TL;DR
This paper analyzes Hour of Code (HoC) activities designed for K-12 students to introduce AI and ML. It finds that while these activities are rapidly increasing, they often focus on perception and machine learning, with limited coverage of other core AI concepts and a tendency towards "telling" rather than "doing." The authors suggest improvements for more comprehensive and engaging AI/ML education.
What Problem Does This Paper Address?
As AI and ML become increasingly integrated into everyday life, there's a growing need for AI literacy across all age groups, especially for K-12 students. However, current K-12 computing education often neglects AI/ML concepts. While initiatives like Code.org's Hour of Code (HoC) offer short, accessible introductions, it's unclear how effectively these activities cover the fundamental "big ideas" of AI and machine learning. This paper aims to bridge this gap by examining the content and pedagogical approaches of HoC activities specifically focused on AI/ML, identifying areas for improvement in introductory AI/ML education.
Methods
The researchers conducted a content analysis of HoC activities. They used a deductive coding approach based on Touretzky et al.'s (2019) five big ideas of AI (perception, learning, natural interaction, representation and reasoning, societal impact) and Touretzky et al.'s (2023) five aspects of ML. They also inductively coded societal impact topics and instructional methods (hands-on vs. telling). The analysis involved three researchers collaboratively coding the activities to ensure consensus.
Datasets
Not discussed in the paper. The study analyzed publicly available Hour of Code activities.
Key Insights and Ideas
- Rapid Growth, Uneven Coverage: The number of AI/ML-related HoC activities has surged dramatically since 2021, but the coverage of AI's "big ideas" remains uneven. Perception and learning are frequently addressed, while representation, reasoning, and natural interaction are less common.
- "Telling" Over "Doing": A significant portion of HoC activities rely on videos or teacher explanations ("telling") rather than hands-on engagement ("doing"). This is a missed opportunity, as research suggests project-based learning is crucial for novice learners.
- Misleading Labels: Some activities labeled as AI/ML by HoC actually contained little to no AI/ML content, potentially misinforming students, educators, and parents.
- Societal Impact on the Rise: There's a notable increase in activities addressing societal impact, a positive trend compared to earlier HoC research. However, the focus often remains on high-stakes issues like self-driving cars, potentially overlooking more everyday algorithmic biases.
- Technosolutionist View on Environment: When environmental impact is discussed, it's often framed through a "technosolutionist" lens, potentially ignoring the significant energy consumption of AI/ML models.
- Accessibility Challenges: Some activities, despite being labeled as introductory, require substantial programming background, making them inaccessible to true novices.
Key Results
- Activity Growth: From only six AI/ML activities in 2021, the number grew to 47 by April 2024.
- Big Idea Coverage: Among activities addressing at least one big idea, perception was most common (83.33%), followed by learning (75%), natural interaction (41.67%), societal impact (41.67%), and representation/reasoning (13.89%).
- ML Aspects: 75% of activities incorporated learning, with 17 activities discussing how training data shapes model behavior. Less frequently, activities addressed learning algorithms (9 activities).
- Societal Impact: Fifteen activities (41.67%) addressed societal/environmental impact, with eight focusing on privacy issues and eight on harmful biases.
- Hands-on vs. Telling: While 30 activities addressed perception, many did so through videos or explanations rather than hands-on tasks. Similarly, learning was often "told" rather than "done."
Limitations and Future Work
The authors acknowledge the limitations of their exploratory study, particularly the small sample size of AI/ML-specific HoC activities. They also note that the "one hour" constraint of HoC activities naturally limits the depth of engagement. Future work should move beyond content analysis to observe and evaluate how these activities are actually implemented with students.
The authors recommend several improvements for future HoC activities:
- Broaden AI Ideas: Cover the full spectrum of AI ideas, not just perception and learning, and clearly define what AI is and isn't, especially regarding generative AI.
- Offer Unplugged and Collaborative Activities: Repurpose existing unplugged activities and encourage collaborative designs to reflect the social nature of AI development.
- Go Beyond Data-Centric Focus: Develop approaches that help students understand learning algorithms, not just data.
- Integrate Societal Impact: Weave societal impact into the core AI/ML concepts rather than treating it as a separate topic, and address everyday algorithmic biases.
- Develop Accessible Tools: Create more accessible tools for hands-on AI/ML model building and exploration, aligning with the "Creativity with AI" theme.
Summary updated