Abstract
Artificial Intelligence (AI) technologies are becoming ubiquitous, transforming the workforce by altering or creating jobs and influencing decisions that affect minority communities. The necessity of AI literacy—comprising the knowledge and skills for critical interaction with AI—is increasingly important. Multilingual learners, who engage with both everyday and domain-specific vocabulary and syntax, must acquire AI literacy, necessitating tailored pedagogical practices. This article outlines three strategies implemented for multilingual middle school students during a summer camp in Southern California. We argue for a mutually beneficial relationship between acquiring second language proficiency and AI literacy. Supporting these processes, we present three pedagogical strategies: (1) storytelling to model AI decision-making, (2) role-playing as an AI to demonstrate programmability and learning from data, and (3) programming text-to-speech-to-text AI to illustrate sensor functionality and action-reaction concepts. These strategies potentially foster linguistic scaffolding and translanguaging, aiding multilingual learners in acquiring new literacies.
Introduction: Addressing the AI Literacy Gap
As AI becomes embedded in everything from job applications to social media, AI literacy has transitioned from a technical elective to a critical survival skill. However, minoritized and multilingual learners often face a "double hurdle": they are frequently underserved by standard technology curricula and may be more vulnerable to the biases and misinformation inherent in AI systems.
Traditional AI education rarely accounts for the unique "linguistic repertoires" of multilingual students. This research addresses this gap by proposing that the process of learning a new language—navigating syntax, vocabulary, and translation—is structurally similar to understanding how AI processes data. Instead of viewing a student’s native language as a barrier, this framework treats "translanguaging"—the fluid use of multiple languages—as a powerful tool for teaching computational logic.
Pedagogical Framework
The study details a learning design implemented during a two-week summer camp in Southern California for over 40 middle school students, 90% of whom were Latine and Spanish-speaking. The authors developed three core pedagogical strategies to demystify conversational agents like Siri and ChatGPT:
1. Storytelling via Decision Trees
To explain AI decision-making, instructors used "Choose Your Own Adventure" narratives. Students mapped out "if-then" logic (e.g., If the user says "Oye Siri," then activate) using visual decision trees. This bridged the gap between human reasoning and machine logic, allowing students to see the algorithmic structures underlying everyday interactions.
2. Embodied Role-playing
To teach "Programmability" and "Learning from Data," students acted as the AI themselves. One student created a "database" of prompt-response pairs (the training phase), while their partner tried to memorize and recall them (the testing phase), eventually calculating their own "accuracy percentage." This physicalized the concept of a model being only as good as its training data.
3. Block-based Programming (Scratch)
Students built a functional translation pipeline using Scratch. They created a program that captured audio (Sensors), converted it to text, translated it, and vocalized the result (Action-Reaction). This allowed them to see the "modality conversion" happening under the hood, transforming abstract concepts into tangible code.
Key Insights and Results
The implementation revealed that translanguaging serves as a vital cognitive scaffold. Allowing students to switch between English and Spanish helped them grasp abstract logic; for example, comparing the English "If/Then" to the Spanish "Si/Haga" reinforced the universality of algorithmic structures.
By role-playing the training and testing phases, students moved away from personifying AI as a "magical being" and began to see it as a system limited by its database. Conversational agents served as the perfect "hook" because multilingual learners are already highly tuned to language nuances. Students were particularly engaged when questioning why an AI might struggle with specific accents or translation inaccuracies, leading to critical discussions about AI ethics and biased data.
While this was a narrative account of a pilot program, the qualitative results were promising. Students successfully navigated complex computational concepts by relating them to their own bilingual identities, transitioning from technology users to technology designers.
Limitations and Future Work
This article serves as a narrative description of a pilot program rather than a formal empirical study with a control group. Future work will focus on:
- Quantitative Evaluation: Systematically measuring the impact of these strategies on long-term AI literacy and language proficiency.
- Scaling the Curriculum: Investigating how these "unplugged" and "plugged" activities perform in diverse, formal classroom settings.
- Translanguaging Dynamics: Conducting deeper research into how specific linguistic shifts (code-switching) during technical tasks aid the retention of computational concepts.
Reference: Ojeda-Ramirez, S., Ritchie, D., & Warschauer, M. (2024). AI Literacy for Multilingual Learners: Storytelling, Role-playing, and Programming. The CATESOL Journal, 35(1).
Summary updated