Apparently, Artificial Intelligence is going to kill us.
Depending on which headline you read, superintelligent machines may escape our control, outsmart their creators and eventually decide that humanity was an interesting experiment that should probably be discontinued. Some prominent AI researchers have even attached probabilities to catastrophe, with figures of 10 percent or more occasionally appearing in public discussion.
There is just one inconvenient detail: science cannot currently tell us that humanity has a 10 percent probability of being wiped out by AI.
That does not mean the risk is zero. The International AI Safety Report 2026, produced by more than 100 experts, takes potential loss of human control extremely seriously. But it also makes something equally important clear: expert estimates vary enormously, current AI systems do not possess the capabilities required for such a takeover, and the probability, timing and nature of catastrophic loss of control remain deeply uncertain.
Ten percent, therefore, isn't the equivalent of saying there is a 10 percent chance of rain tomorrow. There is no historical database containing 10,000 previous encounters between humanity and superintelligence from which we can calculate the number ending badly. These are informed judgments about unprecedented future events, not actuarial tables.
Meanwhile, a considerably less cinematic danger is sitting directly in front of us.
We may simply stop thinking.
Research is beginning to provide evidence for this concern. A 2025 Carnegie Mellon–Microsoft Research study examined 319 knowledge workers and 936 examples of AI use. Greater confidence in AI was associated with less critical thinking, while greater confidence in one's own abilities was associated with more.
There lies the real danger.
AI doesn't have to overthrow humanity if humanity cheerfully hands over the keys.
We already have a spectacular demonstration of what happens when people abandon verification and substitute tribal loyalty for evidence. Consider the MAGA movement. After the 2020 American election, courts rejected claims of widespread fraud and every state certified its results, yet Pew found that three-quarters of Trump supporters still believed Trump was definitely or probably the rightful winner. Research has also found Trump supporters disproportionately visiting unreliable websites, while political identity has increasingly influenced trust in scientific institutions.
This is not an argument that stupidity belongs exclusively to one political tribe. Humans across the ideological spectrum are remarkably talented at believing information that confirms what they already want to believe. We apparently arrived equipped with confirmation bias long before Silicon Valley thoughtfully automated it for us.
But MAGA provides an extraordinary contemporary case study: when identity becomes more important than evidence, evidence becomes optional.
Now add AI.
Imagine billions of people accepting machine-generated answers because they sound authoritative, outsourcing research because it is inconvenient, surrendering judgment because thinking takes time, and gradually losing the intellectual habits required to challenge the machine in the first place.
That is a far more plausible road toward losing control.
The irony is that AI itself need not become evil. It merely needs humans to become intellectually lazy.
There are genuine risks from increasingly powerful AI: autonomous weapons, cyberattacks, biological misuse, manipulation, surveillance and potentially systems we cannot adequately control. Pretending otherwise would be irresponsible. Stanford's 2026 AI Index describes capabilities as accelerating while our ability to govern and evaluate them struggles to keep pace.
But our future has not yet been outsourced.
Which brings us inevitably to schools and homes.
If we want children prepared for an AI-dominated world, teaching them how to prompt ChatGPT isn't nearly enough. They must learn to question answers, demand evidence, recognize nonsense, compare perspectives, tolerate uncertainty, change their minds and occasionally have the courage to say, “No. That doesn't make sense.”
Parents and educators therefore have a rather important assignment.
We must develop Human Intelligence capable of controlling Artificial Intelligence.
The machines may become astonishingly powerful.
Whether we surrender our thinking to them remains entirely up to us.