The Science of Superforecasting: How Humans and AI Are Learning to Predict the Future

The future is inherently uncertain, yet a select group of individuals consistently predict global events with astonishing accuracy. These individuals, known as "superforecasters," are not necessarily subject-matter experts or intelligence insiders with access to classified data. Instead, they are methodical thinkers who treat prediction as a rigorous, learnable skill. The concept gained prominence through the Good Judgment Project, a massive geopolitical forecasting tournament launched in 2011, which revealed that the top tier of forecasters could consistently outperform professional intelligence analysts. Today, the science of superforecasting has evolved from a niche academic experiment into a critical tool for governments, philanthropic organizations, and corporate boardrooms seeking to navigate an increasingly volatile world.[7][5][8]

The stakes for accurate forecasting are immense, driving institutions to seek out the best predictive minds. For example, major philanthropic evaluators like GiveWell now contract professional superforecasters to predict future U.S. government foreign aid funding levels, using these probabilistic estimates to guide millions of dollars in global health grants. By relying on structured foresight rather than gut intuition, organizations can anticipate funding gaps, stress-test their strategic assumptions, and allocate resources more effectively. This shift represents a broader recognition that true predictive power lies not in who has the most information, but in who processes that information most objectively.[5][8]

The secret to superforecasting lies in a disciplined cognitive process rather than innate clairvoyance. Elite forecasters rely heavily on decompositional reasoning, a technique that involves breaking a massive, complex question into a series of smaller, more tractable sub-problems. Instead of trying to guess the outcome of a distant election or an economic shift in one leap, they estimate the probability of the underlying variables that would lead to that outcome. This forces the forecaster to articulate their explicit assumptions, making it easier to identify flaws in their logic and adjust their predictions as new data emerges.[1][8]

Equally critical to the superforecaster's toolkit is the practice of Bayesian updating. For the average person, beliefs are often treated as treasures to be guarded; for a superforecaster, beliefs are merely hypotheses to be tested. They practice active open-mindedness, continuously revising their probability estimates up or down the moment new, credible information arrives. They do not strive for absolute certainty, which often leads to overconfidence and error. Instead, they aim for granular accuracy, assigning precise probabilities—like 63% or 18%—rather than relying on vague terms like "likely" or "doubtful".[1][8]

To quantify this accuracy, the forecasting community relies on a rigorous mathematical metric known as the Brier score. The Brier score measures the mean squared difference between a predicted probability and the actual outcome, with scores ranging from 0.0 (perfect accuracy) to 1.0 (perfect inaccuracy). For a simple yes-or-no question, a forecaster who flips a coin and predicts 50% every time will earn a Brier score of 0.25, which serves as the baseline for random guessing. In elite tournaments, human superforecasters consistently achieve Brier scores below 0.12, placing them in the 90th percentile of all participants and demonstrating a persistent, measurable skill that defies sheer luck.[1][3]

As the science of individual forecasting has matured, the financial sector has scaled the concept through prediction markets like Polymarket and Kalshi. These platforms allow participants to buy and sell contracts tied to the outcomes of real-world events, effectively transforming collective uncertainty into tradable financial instruments. By attaching real financial stakes to predictions, these markets incentivize participants to gather diverse, fragmented information and express their beliefs with capital. The resulting contract prices serve as a real-time, aggregated probability estimate, often rivaling traditional polling data in speed and responsiveness to breaking news.[4]

However, the "wisdom of the crowd" harnessed by prediction markets is not infallible, particularly in highly specialized domains. A recent evaluation of prediction markets during 2025 and 2026 found that platforms like Polymarket failed to outperform standard statistical baselines when forecasting infectious disease dynamics, such as influenza hospitalizations and measles outbreaks. In these complex public health scenarios, the markets suffered from low trading volume and a lack of epidemiological expertise among self-selected participants, ultimately falling short of expert-curated ensemble models. This underscores a critical limitation: financial incentives alone cannot replace domain-specific data and structured modeling when the underlying mechanics of an event are highly technical.[6]

Recognizing the limits of both human crowds and individual experts, the forecasting frontier has rapidly shifted toward artificial intelligence. Early iterations of large language models were notoriously poor at predicting the future, often scoring worse than random guessing due to persistent overconfidence and an inability to calibrate their probabilities. However, the landscape shifted dramatically with the development of "Superforecasting LLMs"—models explicitly architected and trained to emulate the reasoning discipline of elite human forecasters. These advanced systems utilize multi-agent search layers to ingest real-time data, apply Bayesian calibration routines, and synthesize evidence through structured chain-of-thought prompts.[1][3]

The empirical results of these specialized AI models have been striking. By late 2025, advanced LLMs had surpassed the accuracy of the median public forecaster, achieving Brier scores that represent a massive leap over previous generations of AI. Researchers found that by using ensemble approaches—aggregating the predictions of multiple diverse LLMs—the artificial crowd could achieve accuracy levels statistically indistinguishable from the human crowd. In some benchmarks, these AI ensembles successfully mitigated the overconfidence effects that plagued earlier models, proving that silicon-based reasoning could be calibrated to handle real-world uncertainty.[1][2][3]

Despite these rapid advancements, the very best human minds still retain a measurable edge over the machines. In recent benchmark evaluations, elite human superforecasters maintained a difficulty-adjusted Brier score of 0.081, while the most capable AI model scored 0.101. This translates to a roughly 20% accuracy advantage for the top humans. The human edge is particularly pronounced in categories with sparse data that require highly subjective judgment, nuanced contextual understanding, and the ability to weigh the credibility of conflicting human sources. While researchers project that simple linear extrapolation could see AI reach parity with superforecasters by late 2026 or 2027, the gap remains significant for the most complex geopolitical and economic questions.[2][3]

Consequently, the most powerful forecasting paradigm emerging today is not a competition between humans and machines, but a hybrid collaboration. Studies have shown that when human forecasters are paired with specialized AI assistants, their prediction accuracy increases by 23% to 43%. The AI assistants excel at rapidly synthesizing vast public datasets, identifying historical base rates, and enforcing structured reasoning, while the human forecaster provides the nuanced judgment and contextual intuition that the model lacks. This synergy effectively combines the tireless data-processing scale of silicon with the qualitative wisdom of the human brain.[1][8]

For corporate leaders and policymakers, the value of this hybrid approach extends far beyond the final probability number. What decision-makers truly value is the rationale behind the forecast. Superforecasters—and increasingly, their AI assistants—provide a transparent chain of reasoning that details exactly how they arrived at their conclusion. This allows leaders to scrutinize the underlying assumptions, trace causal links, and stress-test strategic scenarios by identifying hidden risks. In an era defined by rapid change and complex global challenges, the ability to methodically map the future is no longer a parlor trick; it is an essential discipline for survival and success.[2][8]