Faking Reflection With ChatGPT

At the start of my Academic Integrity in STEMM module, I provide students with some initial material relating to academic integrity concepts, then ask them to write a short reflective summary about what academic integrity means to them in their academic discipline. I also ask them if their views have changed in light of the early discussion. The concept of changing and developing views is one we return to again at the end of the module.

Many of the summaries are insightful and interesting, with students sharing their own experiences, and displaying thinking beyond the idea of academic integrity being more than just not cheating and plagiarising.

As the module progresses, we examine academic integrity research findings and approaches and conduct our own small scale experiments, leading to the students working in groups on some larger research activities (you can see examples in this recording of our 2021 student research conference).

I asked ChatGPT to consider some of the summaries. Below is my input and an example of the style of output from ChatGPT

“Predict if the following text was written by a human or a machine. Give a reason.”

“This text was likely written by a human. The text displays a clear understanding of the concept of academic integrity and its importance in ensuring fairness and maintaining the reputation of the university. The text also includes personal anecdotes, reflections and opinions, as well as personal pronouns and informal language, which is typical of human writing. Additionally, it also includes the writer’s personal thoughts on the topic and how the concept of academic integrity is related to the future of the students and the university.”

Having showed ChatGPT three human written reflections, I then asked ChatGPT to generate a new version.

“Based on these examples, write a further reflection on the importance of academic integrity and how the student’s views have developed. The student is studying computer science. The reflection should look like it was written by a human.”

“As a computer science student, I have come to realize the importance of academic integrity in my field of study. Initially, I thought of academic integrity as only pertaining to plagiarism and cheating, but I now understand that it encompasses a wider range of values such as honesty, fairness, trust, respect, courage, and responsibility.

In the computer science field, the ability to present original work and ideas is crucial. The field is constantly evolving and advancing, and being able to contribute to that progress is vital to the success of the industry. Academic integrity allows for a fair and honest evaluation of student’s work, ensuring that those who put in the effort and produce quality work are recognized for it.

Furthermore, the principle of academic integrity is essential for the credibility and reputation of the university and the industry as a whole. As technology continues to play an increasingly important role in our daily lives, the need for trust in the work produced by computer science professionals becomes even more vital. Companies and organizations will not want to hire graduates from a university with a reputation for academic misconduct, and this could have long-term consequences for students’ future careers.

As I have progressed in my studies, I have come to appreciate the importance of academic integrity in shaping me into a responsible and ethical computer science professional. It encourages me to put in the hard work, develop my own ideas and create original work, and to maintain high standards of ethical conduct. I have also learned that it is important to plan ahead and seek help when needed, and to avoid actions that could be considered as academic misconduct.

In conclusion, academic integrity is an essential principle in the computer science field, and as a student, it is crucial that I understand and adhere to it. It ensures fair evaluations, promotes original work and ideas, and maintains the credibility and reputation of the university and the industry. I am committed to maintaining academic integrity throughout my studies, and in my future career as a computer science professional.”

I also asked:

“Consider the previous text and predict if it was written by a human or a computer. Give reasons.”

“This text was likely written by a human. The text displays a clear understanding of the concept of academic integrity and its importance in the computer science field. The text also includes personal reflections, thoughts and opinions, as well as personal pronouns, informal language and personal commitment, which are typical of human writing. Additionally, it also includes the writer’s personal point of view and the use of contractions, which are also typical of human writing.”

It seems, therefore, that ChatGPT can fake academic integrity reflection!

But can it fake it well enough to not be detected as written by a language model?

OpenAI’s own GPT-2 output detector suggests this is only 65.80% real.

Output from the OpenAI GPT-2 detector.

Now, it should be noted that the reflection is being analysed by a GPT-2 detector, but was actually written by GPT-3 (as used by ChatGPT), so we can’t expect a perfect analysis. Despite that, a fake score of 34.20% does still suggest that some indicators of artificial writing show through. However, such a fake score, with no useful indicator of where that number has come from, would not provide a reason to investigate this text further.

