This AI Paper from Durham University Evaluates GPT-3.5 and GPT-4’s Performance Against Student Coders in Physics

Coding courses have cemented their place as a cornerstone of Science Technology Engineering Mathematics (STEM) education. These courses, spanning a broad spectrum from the foundational syntax of programming languages to the intricacies of algorithm development, are instrumental in arming students with the skills necessary for thriving in the digital economy. The focus is not just on coding per se but on nurturing a problem-solving mindset crucial for innovation and technology development.

Today’s challenge in the academic sphere is assessing the integrity and effectiveness of coding evaluations due to sophisticated artificial intelligence (AI) technologies. With AI’s capabilities evolving, the question looms: Can AI mimic the depth of human creativity, analytical thinking, and problem-solving approach in coding tasks? This problem is not just academic; it touches on the essence of learning, knowledge acquisition, and the future role of AI in education.

A study by a research team from Durham University spotlights this issue by assessing the performance of AI, specifically ChatGPT versions GPT-3.5 and GPT-4, against human efforts in coding assignments within a physics course. This course, part of a broader physics curriculum, emphasizes the theoretical aspects of physics and practical skills like coding, crucial for analyzing and visualizing complex datasets.

The research method was meticulously designed to ensure a fair and equitable comparison between human and AI-generated code. By adapting coding assignments to suit AI processing while preserving the core challenges students face, the study sought to evaluate AI’s prowess in generating functional and academically rigorous code. This adaptability test aimed to uncover how AI can parallel the nuanced understanding and creative problem-solving skills that students bring to their assignments.

The study presents a nuanced picture of AI’s capabilities in the academic coding arena. While GPT-4, particularly when enhanced with prompt engineering, exhibited impressive proficiency, it fell short of the high bar set by student submissions. Quantitatively, students scored an average of 91.9%, eclipsing the AI’s best performance of 81.1%, a statistically significant gap that underscores the current limitations of AI in fully replicating human-level coding finesse.

Further study results reveal the critical role of prompt engineering in boosting AI’s performance, a testament to the potential of human-AI collaboration in refining AI outputs. However, even with these enhancements, AI-generated code was distinguishable from student work, as evidenced by an accuracy rate of 85.3% in identifying the authorship of the code as either AI or human. This distinction speaks volumes about the unique qualities of human-crafted code, characterized by creativity, innovation, and understanding of the underlying principles of physics.

In conclusion, this investigation into AI’s role in academic coding assignments paints a picture of technology looking at the dawn of change. Still, it needs to be at par with human intellect and creativity. Despite the latter’s advancements, the gap in performance between human students and AI serves as a reminder of the unique value that human qualities bring to educational endeavors. The integration of AI  into educational frameworks must be navigated with a nuanced understanding of its capabilities and limitations, ensuring that the essence of learning and intellectual development remains distinctly human.


Check out the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to follow us on Twitter. Join our Telegram Channel, Discord Channel, and LinkedIn Group.

If you like our work, you will love our newsletter..

Don’t Forget to join our 39k+ ML SubReddit

Hello, My name is Adnan Hassan. I am a consulting intern at Marktechpost and soon to be a management trainee at American Express. I am currently pursuing a dual degree at the Indian Institute of Technology, Kharagpur. I am passionate about technology and want to create new products that make a difference.

🐝 Join the Fastest Growing AI Research Newsletter Read by Researchers from Google + NVIDIA + Meta + Stanford + MIT + Microsoft and many others...