Past, present, and future Venn diagram (Source: AI-Gemini)
Throughout our lives, all personal, collective, global, tribal and national issues consist of an interplay of past, present, and future. This is formalized in the top picture, which was constructed for me by AI (Gemini). We can mostly factualize the past (if we can still find relevant evidence). We can experience the present and try to predict the future as best as we can. I have tried, to the best of my ability, to predict the future throughout this blog (put “predictions” into the search box), with an emphasis on the tools that can help me to quantify these predictions. This is important because some of the predictions are for existential issues and one needs time to act on them to mitigate them. My book, Climate Change: the Fork at the End of Now, focused on such scenarios (See the “My Now” blog from March 9, 2021). Regarding a different threat, a recent Nobel Prize–winning nuclear physicist predicted that humanity has about 50 years to survive:
The chances of you living 50 years are very small’: Theoretical physicist explains why humanity likely won’t survive to see all the forces unified. Live Science spoke with Nobel prize-winning physicist David Gross, who recently received the $3 million Special Breakthrough Prize in Fundamental Physics, about the quest to unite all the forces and why humanity might not live to see a unified theory.
50 years is well within my definition of “Now” (within the lifetime of my grandchildren), which takes us to the end of the century.
Prof. Gross’ conclusions are based on the likelihood of nuclear war, which recently increased from 1% per year to 2%. I didn’t see the origin of the change in the yearly estimate in his interview, but once you accept the increase in yearly estimate, the math is very clear and follows simple exponential decay that he is very familiar with.
My guess is that Prof. Gross’ estimate of the yearly doubling the chance of nuclear confrontation is based on the Doomsday Clock, which was discussed in an earlier blog (June 18, 2025).
The two main events that are pushing forward the Doomsday Clock are the Russian–Ukrainian war and the US-Israel-Iran war. Recent developments in the Russian–Ukrainian war have given the sense to many (including me) that—against all odds—the Ukrainians are gaining the upper hand:
Ukraine Bombards Moscow With One of the Biggest Drone Attacks of the War. The attack, which shut down the capital’s airports for several hours, was part of an escalating campaign to bring the conflict home to Russians.
It seems to many that at some point, the Russians will either choose to end the war with an agreement or destroy Ukraine with nuclear weapons. However, the nuclear destruction of Ukraine would not stay in Ukraine.
AI is now developing to help us with quantitative predictions about this complicated (and frightening) picture. I suspect that quantitative predictions of the future will be one of the largest applications of AI. This blog will focus on the general abilities of the technology right now; next week’s blog will focus on how it understands climate change (which my background gives me the capability of scrutinizing). In two weeks, I will write about the guardrails that are now under discussion for AI.
Before shifting to these details, I thought that it would only be“fair” to let AI outline its own capabilities in quantitative prediction of the future. The AI systems that I used in this exercise were Gemini (Google) and CoPilot (Microsoft). My prompts (as well as my answer to a question that AI posed to me) are marked in red.
Through Gemini:
“AI as a predictor of the future?”
AI is rapidly transitioning from an analytical tool into an active predictor of the future. By shifting forecasting from a human craft to an automated utility, machine learning models can process massive, complex datasets to predict everything from market shifts to weather events. In live forecasting platforms like Metaculus, advanced AI systems have already achieved over 80% of the accuracy of elite human forecasters on geopolitical and technology trends. [1, 2, 3]
The capabilities, limitations, and societal impacts of using AI as a “crystal ball” involve several key factors:
Where AI Excels at Predicting
• Time-Series and Pattern-Heavy Data: AI thrives in environments with minimal “distributional shift”—where past data patterns remain relatively stable. [1]• Scientific & Medical Discoveries: AI models analyze vast literature and biological data to predict valid molecular structures for drugs or map out evolving scientific fields. For instance, a Mayo Clinic team developed an AI capable of detecting pancreatic cancer on CT scans up to three years before clinicians can see it. [1, 2, 3, 4, 5]• Hyper-Local Climate and Weather: AI drastically speeds up environmental calculations, identifying patterns linked to storms, floods, and wildfires with remarkable short-term accuracy. [1, 2]• Crowdsourced Forecasting: Benchmarks like the University of Chicago’s Profit Arena evaluate AI models live on real-world prediction markets, showing that certain AI models find statistical edges that humans miss. [1]Severe Bottlenecks to AI Forecasting [1]
• The “Data Gap”: AI struggles significantly when there is a time gap between its training data and a current event. It cannot naturally account for sudden, unprecedented “black swan” events, such as geopolitical outbreaks or unrecorded financial crises. [1, 2]• Overconfidence and Hallucination: Evaluation studies by institutions like Knowledge at Wharton show that AI models frequently exhibit overconfidence, assigning dangerously high probability scores to outcomes that lack factual evidence. [1]• Lack of True Intuition: AI relies purely on historical data correlation. Unlike humans, it lacks genuine creative insight or “judgmental forecasting” capabilities—the ability to assign probabilities based on abstract intuition and personal domain expertise. [1, 2]The Dangers of Algorithmic Future-Telling
• Self-Fulfilling Prophecies: When a dominant AI system predicts a market crash or a political outcome, human actors change their behavior to match that prediction, inadvertently forcing it to come true. [1, 2, 3, 4, 5]• Pre-Crime and Surveillance: Authoritarian governments are actively testing AI to analyze location history and internet traffic to predict who might become a political dissident before any public act of protest occurs. [1]• Feedback Loops: As AI-generated content floods the web, newer forecasting models risk being trained on data predicted by older models, leading to a closed-loop degradation of accuracy. [1, 2, 3, 4]Ultimately, AI is a powerful enhancer of human prediction, not a replacement for it. The most reliable forecasts still occur when AI handles the brute-force data crunching, leaving humans to provide contextual oversight, ethical boundaries, and strategic judgment. [1, 2, 3]
Through CoPilot:
“AI as a predictor of the future?”
