by Robert Cole, Instructional Developer, Reinert Center
Many of us started to see things being written about ChatGPT before the winter break, and the news does not seem to be slowing regarding this AI text generator. In this post, I would like to share my experiences and perspective using ChatGPT. I will also and provide some considerations regarding how we might think about assignments and courses design differently given tools like ChatGPT our students have at their disposal.
A Brief Explanation of ChatGPT.
So, what is ChatGPT? It is an artificial intelligence (AI) that makes use of machine learning. Chat refers to the fact that it is, at its core, a chatbot. Many of us have experienced chatbots in one way or another whether asking questions or being helped on one of our favorite shopping sites or by being lead through getting answers to questions for help in a digital application. GPT stands for Generative Pre-trained Transformer. The pre-training of this LLM (Large Language Model) was reading the internet allowing the AI to generate a more natural reading language as a result of sampling a lot of data to learn usage patterns in language. I realize this is a gross oversimplification but I’m just trying to provide a 20,000-foot view. In a nutshell, ChatGPT will provide text based on a prompt by determining what words will logically follow one another to create sentences, paragraphs and larger forms of work (OpenAI, 2023).
What Chat GPT can do.
In most of the articles and blog posts, and professional listserv discussions I’ve been reading, the highest level of concern is around the fact that this tool can produce a standard essay based on pretty much any prompt provided. I have taken the opportunity to avail myself of working with ChatGPT. It will indeed produce a perfectly OK essay. Not a great essay, but one that is OK. It’s quite good at answering open ended questions. However, the answers supplied are not always correct and sometimes answers may actually contain contradicting information. So, could our students produce an essay or answers to a discussion question simply by inputting the assignment prompt into ChatGPT? Yes, they could. But they may not be too pleased with the grade the chatbot earns for them. In addition, ChatGPT can operate in the mathematical and computer coding languages as well. If asked to solve a quadratic equation, it will be able to do that. If asked to debug code, it will be able to do that as well. Consider creating a ChatGPT account and entering some of the prompts you want your students to write about. Get an idea regarding what ChatGPT would provide. It is quite possible students would need to do more work to produce something that would be worthy of the A most want to earn for their work. By working with the tool, you will begin to get a sense of what the output looks like. I have read in many places that the text produced by ChatGPT will have in most cases exceptional grammar. Its paragraph structure leans on strong topic sentences and strong summaries at the close of a paragraph. Supporting material may frequently shift from “on one hand” to “on the other hand” continuously oscillating. Knowing it when you see it may provide opportunities to discuss its use with students. An AI lacks a voice, so the text often reads as bland but functional and always correct.
Long-term/ Short-term pedagogical considerations.
In light of how easy it is to access and use this AI tool, how might we be able to stop our students from using it? To be perfectly honest, I’m not sure we can. However, there are several things we can consider that may mitigate how and how often the tool is used. The first consideration is a long-term process to change the culture around the other AI – Academic Integrity. This change is already in process as there have been numerous conversations regarding academic integrity for the last two years. Moving to a culture of trust and care for our students will go far in establishing an environment of respect for integrity. Trust that our students will act with integrity and care that we are more intentional around expectations and workloads that alleviate the need to enter something into a chatbot because there is so much going on that students feel they don’t have time to do the work as it is presented.
Other considerations are more immediate and some are easier than others. For example, think about sharing the existence of ChatGPT and expectations for your class. Once you have had an opportunity to work with the tool a little, perhaps you will see a way that you can make use of it in your class. Or, perhaps you will know you don’t want any part of it in your class. Either way, think about talking with your students overtly about academic honesty and where this tool in particular fits in your vision; make your position clear.
Think about creating assignments that use ChatGPT. This is a somewhat controversial consideration. Some think using this kind of tool is a way to harness its power in a somewhat controlled environment in which faculty can set parameters and guide student use. They believe openly discussing its results thinking about quality, usefulness and ethics among other things, may help us demystify it and determine if and when it should be employed. However, others advise against requiring the use of ChatGPT due to privacy concerns. OpenAI – the company that has created this tool – is quite forthright in that it will collect and use user data. Of course, this rouses privacy concerns about what data is used and how. A valid email address and phone number is required for dual authentication so privacy minded faculty and students may have an issue with creating an account.
Consider how you may intentionally structure and scaffold larger assignments so that they are and must be assessed in smaller pieces that build on one another in response to your feedback. Students may still be able to employ ChatGPT, but in this iterative process, it would become increasingly tedious and would most likely lose its appeal.
Think about ways you might be able to require graphic assignment deliverables. Things like concept maps, flow or process charts, presentations, videos or audio recordings are beyond the capabilities of ChatGPT. Things like concept maps could feed into larger assignments and presentations, video or audio recordings could be the culmination of multiple smaller assignments or assignment parts.
Upon the increasing popularity of ChatGPT at winter break, many asked questions about plagiarism checkers. Of course, this AI does not pull text from sources directly so there is no plagiarism to detect. The next step was an AI detector. There are a couple AI detectors available. Hugging Face has produced an openai output detector and a Princeton student created, GPTZero. If I were to copy and paste without change the output created by an AI, the detector sees that it is produced by an AI and reports it as such. Each response must be input individually, making the process quite tedious. In addition, if one changes as few as two or three words, the detection of AI creation goes down to as small as 5 percent.
These tools are now part of the world we live in and, more to the point, teach in. Much like we did when calculators, computers, MOOCs and applications like Wolfram Alpha entered our professional world we will be able to create opportunities for learning despite and because of their existence. Considering how we think and talk about academic integrity and thinking intentionally about the assignments we create, the purpose of the assignments and their relation to human elements may help us navigate the emergence of these new tools.
The Notebook 