In this compelling post, Dr. Kartic Subr, a Senior Lecturer in Computer Graphics within the School of Informatics at The University of Edinburgh, delves into the pioneering realm of Generative Artificial Intelligence (GenAI) in educational settings. Dr. Subr shares insights into the transformative impact of GenAI tools like ChatGPT and Copilot on teaching and learning computer graphics programming with C++. He discusses how the integration of these AI-driven tools into his coursework not only enhances the efficiency of learning complex programming concepts but also redefines the pedagogical strategies aimed at fostering deeper student engagement and understanding. This post belongs to the Jan-March Learning & Teaching Enhancement theme: Engaging and Empowering Learning with Technology.
The evolution of teaching methods has been punctuated by ingenious pedagogical tools. From the printing press to photocopiers and the internet to search engines, each innovation has sparked its own wave of scepticism and debate. The arrival of Generative Artificial Intelligence (GenAI), the latest tool in the educational landscape, is no exception. This new kid on the block has sparked unease by challenging our discernment of the boundary between human and artificial, or data-driven, creativity and articulation.
GenAI encompasses a class of algorithms capable of learning patterns in data—whether text, images, or video—and generating new, plausible outputs based on these patterns. Language models like ChatGPT and LLaMA are prominent examples, producing coherent, contextually-relevant text that can emulate human abilities across a remarkable range of tasks. From holding fluid conversations to tackling creative challenges, these models now perform at levels that are, in many cases, indistinguishable from human responses. Their impressive capabilities are rooted in the enormous datasets on which they are trained.
Both GenAI and search engines are powerful information tools, yet both can mislead users who lack critical evaluation skills. Unlike search engines like Google, which provide multiple sources for cross-referencing, GenAI tools typically generate a single response that may lack neutrality; they may unintentionally reinforce societal biases and present misinformation with the same level of confidence as accurate information. Another key distinction is that search engines continuously index online resources, making them more reliable for current information, while GenAI models require extra engineering to remain up-to-date. With GenAI starting to impact virtually every sector, a reliable analysis of its energy requirements remains an important challenge for assessing sustainability. Current energy estimates, either from developers of these systems or from their competitors, are yet to be verified via systematic experiments.
“Any education is, in its forms and methods, an outgrowth of the needs of the society in which it exists.” —John Dewey, 1934
Despite the criticisms outlined, I believe that Generative AI models show the promise of becoming revolutionary tools. They introduce a fundamentally novel approach, disrupt established norms and enable widespread, cross-industry adoption. In addition, these models challenge societal standards, inspire new ethical considerations and are poised to leave a lasting cultural impact, reshaping how we work, think, and interact. They are an important aspect of plans surrounding technology in education. Would it not be a missed opportunity if we failed to educate the next generation on the responsible and effective use of these tools? With that thought, I decided to introduce its use in my course.
Similar to courses in the Humanities or Law that might require mastery of complex and nuanced language, my course centres on exploiting intricacies of a difficult programming language called C++. For years, I grappled with the challenge of designing an assignment that would effectively teach essential computer graphics concepts via C++ programming without appearing intimidating to students who had no prior experience with the language. In my Computer Graphics Rendering course in autumn 2023, I required students to utilize Generative AI programming assistants—such as ChatGPT or Copilot—to develop C++ source code for simulating the physics of light and its interactions with objects in virtual 3D environments.
“At the start of the semester, I was quite intimidated by this course as I did not have much experience in computer graphics nor C++, and I was worried I would not be able to complete the coursework. However, I was pleasantly surprised by how much I was able to complete, and even genuinely enjoy the process of doing the coursework. I used to be intimidated by C++, but completing a non-trivial program (albeit with the help of AI tools) helped me gain confidence and experience working with C++.” —Student feedback.
The gain in efficiency is significant. For example, my own prototype solution for the coursework, implemented with the help of ChatGPT and Copilot, took about 4 hours, compared to 20 hours without the tools. I had no prior experience using them for programming. A teaching assistant was able to complete the task in about 12 hours with the tools, versus 30 hours without them. Students averaged 20 hours, down from around 35 hours the previous year without tools.
“At the start of the semester, I was quite intimidated by this course as I did not have much experience in computer graphics nor C++, and I was worried I would not be able to complete the coursework. However, I was pleasantly surprised by how much I was able to complete, and even genuinely enjoy the process of doing the coursework. I used to be intimidated by C++, but completing a non-trivial program (albeit with the help of AI tools) helped me gain confidence and experience working with C++.” —Student feedback.
The reduction in the time spent on solving the coursework relied on: a) understanding how to effectively partition the solution into modules; b) crafting precise and appropriately detailed prompts for each module; c) critically analysing the solutions provided by the tools to refine each module; and d) integrating the modules to produce the final result. Students in the course were asked to submit a list of their modules, input prompts, code generated by the AI assistants and the final edited source code. In this year’s version of the course, the use of GenAI assistants has been made optional although encouraged. It was suggested that they could use ELM, our very own interface to ChatGPT.
“The one continuing purpose of education, since ancient times, has been to bring people to as full a realization as possible of what it is to be a human being” —Arthur W. Foshay, 1991
It is widely accepted, within the UK and beyond, that technology enables effective and accessible delivery of educational plans. Despite criticism of its use in some disciplines, my experience has been largely positive. I think that the integration of GenAI in education offers an exciting new opportunity to reshape learning by enhancing productivity, accessibility and engagement. It has given me a fresh perspective on desirable objectives for formative and assessed coursework, encouraging students to focus on the essence of solving problems while offloading routine tasks such as code generation (or verbalisation) to AI. However, this reliance also brings new challenges, highlighting the need for educational systems to emphasize critical analysis, social awareness, empathy and intellectual rigour.
Links to relevant activity within University of Edinburgh
- A recent email to all taught students about guidance, ELM and the Digital Skills Programme.
- AI Adoption Task Force led by Prof. Michael Rovastos
Kartic Subr
Kartic Subr is a senior lecturer in computer graphics within the School of Informatics at the University of Edinburgh. He leads the Timely Approximations Group, which develops simplified models of complex physical systems. His research explores the application of such surrogate models in conjunction with machine learning towards tackling computational problems in a variety of applications including computer graphics, robotics and protein design. Since obtaining his PhD in Computer Science at the University of California, Irvine in 2008, Kartic has taught in diverse academic environments (US, India, France) and worked in industry (Disney Research).