AI can predict aspects of the future—but only in specific, data-rich domains, and never with perfect certainty. Across science, business, and geopolitics, AI is already outperforming humans in some forecasting tasks, yet it also introduces new risks when misused. Below is a clear, evidence-based breakdown of what AI can and cannot do as a predictor of the future.
🧠 What AI Can Predict Reliably
1. Natural disasters (partially)
AI improves early warning systems by analyzing massive streams of satellite, weather, and environmental data far faster than humans. It detects patterns linked to hurricanes, floods, droughts, and wildfires, helping scientists issue
earlier and more accurate alerts. However, AI cannot eliminate uncertainty, especially for chaotic systems like earthquakes.
2. Business and technology outcomes
Large language models have begun outperforming trained human experts in predicting which new tech ventures will succeed. In a controlled tournament, top AI models correctly identified winners in nearly 4 out of 5 cases, beating MBAtrained managers. This suggests AI is becoming a powerful tool for strategic foresight in business.
3. Geopolitical forecasting
AI systems have competed in forecasting tournaments (e.g., Metaculus) and achieved accuracy levels comparable to top human forecasters—sometimes surpassing them. One AI system scored over 80% of top human performance, far above expectations.
Where AI Prediction Becomes Dangerous
4. Predicting human behavior for surveillance
Some governments are experimenting with AI to forecast political dissent or identify potential future critics by analyzing online activity, location data, and behavioral patterns. This raises severe ethical and civilliberty concerns and demonstrates how predictive AI can be weaponized.
5. Predicting social outcomes at national scale
Argentina attempted to build a “social digital twin” to simulate future poverty, subsidy impacts, and humancapital development. While ambitious, the system’s rollout was flawed and sparked skepticism—highlighting the difficulty of predicting complex social systems.
What AI Might Predict in the Future
Experts and analysts foresee AI becoming deeply embedded in forecasting across society:
• Predictive healthcare: anticipating diseases years before symptoms.• Economic and societal modeling: more accurate simulations of policy impacts.• Everyday anticipatory systems: AI that predicts your needs before you express them.• Longterm societal shifts: experts are divided—56% of AI experts expect a positive impact over 20 years, while the public is more skeptical.The Bottom Line
AI is not a crystal ball. But it is becoming a powerful forecasting engine—especially where:
• There is abundant, highquality data• Patterns are stable and quantifiable• Predictions can be updated continuously• Human bias or slow analysis is a bottleneckAI struggles or becomes dangerous when:
• Predicting complex human behavior• Used for political or social control• Data is biased, incomplete, or invasive• Systems are deployed without oversight• A question for you, Micha• Are you thinking about AI as a tool for personal forecasting (e.g., career, finance), societal forecasting, or philosophical questions about the future of humanity?
My answer: “All the above and the interactions between them. Specifically – how can I check the probabilities of these predictions.”
Following my response, AI gave me the following links:
Can AI Help Scientists Predict Natural Disasters More Accurately in the Future?
AI beats human forecasters in tournament predicting 30 tech ventures
Chinese Firm Develops Technology That Uses AI To Predict Future Protestors
https://www.yahoo.com/news/politics/articles/argentina-launched-ai-predict-future-150104368.html
AI May Soon Understand Humans Better Than We Understand It, Scientists Warn
3. Public and expert predictions for AI’s next 20 years
Meanwhile, The Guardian covered an optimistic vision of the future that is “the product of 45 authors based on databases compiled by more than 200 researchers from around the world”:
Humanity can raise living standards, reduce inequality and keep global heating within a 2C rise, according to a sweeping vision for planetary survival. The report by the World Inequality Lab (WIL) aims to be the most comprehensive attempt yet to navigate the polycrisis that is pushing the world toward climate breakdown, political extremism and ever greater economic and social tension. It offers a set of bold policy proposals, including hefty wealth taxes on billionaires, sharp reductions in working hours, a change in diets and a shift of investment from materially intense sectors, such as industry and mining, to education and health.
While the example above is human-driven, this could also be the start of a different conversation with AI about how to make the world a better place.
Stay tuned!