Human forecasting accuracy in practice relies on the 'wisdom of the crowd' effect, in which predictions about future events are significantly improved by aggregating across a crowd of individual forecasters. Past work on the forecasting ability of LLMs suggests that frontier LLMs, as individual forecasters, underperform compared to the gold standard of a human crowd forecasting tournament aggregate. In Study 1, we expand this research by using an LLM ensemble approach consisting of a crowd of twelve LLMs. We compare the aggregated LLM predictions on 31 binary questions to that of a crowd of 925 human forecasters from a three-month forecasting tournament. Our preregistered main analysis shows that the LLM crowd outperforms a simple no-information benchmark and is not statistically different from the human crowd. In exploratory analyses, we find that these two approaches are equivalent with respect to medium-effect-size equivalence bounds. We also observe an acquiescence effect, with mean model predictions being significantly above 50%, despite an almost even split of positive and negative resolutions. Moreover, in Study 2, we test whether LLM predictions (of GPT-4 and Claude 2) can be improved by drawing on human cognitive output. We find that both models' forecasting accuracy benefits from exposure to the median human prediction as information, improving accuracy by between 17% and 28%: though this leads to less accurate predictions than simply averaging human and machine forecasts. Our results suggest that LLMs can achieve forecasting accuracy rivaling that of human crowd forecasting tournaments: via the simple, practically applicable method of forecast aggregation. This replicates the 'wisdom of the crowd' effect for LLMs, and opens up their use for a variety of applications throughout society.
Ensemble LLMs have been shown to match human crowd performance in probabilistic forecasting.
A study compared an ensemble of twelve LLMs against the aggregated predictions of 925 human forecasters, finding statistical parity in accuracy.
Incorporation of human cognitive outputs into LLMs refined prediction confidence levels and enhanced forecasting accuracy.
The research identifies potential for practical applications and future investigation into hybrid human-AI forecasting models.
Significant strides have been made in the capabilities of LLMs, notably through the utilization of ensembles of diverse models to imitate the human 'wisdom of the crowd' phenomenon. This approach has now been rigorously tested against human forecasting accuracy, revealing LLMs' potential to match human crowd performance in the domain of probabilistic forecasting. Through two distinct but interconnected studies, researchers have explored the ensemble method's efficacy and the influence of human-derived forecasts on LLM predictions.
In the first study, an ensemble of twelve LLMs was compared against the aggregated predictions of 925 human forecasters in a forecasting tournament. The critical findings include:
Exploring further, the second study delved into the potential of enhancing LLM predictions with human cognitive outputs. Key results include:
The collectively drawn conclusion from these studies not only heralds a significant benchmark in LLM capabilities but also opens up avenues for practical applications and further academic inquiry:
Through an ensemble approach, LLMs have exhibited a capacity to match and potentially surpass human forecasting accuracy. These findings do not just signify a milestone in artificial intelligence but also offer a glimpse into future interdisciplinary research and application pathways. As LLMs continue to evolve, the integration of human cognitive outputs may serve not only to refine predictive accuracies but also to harness the collective strengths of both human and machine intelligence, forging a new frontier in forecasting methodology.