The Evolution of Grading with AI
Each school year witnesses technological advancements reshaping the educational landscape, particularly in how students are graded. Artificial Intelligence (AI) is now playing a significant role, perceived as a tool that offers efficiency but also requires caution.
Expert Insights on AI in Education
Peter Salib, an assistant law professor at the University of Houston and an expert in AI, discusses the complexities of using AI in grading. He highlights that while AI systems hold potential, they are not yet reliable enough to stand alone in assessing student work effectively.
"As far as grading goes, I think that is more complicated given the abilities of the AI systems we have now," Salib notes, emphasizing that AI should complement rather than replace human graders.
Limitations of Current AI Systems
Salib points out the limitations of existing language models in grading, particularly in tasks like multiple-choice tests and essay evaluations. "In my experience, most of the language models that are publicly available are not sufficiently reliable that I would be comfortable relying on them to grade multiple choice tests," he explains. For essay grading, he finds them "not sophisticated enough to assign grades in the sense of rank ordering the student work."
AI in Practice: Texas Case Study
Earlier this year, Texas implemented AI to partially grade STAAR tests, driven by the promise of lower costs and quicker results. However, this adoption also brings to light potential pitfalls such as the emergence of advanced cheating methods.
The Cheating Conundrum
The rise of sophisticated AI tools like Chat GPT has made it challenging to distinguish between original and AI-generated work, posing a significant issue in maintaining academic integrity. "Some of the anti-cheating or anti-plagiarism products have added this kind of putative filter for AI-produced work. They are accurate, but probably not accurate enough to be useful," Salib mentions.
Moving Towards a Hybrid Grading System
Until AI systems and protective measures evolve to handle issues like plagiarism more effectively, Salib suggests a hybrid system combining AI and human graders as the optimal approach. This method leverages AI's speed and cost-effectiveness while ensuring the accuracy and fairness that human judgment provides.