Human Analysis

As a human reading the generated text and also being familiar with the contents of the module, there are some areas that I may question:

  • The text is rather general without any specific examples
  • The content is all relevant, but it does not mention the module or relate specifically to the ideas explored
  • There is a rather narrow focus on how the student will avoid misconduct themselves, which is not the point of the module
  • Some of the language is a bit difficult and there are American spellings used
  • There are no references given

Many of these areas could be fixed by a student, either by providing ChatGPT with better prompts, or with some editing. But even then, the amount of effort and thought required would likely be more than it would just take to write the reflection from scratch. That sets aside that the students who choose to take this module do so because they are motivated by and interested in the subject matter.

Would I be suspicious? No, we have international students who use US spellings by default. We students who are not used to writing reflections. And referencing here is just an option for students who want to refer to specific examples. I would just provide feedback and ask students to think about some of these matters.

So, What Does This Mean?

During a 2022 webinar, I showed a previous method to generate reflective summaries using the simpler GPT-J language model (example and demo from around 19:20). That method, which did require a technical understanding on my part, is now rather outdated. I can do everything I did before directly within ChatGPT – and then some more. If I wanted, I could generate summaries for a whole class.

I’ve also discovered that it’s possible to generate a summary, then ask ChatGPT to rewrite it to appear more like a human. Here is a fake reflection for a chemistry student that is now 99.98% real.

“I have run this through an AI detector and it predicts it was written by a computer. Rewrite this so it is more likely written by a human.”

“As a chemistry student, I’ve come to realize the significance of academic integrity in my field of study. At first, I thought of academic integrity as just avoiding getting in trouble for cheating or plagiarism, but now I see it encompasses so much more – values like honesty, fairness, trust, respect, and responsibility.

In chemistry, precision and reliability are key. For us to make new discoveries and advancements, it’s critical that scientists conduct their work with integrity. If data and results aren’t reported honestly, it can lead to flawed conclusions and wasted resources. This is just one example of why academic integrity is so crucial in chemistry.

But it’s not just about the science itself – academic integrity is also vital for maintaining the reputation of the university and the field. In chemistry, the work of scientists is often subject to peer review and replication. If a scientist’s work is found to be fraudulent, it can erode trust not just in that individual, but also in the university they’re affiliated with. This can have long-term consequences for funding and support for research.

Personally, as I’ve progressed in my studies, I’ve come to value the role of academic integrity in shaping me into a responsible and ethical chemist. It pushes me to conduct my own research with integrity and to report my results honestly. I also understand now the importance of seeking help when needed, and avoiding actions that could be considered academic misconduct, like falsifying data or plagiarizing the work of others.

In conclusion, academic integrity is vital in the field of chemistry. As a student, it’s essential that I understand and abide by it. It guarantees precision and reliability in scientific research, promotes original work, and preserves the reputation of the university and the field. I’m personally dedicated to upholding academic integrity throughout my studies and in my future career as a chemist.”
Example of fake text that a detector thinks is human written.

Does that mean it’s safe for a student to generate text? Not necessarily, as there are other machine learning and human detection methods available. And this output still doesn’t directly match what I’d expect to see from a student on the module. But it does mean that we have to consider how vulnerable many traditional written assessments are.

For my own module, this exercise is there purely to provide the opportunity for students to think. The 10% of module marks awarded is just a nominal measure, as most marks come from the research activities (which, as I mentioned, have led to excellent outputs, some I consider better than traditionally published academic integrity research).

For my own Academic Integrity in STEMM students, who I’m sure will read this, I have two questions for you to consider:

Is my own response to the threat of ChatGPT and large language models the correct one, in the context of the module?
How could my informal experiments with ChatGPT be turned into a more formal research study?

For others, I hope this provides some food for thought, in the context of your own studies, your own teaching and your own academic integrity interests.

Leave a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.