Science1 publisher2 min readPublished
Two education researchers rebuild AI literacy around four questions students learn to ask
Golnaz Arastoopour Irgens and Sarah Burriss want students asking where an AI tool's data came from and who built it, inside lessons teachers already teach, and their paper offers the design without a classroom test.
The Scientist · Science desk
What happened
- Vanderbilt's Golnaz Arastoopour Irgens and Sarah K. Burriss, of the College of Connected Computing, published their AI literacy framework in the journal Educational Researcher.
- They say their four-part approach differs from existing AI teaching methods, which in their account prioritize technical skills and may carry only a single lesson on ethics.
- Students are taught to ask where a tool's data came from, who designed it, whether it represents them, and whether it is helping them achieve their own goals.
- The paper arrives with the 2025 PISA results, which report that students globally are becoming less creative in how they use digital technologies.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- capability The making component has students build an AI-related object, which puts production back into a curriculum the PISA reading says is drifting toward consumption.
- constraint A framework designed to fit inside any lesson has no timetable slot and no single owner in a school, so uptake rests on what individual subject teachers choose to do.
- decision A school that adopts this is agreeing that a student may decline to use an AI tool and still be completing the lesson.
The case for embedding rests on transfer. "There's a lot of evidence about not just learning things in isolation, but how different topics can connect for students," Golnaz Arastoopour Irgens said [8]. The four components are being, knowing, acting and making, and a teacher can enter at whichever one fits the class [3]. At its simplest level, she said, the work is "questioning, critiquing and imagining" [16].
Refusal is one of the permitted answers. "It's important that we give children pathways to reflect on whether they want to use these tools," Arastoopour Irgens said [20]. She described a component built around "giving them the agency to choose in what contexts they want to use AI" [13]. Sarah K. Burriss set the ambition higher. "Our framework asks adults to facilitate these conversations and critical exploration and design of AI, inviting young people to participate in one of the most consequential issues of our lives," she said [14].
The published account sets out concepts, guiding questions and classroom entry points. It reports no classroom study and no measure of what students do differently afterwards [17]. PISA comes in through timing: the framework appeared as the 2025 results came out [9]. The characterisation of the trend on the record belongs to Andreas Sleicher, the OECD's director for education and skills. In a webinar announcing the findings, he said: "It's a troubling shift from creating to consuming, from producing to scrolling" [10].
Someone could test the embedding claim with a fairly ordinary design. Take two cohorts through the same subject content, one with the four questions woven through the term and one with a single ethics lesson. Then hand both an unfamiliar tool. See who asks who built it and what it was trained on. A study like that would also show whether the questioning survives outside the classroom, which is the behaviour the PISA reading is worried about [9]. The four parts came out of guidelines that the two researchers' own teams already use [21]. "We wanted to help people expand the idea of ethics in AI literacy and make it accessible," Arastoopour Irgens said [12].
What to watch
- An intervention study comparing embedded questioning with a standalone ethics unit, with the outcome measured on what students do with an unfamiliar tool.
- The full PISA 2025 tables, which would put a size and a location on the decline in creative digital use that Sleicher described.
- Whether teacher training providers or districts pick up the four components in non-computing subjects, where the framework says the work belongs.