“AI sandwich” is a term that we’re seeing more often these days as a way to describe human/AI collaboration. The idea is to layer collaboration with Generative AI tools with your own creative and intellectual work in a way that maximizes the strengths of both.
The interesting thing is that these two examples from the field of Education talk about it quite differently:
“Use AI tools for the beginning and end of an assignment, with the middle being grounded in human knowledge and expertise.” AI Sandwich – The AI Pedagogy Project – MetaLAB (at) Harvard
“Start with human inquiry, see what AI produces, and always close with human reflection, human edits, and human understanding of what was produced. … ” Human-centred AI Guidance – Washington State Dept of Education
In the first case, the approach looks like this:
- AI as brainstorming partner Brainstorm research questions on a given topic.
- Human knowledge creation Gather real world research data from multiple sources.
- AI as writing partner Use GenAI to help organize and polish the final product.
In the second case, the approach looks like this:
- Human Input: Humans provide initial insights, creativity, and context.
- AI Processing: AI takes over to analyze data, perform complex calculations, and generate potential solutions.
- Human Refinement: Humans review and refine the AI-generated outputs, adding final touches and ensuring alignment with broader goals.
Clearly, there are many ways to build a sandwich.
Generative AI is fundamentally a sophisticated prediction machine. At its core, it works by using complex mathematical algorithms to predict the most likely next word, image element, or piece of content based on the patterns it has learned from massive training datasets, like an extremely advanced autocomplete system. Just as your phone’s keyboard suggests the next word based on your typing history, Generative AI predicts content by analyzing statistical patterns in its training data. It doesn’t truly ‘understand’ or ‘reason’ in the way humans do but instead calculates probabilities with incredible speed and complexity. This means the AI’s output is always fundamentally a sophisticated form of pattern recognition and statistical inference, not true comprehension or creativity in the human sense. The quality and accuracy of its predictions depend entirely on the breadth, depth, and quality of its initial training data.
The key philosophy behind the ‘AI sandwich’ is that Generative AI is a powerful tool, but it works best when humans provide strategic direction and critical oversight – using GenAI to enhance human creativity and problem-solving rather than letting it operate in complete isolation. This approach helps mitigate somewhat limitations like the hallucinations, bias, or lack of nuanced understanding that can occur in GenAI outputs, while still leveraging its speed and computational abilities.
Is this useful in your field? Should your students learn how to do this? This depends on many things including long-term, equitable access to these tools and the ongoing evolution of their use in your field. Much of the hype surrounding these tools is focused on a future that is very difficult to predict. But we can stay curious and critically engaged with what’s developing and share with our students what we see.
Interested in a deeper dive? Check out these resources:
The AI Pedagogy Project: The AI Pedagogy Project – metaLAB (at) Harvard – Creative and critical engagement with AI in Education
Human-centred AI: Comprehensive AI Guidance Accessible Format – Comprehensive guidance on the use of AI in Education
Recent Comments