AI Toolkit: Libraries and Essays for Exploring the Technology and Ethics Behind AI

Levin Ho, Morgan McErlean, Zehua You, Douglas S. Blank, Lisa MeedenAAAI 2025 · DOI

TL;DR

This paper introduces AITK, an open-source toolkit combining Python libraries and interactive Jupyter notebooks designed to make AI concepts accessible to a broad audience, including those without a programming background. By bridging the gap between explanation and hands-on experimentation, AITK aims to foster a deeper understanding of AI's technology and ethical implications, as demonstrated by its successful piloting in humanities courses and positive usability testing results.

What Problem Does This Paper Address?

The rapid advancement and societal impact of Artificial Intelligence (AI) necessitate a greater public understanding of its underlying technologies and ethical considerations. While numerous resources exist, they often fall into two categories: highly technical toolkits requiring significant programming expertise, or purely explanatory content lacking interactive elements. This leaves a gap for individuals, particularly in non-technical fields, who want to explore AI hands-on without becoming expert programmers, and for educators seeking accessible ways to integrate AI topics into their curricula, especially concerning its ethical dimensions.

Methods

AITK is built upon two core components:

  • Python Libraries: These libraries are designed for easy experimentation with key AI frameworks like deep learning and robotics. A key feature is their focus on visualization, allowing users to see the inner workings of these frameworks, such as displaying neural network structures and visualizing activations across layers.
  • Jupyter Notebooks (Computational Essays): These notebooks provide an interactive way for users to engage with the Python libraries without needing to write code from scratch. They consist of cells containing executable code and formatted text, allowing users to run, modify, and understand prewritten code with accompanying explanations. This format transforms complex AI concepts into "computational essays."

The paper also details the piloting of AITK in humanities courses focused on responsible AI and a usability study to assess its accessibility and effectiveness.

Datasets

The paper mentions the use of a small, simple dataset of 6x6 digits for the "Basic Neural Networks" notebook. This dataset is used to train a neural network to classify digits 0-9. Additionally, the paper notes the generation of random images for training and testing, including checkerboard patterns and images that are not digits, to illustrate concepts like generalization and the impact of training data.

Key Insights and Ideas

  • Bridging the Explanation-Experimentation Gap: AITK's core innovation lies in its integration of explanatory text with interactive code execution within Jupyter notebooks. This "computational essay" format makes AI concepts tangible for non-programmers.
  • Visualizations are Crucial for Understanding: The emphasis on visualizing neural network structures and activations (as shown in Figure 1) is a powerful tool for demystifying complex AI models, transforming them from "black boxes" into understandable systems.
  • Accessibility for Humanities Students: The successful piloting of AITK in humanities courses demonstrates its potential to engage students from diverse backgrounds, enabling them to grasp AI concepts and their ethical implications without requiring extensive prior technical knowledge.
  • Addressing Curricular Shifts: AITK is well-positioned to help computer science programs meet new requirements to focus on the societal impacts of AI, offering notebooks that explore bias in data, word embeddings, and image generation.
  • Novice Challenges: While generally successful, usability testing revealed that novices can struggle with basic notebook navigation and execution, highlighting the need for more scaffolding and clearer instructions.

Key Results

  • Pilot Studies: AITK notebooks were successfully piloted in humanities courses at multiple institutions, with students finding them accessible and helpful.
  • Usability Study:
    • Navigation: Approximately half of the participants could navigate to their desired notebook on the first try.
  • Agreement: For all survey statements, at least 75% of participants strongly agreed or agreed.
  • Understanding: 87% of participants agreed that they understood a topic better after using AITK.
  • Recommendation: 100% of participants agreed that they would recommend AITK to others.
  • Qualitative Feedback: Participants found the labs helpful for contextualizing arguments, enjoyed the hands-on nature, and appreciated the ability to save and refer back to notebooks. Some CS students desired more technical depth.

Limitations and Future Work

  • Scaffolding for Novices: The usability study indicated a need for more scaffolding for novice users, particularly regarding error handling and understanding that code execution is sufficient without deep code comprehension. Future work will involve providing more guidance on these aspects.
  • Technical Depth for CS Students: While accessible, some CS students expressed a desire for more technical challenges and less hand-holding in the notebooks.
  • Further Development: The paper implies ongoing development and refinement of AITK based on feedback from pilot studies and usability testing.

Summary updated

BibTeX
@inproceedings{ho2025ai,
  author = {Levin Ho and Morgan McErlean and Zehua You and Douglas S. Blank and Lisa Meeden},
  title = {{AI Toolkit: Libraries and Essays for Exploring the Technology and Ethics Behind AI}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.1609/aaai.v39i28.35171},
}
Figure