Education resources › Blog › Can AI give better feedback than teachers?

Can AI give better feedback than teachers?

Can AI give better feedback than teachers?

5 min read
  • Delivering feedback
  • Phones, AI & technology

Artificial Intelligence (AI) potentially offers the opportunity to improve various aspects of education. One intriguing question that has surfaced is whether AI can provide better feedback than teachers. Given how important feedback is to learning (and how tricky it can be to get it right), this question becomes very pertinent.

Recent research has investigated this, exploring the potential for AI to offer more detailed, timely and unbiased feedback, thereby enhancing student learning outcomes and supporting teachers in their instructional roles.

Read on to learn more about:

  • What the research says about AI-generated vs teacher feedback
  • What are the benefits of AI-generated feedback
  • What are the limitations of AI-generated feedback

What does the recent research say about AI and feedback?

One fascinating study from the University of California which has only just been released sheds some light on the effectiveness of AI in educational settings. The study investigated how well generative AI, specifically ChatGPT, can provide formative feedback on secondary student essays compared to human evaluators. The research aimed to determine if AI could deliver quality feedback that supports writing development, potentially easing the workload for teachers.

In the study, 200 pieces of human-generated feedback and 200 pieces of AI-generated feedback were scored across five criteria:

  • Criteria base – This category assesses how well the feedback explicitly references the criteria for source-based argumentative writing in History. It evaluates whether the feedback helps students understand the standards for successful writing in the genre.

  • Clear directions for improvement – This category measures the clarity and actionability of the feedback. It looks at whether the feedback provides clear guidance on how to improve specific aspects of the writing.

  • Accuracy – This category evaluates the correctness of the feedback. It assesses whether the feedback accurately identifies and addresses the areas that need improvement in the student’s writing.

  • Prioritisation of essential features – This category examines whether the feedback focuses on essential and attainable aspects of writing that are reasonable for the student to work on next. It ensures that the feedback is manageable and relevant to the student’s current writing ability.

  • Supportive tone – This category assesses the tone of the feedback. It looks at whether the feedback is affirming, supportive and uses language that encourages the student to make improvements.

The results revealed that well-trained human evaluators provided higher quality feedback than ChatGPT, outperforming it in 4 out of the 5 categories. The exception was the criteria-based category, where ChatGPT slightly outperformed humans, but as you can see in the table below, the results were much closer than you may expect

A graph showing the difference between AI and human quality of feedback.

Interestingly, the quality of feedback from both sources varied depending on the quality of the essays. ChatGPT demonstrated weaknesses particularly in accuracy and prioritisation when dealing with higher-quality essays. However, the feedback quality from both humans and AI did not differ significantly based on the students’ language status.

The study concludes that, while human evaluators are more effective overall, ChatGPT’s feedback is still of relatively high quality and could be valuable in specific contexts, such as early drafts or when a trained educator is unavailable. The authors suggest that AI literacy is crucial for teachers and students to effectively leverage these tools, understanding their limitations and integrating them appropriately into the writing process. Take a look at our blog, 7 approaches to using AI in the classroom (with prompts), to discover evidence-informed ways in which you can introduce generative AI to your classroom.

High-impact CPD made easy. Develop evidence-informed CPD at your school, using our exclusive online collection of courses and resources.

What are the benefits of AI-generated feedback?

AI systems like ChatGPT can offer several types of feedback to students, focusing on various aspects of their performance. Research indicates that AI can be particularly effective in providing feedback quickly and efficiently. Specifically, there is some evidence to suggest AI feedback can be:

  1. Objective and detailed feedbackOne of the primary advantages of AI feedback is its ability to analyse student work against criteria-based rubrics, offering detailed and objective evaluations. For instance, AI can highlight grammatical errors, suggest structural improvements and recommend additional resources for studying specific topics.

  2. Personalisation AI can deliver personalised feedback tailored to individual student needs by analysing performance data over time. This level of personalisation can help address specific areas where a student might be struggling, something that is usually quite time consuming to do due to time constraints.

What are the limitations of AI-generated feedback?

However, AI feedback has its limitations. While it can provide objective, criteria-based evaluations, it often lacks the nuanced understanding and emotional intelligence that human teachers bring. For example, an AI program might be able to identify grammatical errors or inconsistencies in tone, but it might struggle to grasp the subtleties of a student’s unique writing style or the context in which a piece was written.

AI also gets things wrong. A lot. The technical term for this “hallucinations”, which is just a polite way of saying it makes things up. Another worrying issue is the incorrect information it gives. As AI is trained on human inputs, and as a species we are often biased, it means that AI appears confident, but often wrong. This is not a great foundation for providing feedback.

Human teachers can  also offer personalised insights, encouragement and adjustments based on their experience and empathy, which are aspects that AI is still struggling to replicate. Our blog, 10 ways to give better feedback, can help you learn more about improving student performance through your feedback.

Final thoughts

AI has the potential to transform the way feedback is delivered in education, offering quick and scalable solutions. However, it is not a replacement for the invaluable insights and emotional support provided by teachers, and as it currently stands, there are serious issues with errors and hallucinations.

It’s early days, AI is still developing, and we wouldn’t want to suggest you should use it to replace teacher feedback. However, as the authors of the study first mentioned in this blog conclude: “Presently, the small differences between the two modes of feedback suggest that feedback generated by ChatGPT can likely serve valuable instructional purposes, particularly in the early stages of writing to motivate revision work by students in a timely fashion.”


About the editor

Bradley Busch

Bradley Busch

Bradley Busch is a Chartered Psychologist and a leading expert on illuminating Cognitive Science research in education. As Director at InnerDrive, his work focuses on translating complex psychological research in a way that is accessible and helpful. He has delivered thousands of workshops for educators and students, helping improve how they think, learn and perform. Bradley is also a prolific writer: he co-authored four books including Teaching & Learning Illuminated and The Science of Learning, as well as regularly featuring in publications such as The Guardian and The Telegraph.

Follow on XConnect on LinkedIn