# Exploring Machine Teaching with Children

**Authors:** Utkarsh Dwivedi, Jaina Gandhi, Raj Parikh, Merijke Coenraad, Elizabeth Bonsignore, Hernisa Kacorri  
**arXiv:** [2109.11434](https://arxiv.org/abs/2109.11434)  
**Date:** 2021-09-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2109.11434v1/

## Key points

- Use of machine teaching interfaces to help children develop creativity, flexibility, and comfort with ML and AI
- Study conducted with 14 children aged 7-13 years and adult co-designers
- Children trained to build and test image classifiers using ML models
- Findings shed light on how children reason about ML concepts
- Visibility of ML metrics (confidence scores) for experimentation is important
- Exchanging models promotes reflection and pattern recognition
- Quick data inspection enhances understanding of data's influence on model training and performance
- Thematic analysis of over 5 hours of video recordings from sessions
- Insights contribute to designing effective machine teaching experiences for children

## Summary

The study titled "Exploring Machine Teaching with Children" focuses on the use of machine teaching interfaces to help children develop creativity, flexibility, and comfort with machine learning (ML) and artificial intelligence (AI). The research team worked with a group of 14 children aged 7-13 years, along with adult co-designers, to explore how children engage with ML concepts. During the study, the children were trained to build and test image classifiers using ML models. They also tested each other's models for robustness. The findings of the study shed light on how children reason about ML concepts and provide valuable insights for designing machine teaching experiences specifically tailored for children. One key insight from the study is that ML metrics, such as confidence scores, should be visible for experimentation. This visibility allows children to explore and understand how different metrics affect the performance of their models. Another important finding is that ML activities should enable children to exchange models with each other. This promotes reflection and pattern recognition as they compare and analyze different approaches. Additionally, the study emphasizes the importance of quick data inspection in the machine teaching interface. For example, allowing children to inspect images or gestures quickly can enhance their understanding of how data influences model training and performance. To gather data for analysis, researchers recorded over 5 hours of videos from the sessions and analyzed them using thematic analysis techniques. Codes were induced through open coding and deductive analysis was performed based on a priori codes related to machine learning and teaching. Overall, this research contributes valuable insights into designing effective machine teaching experiences for children. By incorporating these insights into educational tools and interfaces, we can empower young learners to develop essential skills in creativity, flexibility, and comfort with ML and AI technologies.