Despite ongoing global volatility marked by wars, pandemics, and political fragmentation, advances in data collection and artificial intelligence (AI) have in some cases enhanced the ability to predict future events. While many perceive the world as increasingly unpredictable, experts suggest that emerging technologies are enabling more accurate forecasts in specific domains.

One notable example is short-term weather forecasting, where AI-enhanced models now produce seven-day predictions with accuracy comparable to three-day forecasts from 1980, owing to improved sensors, expanded data sets, and greater computing power. This raises the question of whether similar improvements are possible in anticipating political and economic developments.

Anthony Vinci, a former US intelligence officer and founder of Vico Technologies, argues that AI represents a breakthrough in forecasting complex events. According to Vinci, there are four primary approaches to assessing future probabilities: relying on individual human superforecasters, leveraging the collective intelligence of organizations such as intelligence agencies or political risk consultancies, utilizing prediction markets like Polymarket or Kalshi, and deploying AI models that integrate these data sources to generate refined insights.

Vico Technologies uses a generative AI system supported by human oversight to analyze vast troves of information and estimate the likelihood of events and their potential cascading effects. These forecasts can assist governments, investors, and corporations in navigating uncertainties such as geopolitical tensions affecting key regions like the Strait of Hormuz. A recent demonstration of the platform's capability was its prediction of a US military action against Iran, assigning an 89 percent probability of occurrence by March 31, several weeks ahead of the February 28 attack. This contrasted with a 63.5 percent probability estimated by the Polymarket prediction market. Vico reports a Brier score—a metric evaluating prediction accuracy and confidence—below 0.15, a level surpassed only by top human superforecasters.

However, such AI-driven forecasting faces significant challenges. One concern is the intentional contamination of training data, referred to as "LLM grooming," where influential actors attempt to skew the inputs and outputs of large language models, potentially compromising their reliability. Additionally, the growing introduction of synthetic, AI-generated data risks further degrading model performance over time. Experts warn this could even lead to a phenomenon likened to "model autophagy disorder," where models effectively consume and corrupt their own generated content.

Carissa Véliz, a philosophy professor at Oxford University and author of *Prophecy*, emphasizes that data and predictions inherently reflect human biases and power dynamics. She cautions that predictive tools create selective reflections of reality rather than a comprehensive picture, reminding that "the map is not the territory."

Prediction markets are also susceptible to manipulation. Recent incidents include Israeli journalist Emanuel Fabian, who faced severe pressure and threats related to reporting missile strikes that affected market bets on Polymarket. Unlike these markets, AI models are less vulnerable to direct interference but must continually adapt to a rapidly changing environment to maintain accuracy. Vinci invokes the "Red Queen’s" metaphor from Lewis Carroll’s *Through the Looking-Glass*, noting that forecasting systems must keep evolving merely to maintain their current effectiveness.

As AI forecasting tools develop, their long-term reliability remains under observation, balancing the promise of greater foresight against complex political, social, and technical constraints.