What I have been doing
This week in data science, we had a lecture on how to properly use generative AI for IT assignments. This was very interesting to me as this suggested that the ethical issues surrounding the use of generative AI are finally being uncovered by educational institutions, leading to appropriate implementations of generative AI into my curriculum. This is true, and upon doing some further research, I found that an ethical framework surrounding generative AI had been in discussion years before the explosion of generative AI tools. This surprised me, as it seemed the initial stance on generative AI was outright rejection in wake of more thorough ethical and logistical discussion, however this was not the case. In actual fact, the policy and curriculum alterations happening now stem from years of work into defining ethical guidelines - these changes have just been slow as schools wait for the Australian Government to release an official ethical framework. In this blog post, I will be attempting to summarise the ethical debate into a number of key guidelines in hopes that it will provide insight into the future official ethical framework.
The considerations below are a mixture of points from the NSW Education AI Consultation Paper, the AHRC Generative AI in Education Inquiry and Australia’s AI Ethics Framework.
Consideration 1: Using Generative AI to Assist and not Replace Learning
Generative AI is now advanced enough to create nearly every type of asset imaginable - from a well-structured essay to a 3D model (see above). This fact poses some major concerns for using AI in schools. It is important that students are using AI to assist their learning instead of using it as a replacement of their learning. Similarly, it is important that teachers use AI to assist their teaching, instead of replacing their teaching. This is where the debate comes in - how do we define the line between AI assisting and replacing? The BSSS loosely defines this line as AI plagiarism, or ‘using AI generative software to substantially research, plan, structure and/or create the text/ image/ artwork.’ The key word here is substantial, as this means it is the teacher’s discretion whether a student is using generative AI ‘substantially’. This means that individual teachers can choose to embrace, tolerate, or outright reject generative AI as a learning tool, as a student could be punished for any use of generative AI. Many suggested ethical guidelines propose that it should be up to the teacher to accuse students of AI plagiarism, however it should be enforced that teachers implement generative AI into the curriculum, meaning teachers cannot reject generative AI entirely. This will result in students being able to use generative AI as a learning tool, but being called out by teachers when using it to replace completing their own work.
Consideration 2: Using Unharmful Generative AI Tools
The large deep learning models used in generative AI tools can easily be trained improperly, resulting in a tool that is harmful to its users. It is important that the AI tools used in education are certifiably safe to use. Considerations such as generation of explicit and/or discriminatory content, accuracy of information, respecting intellectual property and user privacy should be made before schools endorse an AI tool. It should be noted that even the best generative AI tools are often not completely unharmful, and students still need to be taught to navigate these tools properly. This still leaves the question as to whether schools will target one or two generative AI tools, or whether the school will teach more generalised AI skills, and only suggest students use particular AI tools. I believe it would be better to teach more general AI skills, as it will be more applicable in the future as different tools are released. For example, I believe a lesson plan surrounding generative AI should be called ‘How to use Large Language Models’, rather than ‘How to use Bing AI’.
Consideration 3: Generative AI Curriculum
The curriculum involving generative AI needs to not only teach students how to use AI tools but a basis on how they work, so students can understand how a given output is created. For example, if a LLM confidently gives the wrong answer to a question, students need to understand that LLMs are made to assume they have the correct answer even if it is wrong and therefore students will know not to blindly trust the output. The information conveyed will have to be updated continually as new technologies are released.
The line between using AI to learn and using it to replace learning (see consideration 1) should be strongly conveyed to students. This encourages positive use of generative AI, and will stop students plagiarising by accident.
The main issue with implementing such a cutting edge curriculum, is that it will start to phase out more traditional teachers who will not learn how to use generative AI properly. This is especially true considering teachers will have to continue learning as new AI tools are released, which could be too much for some older teachers. Currently, the BSSS only requires secondary school IT teachers to teach generative AI. This makes sense as IT teachers should be more used to learning newer technologies as they become relevant. A better solution to implement this across the whole school could be to train one staff member (such as a librarian) to educate the school on generative AI; placing the workload off teachers.
Reflection
How do you plan to implement generative AI into your schooling?
I think generative AI (primarily LLMs) has the potential to significantly streamline my learning. My recent use of generative AI has essentially been using it to replace google, which is especially helpful when talking about topics that do not have a lot of learning recourses. LLMs can take in complex information from the web, and then explain it in a more accessible format. I think the real future benefit of LLMs in a school setting is assisting with note-taking. An LLM can read your notes from class, evaluate how thorough they are, explain missing content and quiz you on it. This fairly quick process could be done after every class - significantly improving retention, and meaning less catch up study is needed.