How much crit­ic­al think­ing can we out­source?

AI and education expert Sine Zambach highlights three essential questions: How does it work? How can it be used? And how can we approach it critically? These are not only current challenges – they will only become more pressing over time. “In five years, we will have students who have relied on language models as constant study companions throughout secondary school. In ten years, they will have used them since primary school. Some will have learned to use the technology sensibly, others not. And much depends on how quickly teachers and ministries decide to act,” she says.

“The most important skill young people need is critical thinking. But critical thinking likely requires a foundation of knowledge,” Zambach adds. She warns that students risk cheating themselves if they use ChatGPT as a shortcut to ready-made answers – unless that is the explicit purpose of the exercise: “They lose the basic skills needed to engage critically with the output.”

When teachers cannot – or will not – join in 

As the technology races ahead, many teachers remain hesitant. Some reject students’ use of AI because they themselves lack sufficient knowledge. Others are sceptical about the quality of language technologies. The consequence is that students often feel left on their own. “It becomes incredibly easy to end up using it in the wrong way,” says Zambach.

She points out that scepticism also varies across subjects. “I’ve seen quite a few maths teachers being critical, perhaps because language technologies are still poor at solving equations. By contrast, I’ve met teachers in subjects like religion or classical studies who see possibilities, because the models can draw on vast amounts of knowledge and present students with new perspectives.”

This difference shows that the barriers are not only about technology but also about subject cultures and pedagogical traditions. And that leaves students in a skewed situation: the strongest figure out how to use the technology constructively, or receive help from their parents, while the weakest risk ending up with bad habits and uncritical use. In this way, a technology with great potential to level inequalities may instead create new forms of disparity.

Critical thinking across disciplines

Critical thinking has become a mantra in the education debate. But according to Zambach, it is not a single skill that can be taught across all subjects. “In programming, critical thinking is about optimising code, making it more precise and understanding why an output looks the way it does. In international politics, it is something else entirely – it’s more about evaluating sources and perspectives,” she explains.

What all subjects have in common, however, is that critical thinking requires a knowledge base. Without it, the output of language technology is difficult to assess. Generative AI produces plausible answers – not necessarily correct ones. And that becomes problematic if students lack the tools to detect mistakes.

Zambach gives a concrete example: when OpenAI launched GPT-5 live, the model was asked to explain how an aircraft wing works. The answer sounded convincing – but was factually wrong. “It’s an explanation that circulates widely online, even though NASA has shown it to be false. And precisely because it is repeated so often, it becomes reinforced in the models,” she says.

Thus, AI raises an old challenge in a new form: how do we teach students to distinguish between plausibility and knowledge? Where once students had to learn source criticism in texts, they now must also scrutinise probability-based answers from machines. And here it is not enough to read the result – one has to understand the principles behind it, and why the error occurs.

From learning to formation

The discussion is not only about academic content. Education is also about personal formation. It is about meeting others, debating and becoming part of networks that extend into future working life. Here, AI can support, but (for now) not replace.

If AI is reduced to a quick shortcut, we lose the social and critical dimension of learning. The risk is that we confuse access to information with insight – and knowledge with wisdom. And in the long run, that may undermine the part of education that shapes young people into independent citizens in a democracy.

AI literacy as the key to the future

AI literacy is therefore the key to future education: understanding, use, critique. The question is no longer whether generative AI should be part of teaching – but how.

Today’s students are already experimenting with AI in their learning. Soon, an entire generation will arrive at university with the technology embedded in their habits. Whether it becomes a tool for deeper reflection or just a shortcut depends on how we choose to teach with it now.

“Generative AI can become a powerful extension of human thought. But without a critical and formation-oriented framework, we risk hollowing out precisely what education is meant to provide: the ability to think independently,” concludes Sine Zambach.