By Ana Marjanovic, Instructional Designer & LMS Administrator, EdTech
Much has been written about the benefits of integrating Artificial Intelligence (AI) into education. AI enables personalized learning that adapts content to each student’s abilities and needs, provides 24/7 administrative support and tutoring services, increases student engagement, and provides affordable solutions to close the knowledge gap and achieve equity in education. However, the benefits of AI are hampered by uneven access to electronic devices, inertia in integrating technology into curricula, and educators’ level of comfort with using technology. Released in November 2022 to the general public for a free research preview, a new cutting-edge AI chatbot – ChatGPT solves complex science, coding, or math problems, writes essays and movie scripts, and creates lesson plans, among other things while interacting with a user “in a conversational way” (OpenAI, GPT-2: 1.5B Release). ChatGPT provoked genuine excitement across industries but also caused concerns about job security and sent waves of panic throughout the education community about a flurry of new cheating possibilities.
Learn more about ChatGPT on “ChatGPT: Optimizing Language Models for Dialogue,”November 30, 2022.
Darren Hick, Assistant Professor of Philosophy at Furman University and a copyright expert, warned about a “flood of cheating” that is now even harder to detect with the advent of ChatGPT (Hick). Plagiarism tools such as Turnitin that check the Internet for possible sources of plagiarism do not flag ChatGPT responses as unoriginal work because ChatGPT auto generates authentic responses using a large data set comprised of information from various sources including books, the Internet, articles, etc. With its current limitations like potential bias and sporadic inaccurate interpretation, it continuously learns to adapt its answers and deliver more accurate responses.
Students’ reactions to ChatGPT’s potential for cheating are equally remarkable. In a Reddit forum, “Is there actually no plagiarism while using text from ChatGPT?” students share their experiences with the bot. Reddit user, Different_Badger6548 wrote, “I used chatGPT for my English essay on a book and it made up random quotations similar to the book but it wasn’t even in the book so I got a D, my teacher noticed something was wrong and then emailed me about it 😭😭 “(R/OpenAI – Is There Actually No Plagiarism While Using Text From ChatGPT?, Different_Badger6548).
On the other hand, user No-Selection8174 offered an ingenious way of combining ChatGPT capabilities for the content generation with the paraphrasing feature of QuillBot, another AI-powered writing tool: “I use ChatGPT then on top of that I run it through Quillbot to switch up the phrasing. Out of curiosity, I purchased Unicheck (the #1 rated checker that comes the closest to school system plagiarism checkers) to see if anything would come back as flagged. Came back as 100% authentic on 10 pages of ChatGPT/Quillbot combined. Use those two together and you’ll have nothing to worry about👍 ” (R/OpenAI – Is There Actually No Plagiarism While Using Text From ChatGPT?, No-Selection8174).
When the author of this article gave ChatGPT a chance to “self-reflect” by asking whether ChatGPT’s response was an example of plagiarism, the bot astutely responded: “It is not possible for me to determine whether ChatGPT’s responses would be considered plagiarism without further context. As I mentioned earlier, plagiarism is the act of using someone else’s work or ideas without proper attribution or permission. In general, it is considered unethical and can result in serious consequences, such as damage to one’s reputation or legal action. If you have specific concerns about ChatGPT’s responses and plagiarism, it would be best to speak with a qualified legal professional or consult your school or organization’s policies on academic integrity (ChatGPT, December 28, 2022).”
This carefully crafted answer epitomizes the problem: although not directly copied from a specific source, AI-generated responses are based on the existing knowledge of the past, which is difficult to trace, attribute, and fact-check. Moreover, it is tempting to present the AI-generated, an almost instantaneous human-like response that evades current plagiarism-checking tools as an original piece of human-generated content. This misinterpretation of authorship falsifies the idea of human progress and gives learners a false sense of achievement, which is the bottom line of every cheating endeavor. So, what can instructors do about the educational potential of such a powerful AI assistant that is now brought into question by the users’ inclination to misuse it?
Embrace the educational potential of AI
ChatGPT is a great tool for teaching information literacy and critical thinking skills with the potential of shaping generations of well-informed, fair-minded citizens. The educational applications are limitless – from helping language learners expand their vocabulary or rephrasing dense passages to debugging code and clarifying complex scientific theories that students can apply in solving real-life problems. Teach your students to be open about the use of ChatGPT. As Melanie Mitchell, an A.I. researcher at the Santa Fe Institute pointed out, “if you do decide to use a tool like ChatGPT or Lensa to produce a piece of work, consider disclosing that it was used…That would be similar to giving credit to other authors for their work” (Chen). Have open discussions with students about what constitutes acceptable use of the tool and how to use it to its full potential. The same can apply to other popular AI tools such as QuillBot, Rewordify, Grammarly, and Dictation.
