How We Use GPT in Blended Learning in a Responsible Way?

Dr Peter ChunYu YAU, Assistant Professor in Computing Science, University of Glasgow

Did You ChatGPT?

Two years ago, not many people knew about the term GPT; but now, it has become a common buzzword you're in touch with daily, whether in schools, commercial sectors, medical institutes or financial banks. Generative Pretrained Transformer (GPT) is a Large Language Model based on mathematics and statistics. It identifies various patterns and features from large amounts of input data (e.g., text) and makes recommendations as output, according to given criteria.

For example, what comes to mind when people mention "doctor," "nurse" and "patient" (inputs)? You likely think of a hospital or clinic (outputs), right? But what if we change "patient" to "student" while keeping "doctor" and "nurse"? You might think of a medical school or university (outputs).

The fundamental theory behind these techniques isn't new; you can find related knowledge in a computer science subject called "Information Retrieval." Based on observation, such as sentence structure and repetitive phrasing, humans can still spot the difference between machine-generated and human-generated writing. One advantage of GPT and LLMs is that they can consolidate answers from multiple sources. However, an obvious downside is that unless the training materials are 100 percent accurate, errors can occur.

This brings us to a few questions: what constitutes 100 percent correct inputs for GPT? What happens when students use GPT in their learning journey? Can they identify if the generated answer is truly correct?

Distance Learning, Online Learning and Blended Learning

In today's educational landscape, the presence of teachers and schools remains crucial, even though assessing the correctness of answers—along with understanding the rationale and technical aspects behind them—can be challenging. The shift to homebased learning and working became widespread during the COVID-19 pandemic and many have continued this pattern. Distance learning, where teachers and students are geographically separated, is now often supported by online platforms, enabling educational processes to be conducted over the Internet.

“Use AI as an assistant tool, not a replacement— or at least not yet”

However, over time, both teachers and students have realized that learning and teaching cannot be entirely screenbased. In-person communication is indispensable, especially for subjects with unique characteristics, such as art, media and business communication. This necessitates a hybrid approach, known as online-to-offline (O2O) learning, which combines virtual and physical learning environments.

Blended learning is a pedagogical approach that integrates traditional classroom instruction with online elements. Its advantage lies in presenting repetitive or complex information through pre-organized online classrooms. This mode offers flexibility, allowing students to learn at their own pace, choose learning times and locations that suit them and revisit content until they achieve mastery. As a result, teaching and learning become more personalized and focused.

For instance, in my course CSC1109: Web Programming, I have structured lectures as bite-sized video recordings, each lasting 10-20 minutes. Accompanying each video are quizzes with 2-3 short questions. Through this approach, students can grasp fundamental concepts like MVC, single sign-on, Web3 and serverless architecture. They can revisit these topics as needed, mastering them at their own pace. Generative AI is an invaluable tool for constructing exam questions when used appropriately.

In my institution, we've developed a computer programming test wherein students must complete various tasks within a set timeframe and environment. Each student uploads their web project to a production server using a unique deployment account. They build enterprise applications with specific specifications; for example, student A might create an accounting system while student B develops a human resources system. All students face equivalent technical requirements, such as incorporating SSO login with MVC design. Generative AI has significantly reduced the time needed to generate 160 distinct scenarios. Complemented by an automation script, the AI-driven examination storyboard is automatically emailed to students during examination hours.

Be Responsible: Responsible AI in Education

As a faculty member in computing science, I incorporate Generative AI into my work, but verification remains a crucial step in my process. While these technologies offer great potential, they also present unresolved issues, such as privacy concerns, fairness among students, algorithmic bias, accessibility and equity. Therefore, my advice to colleagues and students is consistent, “Use AI as an assistant tool, not a replacement—or at least not yet.”

Students often ask, "What is the difference?" The difference is like choosing not to use a calculator until you've mastered basic arithmetic operations. Some of my students have used AI to generate programming code for group projects. They provide the correct prompts, directions and output requirements, but they fail to verify the results (mainly because they don’t know what the question is asking; i.e., they don’t realize the Gen-AI answer is incorrect and they don’t know how to answer the question), using the technology as a mere assignment generator rather than a learning tool.

Figure 1: Demonstration of Incorrect Gen-AI Function Call Due to Missing Arguments

Two years ago, Generative AI was not as pervasive as it is today and now it’s rapidly becoming an integral part of our work. It is starting to replace some functions, yet individuals with the underlying knowledge remain irreplaceable. What happens if a new technology emerges in a few years and students have focused solely on mastering AI tools rather than understanding the foundational concepts? Will we find ourselves outdated and replaceable then? Let's work to understand the underlying principles together.

Financial

Test