FT column says people should challenge AI answers, not rubber-stamp them

In short: A Financial Times columnist argues that AI is most useful when people actively question it, not when they simply approve its answers.

What's going on

Vivienne Ming, a neuroscientist and author, wrote in the Financial Times that many people treat AI outputs as if they are automatically correct because they sound confident and smooth.

She describes an experiment in a classroom where students used the same AI helper while wearing EEG headsets. EEG is a tool that measures brain activity (like a heart monitor, but for your brain). For many students, signals linked with mental effort dropped quickly, suggesting they were closer to “watching TV” than thinking through a problem.

A smaller group showed the opposite pattern. Their brain activity stayed high because they challenged the AI, pushed back on its answers, and asked it to critique their thinking. Ming argues these students would likely learn more over time, even if their final essays looked similar.

Ming also points to earlier education research. A 2019 Harvard study found that students learned more when they struggled with problems, but they felt like they learned less. The column says AI can increase this problem by making information feel easy, even when understanding is shallow.

What to watch

The column criticizes policy approaches that rely on “human oversight” as a simple checkbox, such as parts of the EU AI Act. Ming warns that if a person is tired, rushed, or over-trusting, they may just rubber-stamp wrong AI answers. She suggests testing AI systems based on how they affect users over time, and designing AI to slow people down by asking questions instead of giving direct answers.

Source: Financial Times