WE have always been fascinated by people who claim to know the future. The more accomplished the person, the more seriously we tend to listen. A billionaire technology founder predicts when humans will live on Mars. An economist tells us where markets are headed. An AI researcher warns that machines could destroy humanity within a few years.
Then the date arrives. Nothing happens.
Elon Musk offers a good example. In 2016, Musk said SpaceX should be able to launch people to Mars in 2024, arriving in 2025. That did not happen. In 2024, he offered another ambitious timeline, suggesting humans could reach Mars within four years. SpaceX has since continued developing Starship, but its own current Mars information says cargo flights will start no earlier than 2028.
This does not mean Musk knows nothing about rockets. SpaceX has changed the economics and engineering of spaceflight. It simply tells us something important: being very good at building the future does not necessarily make someone very good at dating its arrival.
We should remember this as warnings about artificial intelligence (AI) become more dramatic.
Daniel Kokotajlo, a former OpenAI researcher, became known for the “AI 2027” scenario, which described a path in which rapidly advancing AI could eventually lead to humanity’s destruction by around 2030. But the timeline has already changed. In early 2026, his team pushed its expectation for fully autonomous AI coding into the early 2030s and superintelligence to around 2034.
More recently, former OpenAI and Anthropic researcher Jacob Coxon publicly warned that advanced AI could kill humanity by the end of the decade. Even Sam Altman has spoken about a meaningful probability of catastrophic outcomes while arguing that companies and governments can take measures to reduce the danger.
These warnings should not simply be dismissed. AI can create serious risks. Cyberattacks, biological threats, misinformation, fraud and autonomous systems deserve attention. But there is a big difference between saying a technology presents a risk and saying we know what will happen by 2030.
History should make us humble.
Just before the 1929 stock market crash, celebrated economist Irving Fisher said stock prices had reached what looked like a “permanently high plateau.” Days later, the market collapsed. By mid-November, the Dow had lost almost half its value.
In 1968, Stanford biologist Paul Ehrlich's best-selling book “The Population Bomb” warned that hundreds of millions of people would starve during the 1970s and 1980s. The predicted global catastrophe did not occur. Agricultural technology, better seeds, fertilizers and the Green Revolution changed the food equation in ways the forecast did not adequately anticipate.
Energy forecasting provides another lesson. In the late 1990s, influential predictions warned that world crude oil production would peak around 2004 or 2005 and then decline inexorably. What these models did not foresee was unconventional oil and the technological and economic shifts that would make new sources commercially viable.
There is a pattern here.
Forecasters often extend what they can see today into tomorrow. But the future introduces variables that are difficult to model: new technologies, regulation, wars, economic incentives, human behavior, competing inventions and simple accidents. The further into the future we go, the larger the uncertainty becomes.
Research by Philip Tetlock reinforces this point. His long-running work on expert judgment found that political experts were often overconfident, particularly when making longer-term forecasts. Later research also showed something more nuanced: expertise can help in certain areas, but forecasting accuracy tends to deteriorate as the time horizon gets longer.
This is why I become cautious whenever someone gives an exact year for a dramatic event.
“AI could create catastrophic risks” is a proposition worth examining.
“Humanity will be destroyed by AI by 2030” is a forecast.
Those are not the same statement.
There is another problem. Extreme predictions travel faster than cautious ones. “AI capabilities are advancing quickly, but the timing and consequences remain uncertain” will not generate much attention. “AI will kill us all in four years."
Fear has always been good media.
That does not mean we should swing to the opposite extreme and ignore experts. Experts understand technical details that most of us do not. Their warnings can reveal risks early enough for society to act. In fact, sometimes a frightening prediction fails precisely because people respond to it. Forecasts can change behavior, and changed behavior can change outcomes.
What we need is neither blind trust nor automatic skepticism.
When I hear a bold forecast, I ask a few simple questions. What assumptions have to be true for this to happen? What could change those assumptions? Has the forecaster made previous predictions, and how accurate were they? Is the forecast expressed as a probability or as a certainty? And perhaps most important, what evidence would cause the expert to change his or her mind?
A good forecaster should be willing to update.
The future is not a scheduled event waiting for us to arrive. It is shaped by billions of decisions, inventions, failures, policies and surprises. Experts can help us understand the possibilities. They cannot remove the uncertainty.
So, when someone tells us that humans will live on Mars by a certain year, that AI will eliminate most jobs by another year, or that humanity itself has only a few years left, we should listen.
But we should also ask questions.
Expertise deserves respect. Prediction deserves scrutiny.
The author is the founder and CEO of Hungry Workhorse, a digital and culture transformation consulting firm.
