Teaching Tech to Talk: K-12 Conversational Artificial Intelligence Literacy Curriculum and Development Tools

Jessica Van Brummelen, Tommy Heng, Viktoriya TabunshchykAAAI 2021 · DOI

Abstract

Researchers from MIT have developed a new curriculum and a specialized interface for MIT App Inventor designed to teach K-12 students the inner workings of conversational AI. By moving students from building simple rule-based bots to machine-learning-powered Alexa skills, the study demonstrates that middle and high schoolers can master complex AI competencies, though concepts like algorithmic bias and machine learning generalization remain significant pedagogical hurdles.

Introduction

While children interact with AI-driven agents like Siri and Alexa daily, these technologies remain "black boxes" to them. Most existing K-12 AI education focuses on image classification or basic logic, leaving a gap in understanding Natural Language Processing (NLP) and the societal implications of conversational agents. Furthermore, few tools have been rigorously evaluated against established AI literacy frameworks. This research addresses these gaps by creating a structured 5-day curriculum and a development tool that allows students to build functional Alexa skills without needing professional-grade coding environments.

Methodology and Technical Approach

The researchers conducted remote workshops with 47 students (grades 6-12) and 9 teachers. The core technical innovation was the Conversational Agent Interface for MIT App Inventor, a block-based coding environment for defining "intents" (user goals), "utterances" (user phrases), and "slots" (variables).

Block-based interface for defining intents and slots in the Conversational Agent Interface.

The 5-day curriculum followed a scaffolded approach:

  1. Day 1: Building rule-based agents to highlight the limitations of non-AI systems.
  2. Day 2: Introduction to the "Big 5 AI Ideas" and ethics.
  3. Day 3: Transitioning to Machine Learning (ML) and data representation via "slots."
  4. Day 4: Connecting mobile apps to the cloud (Alexa) using CloudDB.
  5. Day 5: Final project development and presentation.

Key Insights: From Rules to Machine Learning

A pivotal moment in the curriculum was the transition from rule-based to ML-powered agents. By first building a "dumb" bot that failed on simple variations (e.g., "Hi" vs. "Hello"), students immediately recognized the value of ML generalization.

Students experienced "aha" moments during transfer learning, particularly when Alexa understood phrases they had not explicitly coded. However, the study also identified challenges. Students struggled with Competency 12 (Learning from Data); when asked if an AI would recognize a phrase similar to its training data, responses were split 50/50. This indicates that the "black box" of ML generalization remains difficult to visualize. Additionally, students exhibited an "optimism bias," identifying positive AI impacts far more frequently (71%) than negative ones (6%), suggesting a need for more adversarial and ethical critique in future curricula.

Results and Student Outcomes

The curriculum led to significant knowledge gains. Understanding of "What is AI" (Competency 2) and "Decision-Making" (Competency 8) both approximately doubled in post-tests. Teacher engagement ratings were high (4.6/5), and 86% of student responses regarding technological investigation aligned with the "Big AI Ideas" framework.

Examples of student-developed projects: a sign language learning tool and an 'AI Nurse' diagnostic app.

Students successfully created 41 unique projects, demonstrating both technical proficiency and creative application. Projects ranged from medical diagnostic tools like "AI Nurse" to educational aids for learning sign language.

Conclusion and Future Directions

While successful, the study's small sample size and remote format present limitations. Future work will focus on adapting the curriculum for younger (K-5) students, improving the teaching of complex concepts like ML generalization, and exploring scalable delivery methods such as MOOCs.

Summary updated

BibTeX
@inproceedings{brummelen2021teaching,
  author = {Jessica Van Brummelen and Tommy Heng and Viktoriya Tabunshchyk},
  title = {{Teaching Tech to Talk: K-12 Conversational Artificial Intelligence Literacy Curriculum and Development Tools}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/aaai.v35i17.17844},
}
Figure