Unplugged lessons power up children’s AI skills

AI literacy has become a core digital skill for the modern world, no longer limited to coders or engineers. Yet, teaching AI to children poses unique challenges. The researchers note that current educational practices mostly rely on “plugged-in” activities involving coding software, robotics, or web-based AI tools. While these foster practical familiarity, they can overwhelm younger learners unfamiliar with computational logic.

Artificial intelligence education is entering the classroom at a younger age than ever, but how best to teach complex concepts to children remains uncertain. A new study, titled “Developing Artificial Intelligence Literacy Through Mixed Unplugged and Plugged-in Activities in Primary Education” and published in SAGE Open, provides some of the first empirical answers.

The research tests whether combining unplugged activities, games, stories, and physical exercises, with plugged-in, computer-based projects helps primary students build AI literacy more effectively than digital lessons alone. Conducted in a Chinese primary school, the study found that starting with unplugged lessons and then moving to digital exercises produced stronger understanding and more positive attitudes toward AI learning.

Rethinking AI literacy in early education

AI literacy has become a core digital skill for the modern world, no longer limited to coders or engineers. Yet, teaching AI to children poses unique challenges. The researchers note that current educational practices mostly rely on “plugged-in” activities involving coding software, robotics, or web-based AI tools. While these foster practical familiarity, they can overwhelm younger learners unfamiliar with computational logic.

The authors introduce a mixed approach grounded in embodied learning, where physical movement and sensory experience reinforce conceptual understanding. Their study applied this through unplugged role-play sessions before digital programming lessons. For example, children acted out the parts of an autonomous car, drivers, sensors, and obstacles, to simulate how AI systems process information before turning to block-based programming in the later phase.

The researchers designed the course using the “Five Big Ideas” of AI literacy promoted by the AI4K12 initiative: perception, representation and reasoning, learning, natural interaction, and societal impact. Their goal was to strengthen both AI self-efficacy, students’ belief in their ability to work with AI, and perception of AI, the depth of their understanding about how AI functions and affects daily life.

The experiment: Learning in three phases

The nine-week intervention involved 56 fourth graders split into two classes, an experimental group and a control group. Both began with two shared lessons introducing AI concepts. Next, the experimental group completed four unplugged sessions, while the control group worked entirely on digital exercises. Finally, both groups concluded with three identical plugged-in lessons applying their knowledge to machine learning and decision-making tasks.

Researchers used two main measures: an AI self-efficacy questionnaire and a drawing-based assessment. The latter captured how well students could express key AI ideas visually, a developmentally appropriate tool for their age.

Results were revealing. After four weeks of unplugged activities, the mixed group showed a clear boost in AI self-efficacy. Students expressed greater confidence and enthusiasm toward AI, supported by embodied, game-based learning that promoted teamwork and motivation.

However, once they moved to plugged-in sessions, their confidence temporarily dropped. The researchers attribute this to the abrupt transition to block-based programming, which many found technically demanding. Meanwhile, the control group—learning exclusively through digital tools—showed a gradual increase in self-efficacy, but their conceptual understanding developed more slowly.

When the final post-tests were conducted, the experimental group outperformed the control group on AI perception and overall literacy, demonstrating deeper comprehension across all five core AI concepts.

Implications: Mixing methods for stronger foundations

According to the study, sequence and timing are key to AI learning effectiveness. Starting with unplugged, interactive sessions allows children to build a mental model of AI before transitioning to technology-based tasks. This sequencing reduces cognitive overload, fosters inclusion among diverse learners, and helps anchor abstract concepts in concrete experience.

The findings support a growing body of evidence linking physical engagement and embodied cognition with improved understanding in computational education. Students in the mixed group developed not only stronger conceptual awareness but also higher ethical sensitivity toward AI’s societal implications, a dimension often overlooked in early education.

Researchers recommend that primary educators strategically blend unplugged and plugged-in modules, ensuring that unplugged activities precede digital instruction. This design helps children bridge the gap between everyday intuition and technical abstraction, an approach that may also enhance digital citizenship and critical thinking in later schooling.

The authors note some limitations too. The experiment was conducted in a single urban school with a small sample size, restricting generalizability. The study also captured only short-term outcomes. Future research should include larger, multi-school trials and long-term tracking to determine whether early AI literacy training influences students’ STEM engagement later on.