Tim led a three-hour workshop for a group of 14 - 17 year olds as part of the Good Future Foundation’s AI Summer School: a week-long residential event hosted a Holmewood House School in Kent in parallel to a meeting of education professionals and policy makers in Oxford, both exploring approaches to implement the Department for Educations’ AI Safety Standards.
The session combined elements of the Generative AI in Education Workshop in a Box and the Perspectives on AI course.
Tim reflects: (Quotes are paragraphrased from memory based on student inputs)
When I arrived on Thursday, the group had already visited a data centre, and been to the London HQ of Salesforce, as well as taken part in a number of workshops on aspects of AI and on the DFE Geneartive AI Product Safety Standards. Through video link-ups with the parallel summer school in Oxford, the young participants had already been quizzing policy-makers both on how AI could or should be introduced in education, and what could be done to support students who may have been impacted by unregulated introduction of AI before safety standards are in force.
I kicked off the workshop using a collection of pictures from Better Images of AI - inviting students to select images that resonate with them. It’s an activity I’ve used with a number of groups of adults recently - and it always generates good discussions. I’m currently thinking about how this activity could be adapted as an ice-breaker for Citizens’ Track on AI workshops by:
- Making sure the image set represents a diversity of perspectives on AI;
- Providing facilitator notes on some of the issues that can be brought out in discussion of each image;
We then used an activity on researching the history of AI. I set small groups off using different tools to create a short history of AI suitable for 14 - 17 year olds, including:
- Researching using Wikipedia
- Using ChatGPT
- Using PublicAI.co (Apertus - the Swiss-backed open AI model)
- Agent-based coding (A live-demo I ran using Gemini Antigravity to make a mini-game explaining key concepts from AI history)
I’d hoped to also use NotebookLM (to explore Retrieval Augmented Generation), and a voice assistant, but it was tricky to get log-ins set-up.
I was also pleased with how the activity worked, allowing groups to reflect on the difference between the information gathered from Wikipedia vs. an AI ChatBot, and the difference between different chat bot responses - as well as getting exposure to coding agents. At the same time, it allowed us to explore different branches of AI, and to explore early visions of thinking machines, and past AI Winters. It was interesting to hear different views from students on whether this was a ‘long history’ (“more than 70 years of thinking on AI…“) or a very ‘short history’ (“all this change to our way of life in less than 100 years…“).
We then turned to the ‘tool cards’ from the Generative AI in Education Workshop in a Box resources and the issue sheets. One student questioned whether we should really call AI applications ‘tools’ given their outputs are often unpredictable (“A hammer that randomly flipped over and hit the wrong nail or you 10% of the time would hardly be considered a good tool!”).
I had the chance to test out some slides on ‘Who decides how AI is developed?’ that we had cut out of the Workshop in a Box toolkit due to time constraints, and not being sure how easy it would be for teachers to deliver. In this section, we talk about the decisions that can be made at school and classroom level, the decisions made by EdTech providers, and the decisions made by model developers. Whilst I only had time to skim through the material - and didn’t get chance to then use it in the final activity in any depth, it seemed to work conceptually.
We ended with an attitude-finder activity, responding agree/disgree to a set of statements.
The striking thing here was the almost unanimous agreement with the statement:
(1 AI tools for education should be officially checked and approved as safe, accurate and unbiased before they can be used, even if that means it takes longer for the latest models to be used in schools
and the strength of views about existing cognitive harms that introduction of AI may have done, but then on other statements, a range of opinions that were more in favour of AI systems being used, such as spread of views on:
(2) AI should never replace face to face time with a teacher, even if AI might help a pupil learn information faster
and
(3) Important exams should always be marked by a human, even if an AI marker would be fairer
Discussions threw up questions of who should check that AI is safe and unbiased (“It can’t be left to the teacher, because then teachers in poorer schools might not have the time/resources to check everything properly”), and whether it might be justified to pilot tools at scale to work out the potential issues with them in order to make them safe. I’m left reflecting on how a deliberative process might support participants to explore the consistency of views, and trade-offs in more depth. For example, highlighting the likely scenarios that would arise from implementing statement (1).
Overall, the session was a powerful demonstration of the capacity of students to deeply engage with questions of AI governance, and I’m looking forward to hearing what comes out of the parallel Summer Schools overall, and continuing to develop resources that might support discussions on AI in education.