Dragan Gašević, Distinguished Professor at Monash University, Australia, and Director of its Centre for Learning Analytics, delivered a keynote speech titled "Human-AI Collaboration in Learning: Beyond Performance Gains" at the parallel session on "AI Education Development and Evaluation" during the 2026 World Digital Education Conference.

He noted that a recent study by his team identified four major traps learners easily fall into when collaborating with generative AI, which deserve particular attention.

First, the fluency trap. When learners read information that is very fluent, they tend to overestimate their own understanding and therefore reduce deep elaboration of the content. This phenomenon also manifests in their writing tasks. In other words, learners may reduce their engagement during the learning process by as much as 50%, exhibiting metacognitive laziness. Although AI reduces the cognitive friction required for learning, it also makes learners prone to this fluency trap.

Second, the alignment trap. This trap occurs mainly when people try to align with the answers given by AI. When you ask AI a question, it supplies several answers and does the thinking for you. However, numerous studies show that this leads to a decline in learners' creativity. Many learners behave similarly when collaborating with AI: they simply accept the surface information provided by AI and go along with it, rather than truly entering a process of autonomous thinking.

Third, the conformity trap. A recent study found that when learners collaborate with AI on writing tasks, they internalise some of AI's biases as their own views without being aware of it. Even when they have been warned beforehand, they still fall into this trap unconsciously. Learners do not realise the extent of this internalisation; they are unaware of it.

Fourth, the transparency trap. This trap stems from good intentions in designing AI: we want AI to no longer be a black box but to be able to explain its own operations. However, sometimes AI over-explains, generating extensive explanations very fluently. In this process, learners often fail to attend to the uncertainty that lies beyond AI's decisions. Because when using AI, some information is true while some is false, but AI conceals this fact. Many learners do not question what AI provides; they accept everything it offers at face value.