Artificial Intelligence has become one of the defining technologies of our age. Critics warn that tools such as ChatGPT and Claude are making people intellectually lazy by encouraging them to outsource thinking to machines. Others argue that, like every transformative technology before it, AI is changing the way we think rather than replacing thought itself.
The real question is not whether AI can think for us. It is whether we will continue asking the kinds of questions that give meaning to the answers AI provides. Human inquiry has always depended on this principle: ask the right question to find the right answer. During the Second World War, military analysts studied bombers returning from combat. Their wings and fuselage were riddled with bullet holes. The conclusion seemed obvious: reinforce those areas.
Yet Abraham Wald, a mathematician with the Statistical Research Group, saw what others missed. The aircraft under study were only those that had survived. Bullet holes on the wings showed resilience, not vulnerability. The real danger lay in the engines and cockpit—areas where damage meant the plane never returned.
Wald’s insight became a classic example of survivorship bias: the error of drawing conclusions from what is visible while overlooking what has disappeared. According to a popular retelling of the episode, Wald’s advice was simple: “Do not put armour where the bullet holes are. Put it where there are none.” His lesson endures: evidence may be accurate yet incomplete, and incomplete evidence can lead to false conclusions.
Our responsibility is to not compete with machines in processing information... Our responsibility is to question the evidence before accepting the conclusion.
I posed Wald’s scenario to an advanced AI system. Given a spreadsheet of bullet‑hole coordinates, it quickly produced a heat map highlighting the wings and tail. Based on the data, it recommended reinforcing those areas. When pressed, the system explained survivorship bias and warned against treating surviving aircraft as representative of all aircraft. Its reasoning was sophisticated. Yet the insight emerged only after repeated questionings. Left with the original framing, it analysed the evidence without questioning whether the evidence itself was complete.
That distinction matters. AI can process information at extraordinary speed, but it cannot relieve humans of the responsibility to frame problems intelligently. If we feed it incomplete data, it will produce conclusions consistent with those limits. If we ask superficial questions, it will generate polished answers to superficial questions.
For journalists, this lesson is especially relevant. A government may report declining crime while communities stop reporting offences. A security agency may celebrate fewer recorded attacks while intimidation silences victims. A policy may appear successful because only survivors remain visible in the statistics. In every case, the data may be accurate. The picture may still be incomplete. Our responsibility is not to compete with machines in processing information—AI already does that faster than any human can. Our responsibility is to question the evidence before accepting the conclusion.
In countries like Pakistan, the challenge is sharper. Textbooks are treated as holy scripture, and dissent risks being branded blasphemy. AI’s promise of Socratic questioning cannot flourish in classrooms where questioning itself is forbidden. Without cultural tolerance for inquiry, machines will only accelerate silence. Hints and guided questioning can keep persistence alive in classrooms that encourage curiosity. But in societies where questioning itself is discouraged—where a gesture or word that touches religious belief can be branded blasphemy—Socratic modes collapse. Machines cannot teach critical inquiry if the culture punishes it.
Artificial intelligence should not make us intellectually lazy. It should make us more demanding. Yet in countries where textbooks are scripture and dissent is punished, even the most advanced machine cannot rescue us from silence. AI can show us patterns, but only human judgment can see the absences. The danger is not that machines will mislead us, but that we will stop asking the questions they cannot. Who is missing from the dataset? What realities never entered the record? What assumptions shaped the evidence before it reached the machine? And, what false ideas were not questioned? Unless we insist on asking these questions, we risk surrendering not to AI, but to our own fear of inquiry.