Beyond Digital Outsourcing: Crafting Authentic Education in AI-Enhanced Learning [Research Design Outline]
I’ve been wrestling with a question: how do we integrate Generative AI into education without short-circuiting the actual learning process? It’s a tricky balance – these AI tools are incredibly powerful at creation (think top of Bloom’s taxonomy), but what happens if students skip straight to the end, missing all those crucial learning steps along the way? GenAI has the potential to short circuit our current learning process – and in doing so it wont effect just the uptake of [intellectual] knowledge, but have side effects on the personal and social development that education forms too.
Research Context and Objectives
I’m focusing on sixth form students (16-18 year olds) because they’re at such an interesting stage – mature enough to engage critically with AI but still developing their learning strategies. Plus, having read Holmes and Tuomi’s work on Biesta’s educational dimensions [qualification (knowledge), subjectification (personal growth), and socialization (social integration)], I’m convinced we need to look beyond just knowledge acquisition. We need to consider how AI affects personal growth and social development too.
I’m not yet sure of the methods I’ll use, but my idea so far is along these lines:
Either qualitative or mixed methods.
Comparative Analysis: Testing different pedagogical approaches to AI integration across a few classes, with documentation of both successful and unsuccessful strategies from the perspectives of educators and learners.
Technical Implementation: Deploying open-source local Large Language Models to ensure data privacy and security – a crucial consideration in educational settings. This approach allows for customization while maintaining ethical standards in student data handling. If this is not possible then a secondary approach would be building the interface for making API calls to the appropriate LLM. If this too is not possible then we would have to use off-the-shelf LLMs (like Claude or Perplexity).
Data Collection Methods:
– Surveys and semi-structured interviews with learners and educators.
– Teacher observations and feedback sessions.
– Assessment of learning outcomes through academic measures.
– Discussions with the Personal and Social Development team in the school to evaluate possible changes in the students.
Personal Development and Training to support Research
To carry out the research I’ll draw on my technical experience in local LLM development and deployment, complemented by ongoing exploration of education and pedagogy through my study program here. Since this is an area this is constantly developing and unfolding (at quite a fast pace) I plan to engage with emerging research in particular by attending conferences (EFI has a number of conferences dealing with both AI and Education this semester with the Learning Curves events), and following published papers and work of researchers in the field. (Some I have currently identified are: Gert Biesta, Wayne Holes, Rose Luckin, Tristan Harris, Ilya Sutskever, Andrej Karpathy, Andrew Ng)
I need to do some training to develop my knowledge of, and skills in, research methods.
Through the research I hope to develop:
- A framework for meaningful AI integration that supports rather than shortcuts learning
- Technical guidelines for implementing local LLMs in educational settings
- Practical recommendations for teachers and administrators
Other Considerations:
The project’s emphasis on local LLM deployment also addresses growing concerns about data privacy and institutional autonomy in educational AI applications, potentially offering a model for other institutions to follow. Instead of using off-the-shelf AI tools, I’m planning to deploy open-source, open-weights, Local Language Models. In so doing I’m emphasising that educational data privacy matters (for both educators and learners), and this way, we keep everything in-house. Plus, we can tweak the AI to suit our needs.