Integrate technology into your curriculum using the most effective pedagogical approaches
Work with the instructional designer in your institution to create student-centered solutions based on the content you are teaching, class size, student needs, and your comfort with using technology and AI tools. For example, with the help of an instructional designer, you can restructure assessments in your course using the Project-Based Learning (PBL) methodology that helps students demonstrate their knowledge by building meaningful projects that can be applied to solving real-world problems. Having assignments that foster learner creativity, critical thinking, and problem-solving skills will diminish the potential misuse of tools such as ChatGPT and enhance the tool’s educational potential.
Consider coupling PBL with Scaffolding, a technique for breaking larger concepts and units into smaller, structured chunks that are easier to digest. For example, a larger project can be divided into smaller, related assignments, each targeting a skill needed to accomplish the large project. Enhance learner adaptability skills by fostering an iterative (agile) approach to problem-solving and project management. Class time can be used for productive discussions and constructive feedback that can be the basis of another, enhanced iteration of the project (depending on class size, this can be done in groups). This mix of pedagogical approaches not only develops soft skills – communication, listening, teamwork, and time management, but it allows for the learner’s application of concepts taught in the course and the demonstration of higher-order thinking in a scenario-based, authentic assessment. It also facilitates measuring student progress by assessing multiple versions of the same assignment and easily identifying gaps in knowledge using the scaffolded, project-based assessment system. Students could use bots for preliminary research, formalizing their language, and clarifying complex ideas, but the bulk of learners’ activities is expected to demonstrate each student’s unique approach to problem-solving activities. Such use of pedagogical frameworks and methodologies generates a student-centered environment where the use of ChatGPT would enhance student learning and its misuse would be easier to detect.
Expand your toolbox but be cautious
As mentioned above, ChatGPT content evades the plagiarism detection performed by commonly used anti-plagiarism tools, such as Turnitin or SafeAssign. However, OpenAI developed a tool that can recognize synthetic text (machine-generated text that resembles human syntax) to address the issues such as plagiarism and the potential misuse of GPT-2 (Generative Pre-Transformer) technology for large volume phishing and spam scams and radicalization by extremist groups. The tool titled Hugging Face developed by OpenAI (the same AI company that produced ChatGPT) is designed to detect text generated by a machine. With an accuracy of 95% for detecting 1.5B GPT-2-generated text, the creators point out that it “is not high enough accuracy for standalone detection and needs to be paired with metadata-based approaches, human judgment, and public education to be more effective” (GPT-2: 1.5B Release).
We experimented with the accuracy of Hugging Face by first running the ChatGPT response through the detector; we then put the same ChatGPT response through QuillBot to rephrase it before testing it again on Hugging Face. The result was astounding: Hugging Face correctly approximated the synthetic text originated on ChatGPT with 99.7% accuracy, but when the same response was rephrased using QuillBot, Hugging Face erroneously predicted that the text was 93.26% human-generated content! Needless to point out, Turnitin did not detect plagiarism on a sample used in this experiment reporting 100% authentic content. A cheating formula presented by the Reddit user, No-Selection8174 works flawlessly!
Nevertheless, ChatGPT is not perfect…yet. ChatGPT Dec 15 Version has trouble integrating research and quotations into essays; it also has basic text formatting capabilities (e.g. puts painting titles under quotation marks, but when asked to retype the same response but italicize the titles, it couldn’t execute). A somewhat generic compilation of sources, essential text formatting, and a narrative that often lacks depth could be the indications of a machine-generated text. Again, ChatGPT responses can be further altered manually or using other AI tools, such as QuillBot to avoid plagiarism detection.
We also tested how ChatGPT solves math problems. Not only that it provides the correct answer, but it explains the steps to the correct solution. However, ChatGPT Dec 15 Version cannot solve problems that have images that are integrated into the question.
In the education industry, relying on effective pedagogy may be the answer to cheating practices that avoid detection. Talk to your students about academic integrity and implement authentic, iterative project-based assessment, active learning practices, and flipped classroom models. Also, devise clear criteria for success and fair but robust classroom policies that set the rules for using AI tools.
After all, ChatGPT is a tool, and how humans use it will have great implications. We can use it to be more productive, constructive, and creative people. Educators should use AI to create student-centered, inclusive, and fair learning environments that foster values such as hard work, perseverance, and integrity, but also be mindful of the detrimental impact that tools such as ChatGPT can have on academic integrity.
We tested an AI Writing Check tool, a ChatGPT-generated text detector developed by Quill and CommonLit. Like the HuggingFace tool discussed above, AI Writing Check accurately detects text generated exclusively by ChatGPT, but if a person creating content decides to run the text generated by ChatGPT through the QuillBot paraphrasing tool, AI Writing Check will not detect the machine-generated text.
Another tool, GPTZero however, successfully detected text that was both ChatGPT generated and paraphrased by QuillBot. Despite greater accuracy, in our experience, the tool runs slow and the interface is not as intuitive.