AI and digital technology in education is a key research area for LIRNEasia. We are therefore keen to study cutting-edge research and best practices, and to translate these insights into policy and practice in Sri Lanka.
In Journal Clubs, we take an in-depth look at a piece of existing literature to inform our research. On the 25th of August 2025, we evaluated the report titled ‘Understanding the Impacts of Generative AI on Children’, published by the Alan Turing Institute (ATI) in 2025.
The research consisted of:
- Quantitative: Surveyed the perceptions and experiences around Gen AI by:
a) Children and their parents or carers using a nationally representative survey with a sample size of 780 children aged 8-12.
b) Teachers using a survey of 1,001 teachers working with students aged 1-16. This sample was not fully representative, but a quota of 76% female and 24% male was applied to reflect the gender make-up of England’s teaching workforce.The surveys consisted of multiple-choice and free-response questions.
- Qualitative: Consisted of two 3-day workshops which observed how children interacted with Gen AI and recorded their feedback. Both workshops were held in state-funded schools which worked with Children’s Parliament, ATI’s research partner.
During the workshops, researchers provided a series of interactive presentations on: (1) children’s rights in relation to technology; (2) AI and how it works; (3) online safety; and (4) how to use Gen AI for art. The children then participated in activities such as creating an artwork that represented their future self, illustrating how they feel about AI, and creating a zine about the ways Gen AI should be used. Children were given the freedom to choose to use traditional art materials, Gen AI, or both, as they wished.
Key Findings
Gen AI could have a significant impact on children with additional learning needs, who generally used it more frequently, including to assist with socialization. Some children of black and minority ethnic groups found that Gen AI outputs did not represent them, which caused frustration. In the surveys, there were significant differences in usage between private and state-funded schools, income brackets, and region, but not between gender. Many teachers reported using Gen AI, and the majority of them felt that it increased their productivity; however, they were concerned that the same technology, when used by children, could reduce their creativity and critical thinking, which was a concern echoed by parents and carers. During the workshops, many children preferred to use traditional art materials and gave reasons like enjoying the tactility of that medium. Some children felt that Gen AI allowed them to produce more complex artwork than they could have in an analogue form.
Discussion
We recognized that the impact of Gen AI on children in the UK could be very context specific; therefore, the findings would have limited transferability to our region of the world. For this reason, in looking at this report, our focus was on the research methodology and experiences and using it to inform our own research.
1) Research that overemphasizes safety will not produce actionable insights
The ATI and their collaborators placed a heavy emphasis on safeguarding children during their qualitative work. For example, children were not allowed to directly enter prompts into a Gen AI tool. Instead, they gave the prompts to a researcher, who typed it themselves and screened the output before presenting it to the students. This was intended to prevent children from being exposed to potentially harmful images. Some of us felt that this limited the relevance of the findings since it created a barrier which may have affected children’s willingness to use Gen AI and did not reflect how they would interact with the tool in the real world. These kinds of criticisms were levied at several of the safeguarding protocols used in the qualitative research; however, it was also recognized that when dealing with children as research subjects for experiments, it is a good practice to err on the side of caution.
2) Other experimental methods
Meanwhile, the limitation with surveys is that they capture perceptions rather than measure the magnitude of an impact, which may not be very useful when trying to operationalize a technology. One of our staff suggested that a randomized controlled trial might have been more effective, but it was noted that these findings are difficult to scale in a context specific domain like education.
3) Teacher-focused deployment of AI in education
Another suggestion was that the most effective way to mobilize new technology in education is to focus on the teachers. They could be given clear targets, such as producing certain foundational literacy scores or exam scores and given the freedom to design an educational approach that fits their unique contexts. As an intervention, teachers could be provided with Gen AI embedded resources and appropriate training to assist their ongoing responsibilities. Rather than designing a highly localized experiment, it may be more useful to document how different educators use Gen AI to improve learning outcomes and publish these findings, so they can help others develop their own relevant pedagogical approaches.
Overall, there was a concern that the study focused too heavily on recording the perceptions of teachers, children and carers, rather than discovering how AI tools could be used to improve learning outcomes.
By Anish Fonseka (Junior Researcher, LIRNEasia)
Anish is a member of the Data, Algorithms and Policy (DAP) team at LIRNEasia – which participates in the policy dialogue around our algorithmically inclined society by conducting research and developing data science solutions.
Find the slide set of this journal club below.
Download PDF Email
Comments are closed.