Competence domain of the Didactic Framework: Foundational AI knowledge

By the end of this activity, students can… 

• explain “model training” in the machine learning context and describe training procedures. (FLAIR Didactic Framework: LO2) 
• explain what type of data is used in model training and its role in AI. (FLAIR Didactic Framework: LO3) 
• define the algorithmic bias and explain how it relates to AI limitations. (FLAIR Didactic Framework: LO5) 
• examine how different data inputs alter results. 
• assess the potential for bias in specific datasets. 

Before class, students complete a short preparatory task by writing brief definitions of key concepts such as “data”, “types of data”, “personal data”, and “uses of data”. This establishes a baseline understanding and prepares them for in-class discussion. 

In class, students share and compare their definitions. Provide a brief lecture on how machine learning works, focusing on the relationship between training data and model outputs. Emphasise that models do not “understand” the world, but reproduce patterns present – or absent – in the data they are trained on. 

Students form small groups of 3–5 members. Introduce a large dataset and assign different subsets of the data to each group (e.g. data from different regions or demographic groups). Groups use their assigned dataset as input for the machine learning tool/model. The datasets are intentionally varied so that groups can later compare how training data influences model behaviour. 

Example: All groups train a model to classify images with the labels male and female. One group uses mostly images of younger adults, another uses images of older adults, and a third uses a limited and very similar image set.  

After training, groups test their models using the same images across the different models and observe where predictions differ or fail. They document accuracy, errors, and unexpected patterns. 

Students compare results across groups to see how “models” behaved differently based on the individual input. Each group briefly presents their findings to the class. The class collaboratively links these discrepancies to concepts of algorithmic bias and explainability. Students further reflect on how their understanding of data has changed compared to the beginning. 

Possible challenges

• Students might struggle to visualize “data training” without a technical background. 
• Students may find it difficult to interpret differences in results across groups  

How to adress them

• Use visual analogies (e.g., teaching a child to recognize cats using only pictures of black cats). 
• Guide students with targeted questions to support interpretation 

• Machine learning Tool
• Google Teachable Machine https://teachablemachine.withgoogle.com 
• Self-Guided tutorial – The “Bananameter” Walkthrough: https://medium.com/@warronbebster/teachable-machine-tutorial-bananameter-4bfffa765866  
• Lesson Plans & Classroom Activities 
• Hello World: Accuracy and Ethics Activity: https://www.raspberrypi.org/blog/classroom-activity-machine-learning-accuracy-ethics-hello-world-18/  
• MIT RAISE (Responsible AI for Social Empowerment): https://raise.mit.edu/  
• Shark AI (University of Florida/NSF): https://www.nsf.gov/news/preparing-science-educators-use-teach-ai-classroom  
• Faculty-Specific Guides 
• Gemini for Faculty (CMU Guide): https://www.cmu.edu/computing/services/ai/learn-connect/files/gemini-faculty.pdf  
• Demystifying AI for Students: https://www.saskoer.ca/etad873streamlinedinstructionaldesignwithai/chapter/google-teachablemachine/