Introduction
As Artificial Intelligence (AI) becomes ubiquitous, the need for AI literacy in K-12 education has never been more pressing. The AI4K12 initiative identifies "Representation and Reasoning" as two of the five "Big Ideas" in AI. However, teaching these abstract concepts often relies on digital tools that require expensive hardware and stable internet access, potentially widening the "digital divide" for rural or underfunded schools. To address this, researchers at North Carolina State University developed "Guess Whose Face," an unplugged board game that teaches complex AI concepts through physical, hands-on play.
Game Design and Mechanics
"Guess Whose Face" is a competitive board game designed for middle school students (Grades 6-8). It deconstructs the facial recognition pipeline into a collaborative activity between two roles: the Drawer (acting as the Feature Extractor) and the Guesser (acting as the Inference Engine).
The game mimics real-world machine learning through several key mechanics:
- Feature Extraction: The Drawer receives a target face card and must communicate its identity to the Guesser by drawing dots on specific landmarks (eyes, nose, mouth) using tracing paper.
- The Spinner and Data Quality: A central spinner introduces variability. Landing on "Feature Dots" allows for clean data extraction, while "Noise" simulates sensor interference by requiring the Drawer to add random, non-informative dots.
- Chance Cards: These cards simulate algorithmic advantages, such as "Free Guesses" or "Landmark Discoveries," which reveal specific facial structures to the Guesser.
Pilot Study and Key Insights
The game was tested with a group of 10 rural middle school students during a summer camp. The study aimed to evaluate engagement and the effectiveness of the unplugged approach in demystifying AI's "black box."
Demystifying Representation
By physically tracing dots instead of using verbal descriptions, students intuitively learned that AI systems do not "see" a person in the human sense. Instead, they process structured mathematical or spatial representations. This shift from holistic perception to data-driven reasoning is a fundamental step in AI literacy.
Understanding Noise and Salience
The inclusion of "Noise" was a critical learning objective. Students observed firsthand how imperfect data complicates the reasoning process. Furthermore, teams began to develop strategies for "feature engineering," prioritizing salient features—like a unique nose shape—to optimize the Guesser's performance with minimal data points.
Results and Future Directions
The pilot study results were overwhelmingly positive. Students reported high levels of enjoyment (4.625 / 5) and perceived learning (4.375 / 5). Qualitative feedback indicated that students could accurately describe AI as a system that "scans" and "captures" data points to compare against a database.
Future work will focus on scaling the game for broader classroom use, including replacing custom components with standard dice and integrating ethical discussions regarding the misuse of facial recognition and algorithmic bias. By making AI concepts tangible and accessible, "Guess Whose Face" provides a scalable model for inclusive AI education.
Summary updated