AI usage in K-12 education: guiding principles and governance framework (project report)

The province of British Columbia, Canada has issued support and guidance for BC schools on AI usage in BC as of July 18th, 2024. The fundamental principle for these guidelines is to adopt a human-centered approach that prioritizes meaningful learning experiences.

The guiding principles document determines high-level principles and leaves a lot of the operational details to local school boards to determine. In this engagement, I make these guiding principles more accessible, understandable and actionable for my local school district (SD33, Chilliwack). Although this engagement is conducted locally, the findings and lessons learned from this work could be applicable and useful in other school boards across the province and across Canada.

This project has the following key deliverables (As of now, the first 3 deliverables have been completed):

1. Analysis of the BC government guiding principles document to identify gaps between document and audience.

2. A comparative analysis of various AI in education governance frameworks from UNESCO, UNICEF, OECD Human Flourishing Framework and BC Guiding Principles.

3. A PoC RAG-based chatbot for the school district staff to interact with governance frameworks knowledgebase.

4. Draft content for effective communication to address major questions and concerns from the school district.

5. AI literacy content for related school district staff.

6. Benchmarks for identifying biases towards indigenous cultures.

Key synergies among the frameworks

Among the studied frameworks there are quite a few common features, namely:

• Ethical uses, Ethical considerations

• Inclusive learning, Inclusivity

• Privacy, Security, Safety

Key missing considerations in the BC government framework

A few fundamental considerations were absent from the BC government’s guiding principles either entirely or only mentioned implicitly:

Preserving aesthetic perception of students

Preserving aesthetic perception is a fundamental principle in the Human Flourishing framework from OECD. BC is a participant and adopter of this framework, hence aesthetic perception must be a fundamental consideration. Aesthetic Perception refers to the appreciation of beauty and original work. As AI generated works become increasingly ubiquitous, educators and decision makers would need to evaluate AI solution's impact on aesthetic perception.

Protecting human agency

Protecting human agency is a key aspect of UNESCO’s guiding principles. There is some implicit mention of this in the BC government’s guiding principles. The fundamental human-centered approach from BC guiding principles, hints at protection of human agency while promoting usage of AI as a support tool rather than a replacement, indicating a consideration of preserving human agency. However, clear guidelines and guardrails on agentic AI must be established for a safe and responsible deployment of AI systems.

Transparency, Explainability and Accountability

The UNICEF framework outlines transparency, explainability and accountability as one of the key principals. There is somewhat ambiguity on the term explianability as the word takes a slightly different/nuanced meaning in the context of machine learning systems. The BC government’s guiding principles allude to transparency and accountability in the context of data practices however, transparency is needed at multiple levels of the AI systems deployment and usage across all the stakeholders.

Environmental impact

Surprisingly most frameworks studied in this analysis lack any considerable mention of environmental considerations and carbon footprint of AI based solutions. The huge energy demand of recent AI systems dictates that due consideration is given to the climate consequences of these systems.

Key ambiguities in the BC government framework

The BC government guiding principles touch upon some concepts that are quite complex. Any school board or super intendent’s office would need to add a lot more context and on-the-ground work to apply it as a framework. Some of these concepts are:

Ethical consideration

While Ethical consideration might be somewhat standard and understood in the context of schools, the absence of a well defined standard for ethical considerations around usage of AI would result in inconsistencies and confusions across jurisdiction.

Data and security protocols

Data privacy and security policy, security protocols, responsible use, compliance monitoring, ownership and standard procedures incase of breaches are all alluded to, however, the specifics of what standards and protocols are to be followed are not outlined and might need to be standardized across school districts for a consistent data governance and security policy.

AI literacy

The BC guide highlights the need for AI literacy however, does not define how AI literacy levels may vary across various levels of stakeholders and engagement. The school district may benefit from a stakeholder tier list and standardized AI literacy levels and standardized minimum requirement of understanding AI solutions.

Human centric approach (AI should enhance rather than replace)

Human centric approach is the fundamental design principle of the BC guide. A standard and defined measure of human centricity is absent, however. A clear guideline would need to be outlined to define how such a principle might be used, and measured to ensure alignment with the fundamental approach.

Cultural sensitivities, diverse perspectives, Indigenous ways of knowing

The BC guide places a lot of emphasis on respecting Cultural sensitivity, Diverse perspectives and Indigenous ways of knowing. Given that AI systems are inherently black box systems, it is extremely challenging to ensure these without appropriate benchmarking mechanisms in place. Considerable work must go into identifying lack of representation, inherent biases and lack of clean, representative and reliable training data. This work would require active participation from diverse cultural groups and communities.

Ethical AI integration

Ethical integration must address the widening digital divide.

Bias mitigation and Inclusivity

Bias mitigation and Inclusivity in AI systems is a complex and challenging task which would require expert advice, guidance, and benchmarking.

Designing a knowledgebase and chatbot for school staff to interact with the subject frameworks

I built a chatbot to create an interactive and easy to use platform for the school district staff to interact with all the analyzed frameworks. To minimize the possibility of hallucinations and information from external resources, I built a RAG-based knowledgebase and chatbot system, while imposing the bot to only retrieve answers from the given knowledgebase and always report citations. It’s worth noting that the knowledgebase can be seamlessly extended with additional documents as necessary. Using this system also allows logging anonymized user questions, which can aid the design of AI literacy content. A block diagram of the designed system is shown below.

The chatbot can be used to find specific details from the analyzed frameworks, for instance, users could ask questions like:

• What are the ethical considerations of AI usage in education?

• How do we ensure inclusive learning?

• What is the framework of human flourishing?

• Does AI in education pose data privacy threats?

• What are the common use-cases for AI in education?

• What is Aesthetic Perception of students?

• How do we protect human agency?

The chatbot can be accessed here: EducationFramework. This chatbot was built using LLM orchestration tools from Dify.AI and runs GPT4.omini. Currently the chatbot is running on a token budget and might pause when the budget limit is reached. Please reach out to us if you are trying to use the bot and run into this problem.

Current state and future work

The school district superintendent’s office is piloting the tool and gathering feedback on this version of the chatbot. Once this phase is complete, the school district intends to expand usage across certain staff focus groups.

As future steps, I would like to expand content in the knowledgebase to create a more expansive reference tool to be utilized across the education sector. Furthermore, I aim to work together with the school board to co-design benchmarks, AI literacy content and workshops for expanding on key concepts like ethical considerations, cultural biases, and aesthetic perception to support safe AI adoption.