the impact of policy decisions are an essential component of political calculus ( jennings2020election ; wang2015forecasting ; gelman2020information ) . Militaries make use of forecasting for threat assessment and conflict predict
... (truncated, 94 KB total)fde75aac1421b2b6 | Stable ID: sid_JlYZf4leoY