Using Cloud Tools for Literate Programming to Redesign an AI Course for Non-Traditional College Students

Maria L. Hwang, Calvin WilliamsonAAAI 2020 · DOI

Introduction

As artificial intelligence and machine learning (AI/ML) become increasingly relevant across all industries, the need to teach these technical subjects to non-computer science (CS) majors has grown. However, students at trade schools, community colleges, and non-technical universities often face unique barriers to entry. This paper details a pedagogical redesign at the Fashion Institute of Technology (FIT) that transitioned an AI course from traditional local software environments to a cloud-based, "literate programming" workflow. By leveraging automated pipelines and serverless infrastructure, the authors created an accessible, collaborative, and scalable educational model for non-traditional learners.

The Challenges of Traditional Technical Instruction

The authors identified four primary "friction points" that hindered student success in a traditional statistics and machine learning curriculum:

  1. Infrastructure Barriers: Campus IT departments frequently restrict software installations. This often forced students to run tools like R from USB drives, leading to performance issues and configuration errors.
  2. The "Blank Page" Syndrome: For beginners, a standard coding console is intimidating. Without a structured environment, students struggle to document the logic behind their code or save their work effectively.
  3. Collaboration Gaps: Traditional workflows rely on local files and manual submissions (e.g., via Dropbox or email). This makes real-time peer debugging and instructor feedback nearly impossible.
  4. Student Demographics: Non-traditional students—including adult learners and commuters—require a consistent environment that works across various devices, including library computers and Chromebooks, without complex setup procedures.

Methodology: The Literate Programming Pipeline

To address these hurdles, the course was redesigned around the concept of literate programming—a method where executable code is interspersed with natural language explanations. This approach shifts the focus from rote syntax to computational literacy.

The Technical Workflow

The authors developed a "single source of truth" pipeline to streamline content delivery:

  • Authoring: Instructors write all course materials in RMarkdown (Rmd).
  • Automated Conversion: Using a custom toolset leveraging R’s knitr and rmarkdown packages alongside Pandoc, the Rmd files are automatically converted into multiple formats, including HTML, PDF, and MS Word.
  • Cloud Notebooks: Crucially, the pipeline converts the materials into .ipynb (Jupyter/Colab notebooks). A Makefile automates the creation of these outputs and handles the upload to Google Colab and Binder.

A New Submission Model

In this redesigned workflow, students no longer submit static files. Instead, they "make a copy" of an instructor’s notebook in Google Drive. Once they complete the assignment, they share the live link back with the instructor. This allows for real-time feedback and grading within the student's active coding environment.

Key Insights and Pedagogical Benefits

"Serverless" Education

By utilizing Google Colab, the course operates in a "serverless" capacity from the university's perspective. It requires zero IT support or local server maintenance, allowing the curriculum to scale across multiple sections instantly without institutional overhead.

Accessibility as Equity

Moving to the cloud ensures that all students have the same computing power and environment, regardless of their personal hardware. This levels the playing field between a student with a high-end laptop and one relying on a library Chromebook.

Real-time Feedback Loops

Because notebooks are shared via Google Drive, instructors can jump into a student's live environment to leave comments or help debug. This mimics a "pair programming" experience, which is particularly valuable for remote or commuting students who cannot attend traditional office hours.

Results and Outcomes

While the paper focuses on qualitative pedagogical improvements, the authors noted several significant outcomes:

  • Elimination of Setup Errors: The common "it doesn't run on my machine" problem was virtually eliminated by the standardized cloud environment.
  • Streamlined Grading: Instructors can grade by clicking a link, restarting the kernel, and running all cells, which is significantly faster than managing disparate local scripts.
  • Improved Documentation: Students were more likely to engage in "active note-taking," resulting in projects that were better documented and more reflective of their conceptual understanding compared to previous years.

Limitations and Future Work

The authors acknowledge that while technical barriers have been lowered, institutional adoption remains a challenge. Future work will focus on encouraging other non-traditional higher education institutions to adopt similar frameworks and continuing to track the efficacy of this serverless approach as AI tools and cloud environments evolve. Additionally, the authors aim to expand the automated pipeline to support a broader range of programming languages beyond R and Python.

Summary updated

BibTeX
@inproceedings{hwang2020using,
  author = {Maria L. Hwang and Calvin Williamson},
  title = {{Using Cloud Tools for Literate Programming to Redesign an AI Course for Non-Traditional College Students}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2020},
  doi = {10.1609/aaai.v34i09.7069},
}
Figure