Despite widespread perceptions that global events are increasingly chaotic and unpredictable, emerging technologies and data analytics are enhancing the ability to forecast certain developments with greater accuracy. While geopolitical tensions, markets, and public health crises often appear to unfold in an accelerated and turbulent fashion, the use of advanced artificial intelligence (AI) models combined with larger, more precise data sets is improving predictions in some domains.

One example cited is short-term weather forecasting, where improvements in sensor technology and computational power have enabled seven-day forecasts today that match the reliability of three-day forecasts from four decades ago, despite the complicating factor of climate change. This progress raises the question of whether similar advances can be applied to anticipating human-driven political and economic events.

Anthony Vinci, a former U.S. intelligence officer and founder of Vico Technologies, argues that AI represents a significant breakthrough in event forecasting. Vinci emphasizes the limitations of personal judgment in assessing future outcomes and the increasing reliance on AI tools to provide more grounded analyses. His firm’s platform integrates generative AI with human oversight, synthesizing inputs from expert forecasters, organizational intelligence, crowdsourced betting markets, and other data streams to generate probabilistic assessments of geopolitical events and their potential ripple effects.

A notable instance of Vico’s forecasting capabilities was its prediction in mid-February of an 89% probability of a U.S. attack on Iran by the end of March—a prediction that preceded the February 28 strike and outperformed market-based forecasts available at the time. Vico reports a strong track record measured by the Brier score, a metric evaluating forecast accuracy and confidence, asserting performance superior to nearly all individual human superforecasters.

However, challenges remain in maintaining the integrity and reliability of AI-driven forecasts. The datasets used to train and inform these models are vulnerable to intentional manipulation, a phenomenon referred to as “LLM grooming,” where actors seek to distort large language model outputs by poisoning input data. Additionally, the proliferation of synthetic data generated by AI itself poses risks of feedback loops that could degrade model accuracy or cause systemic failures, sometimes described metaphorically as “model autophagy disorder.”

Philosopher Carissa Véliz of Oxford University highlights the inherent limitations of predictive efforts by underscoring that all data represent human constructions influenced by power dynamics. She cautions against equating predictive models with comprehensive depictions of reality, noting that such models are inevitably partial and shaped by what users choose to measure and prioritize.

Moreover, prediction markets, which aggregate the wisdom of crowds through betting mechanisms, have demonstrated susceptibility to manipulation both prior to and following events. The recent case of Israeli journalist Emanuel Fabian, who faced severe harassment aimed at altering his reporting on missile strikes near Jerusalem and thereby affecting related market bets, illustrates such vulnerabilities.

While AI-based forecasting tools like those developed by Vico may not be subject to the same forms of direct manipulation, their long-term robustness in an ever-changing world remains to be seen. Vinci acknowledges the ongoing need for adaptation, referencing the literary metaphor from Lewis Carroll’s Through the Looking-Glass: substantial effort is required just to maintain existing ground in a rapidly evolving environment.