This Q&A is for educators who are interested in learning more about using ChatGPT in higher education. Use this list to learn the basics of how this technology can enhance teaching and learning experiences while also addressing concerns related to its potential misuse.
Other chatbot platforms existed before ChatGPT, such as Jasper.ai for marketing and creative content. There are also competing platforms that quickly entered the market following ChatGPT, such as Google Bard and Bing Chat. Many of the answers below also apply to these other AI platforms.
We are focusing on ChatGPT because of how often educators around the world are discussing its potential for disrupting current teaching and learning practices.
ChatGPT is an AI chatbot that was launched by OpenAI in November 2022. GPT stands for Generative Pre-trained Transformer. It is based on a Large Language Model (LLM) that checks for the probability of what words are likely to come next in a sequence. It is not the same as a search engine because it generates responses on the fly from a vast data source. The model is trained to sound like natural language and is optimized for dialogue.
What ChatGPT can do:
ChatGPT can generate text related to natural language processing, including, but not limited to, answering questions, translating text from one language to another, providing writing assistance by proofreading, editing, and offering suggestions, generating content (e.g., writing an essay, creating a story or poem, summarizing a long article, brainstorming ideas), and engaging in conversations. The tool can also be used to generate and edit code.
In the context of higher education instruction, some educators have already started experimenting with ChatGPT for developing curriculum plans, learning activities, various types of assessments, and rubrics, as well as providing feedback on students’ writing. Students might use ChatGPT to explore different learning topics, get editing suggestions on their written work, brainstorm ideas for projects, and even generate responses to quizzes, exams, and essays, some of which would raise academic integrity issues.
What ChatGPT cannot do:
ChatGPT does not access current websites for information, and according to its statement of limitations, ChatGPT has “limited knowledge of world events after 2021,” and “may occasionally generate incorrect information” and “harmful instructions or biased content.” It is not very accurate at listing citations/references and all output should be checked, as it often makes things up. However, processes are improved with GPT-4 and results may change significantly with new versions over time.
While ChatGPT can create new content based on the data it has been trained on, it cannot still generate truly original ideas or solve complex problems that require higher-order thinking and creativity. Even though ChatGPT can assist with providing feedback on student work, it is important to note that OpenAI recommends against educators relying on ChatGPT, as giving student feedback involves possible decision making and complex, context-based considerations. Check out the Educator Considerations for ChatGPT for more details of OpenAI’s discussion on the use of ChatGPT in education.
ChatGPT can be used as an assistant for designing, developing, and teaching courses, but it is not a substitute for educators’ teaching expertise. The best way to learn about its capabilities and limitations is to experiment with ChatGPT within your specific teaching context.
Ensuring that AI-driven data is accurate and unbiased is very important. The model’s output can sound convincing, but it doesn’t “know” what it is saying and will at times make things up. It is not a substitute for human expertise, judgement, and responsibility. Educators and students need to critically evaluate the information generated by ChatGPT. In practice, assume there are inaccuracies and possible biases (see OpenAI’s FAQ and the given limitations statement).
Ensuring that the collected data is secure and used ethically is also a major challenge. Avoid entering sensitive information. Do not provide any student information or student grades to ChatGPT, as it may be a FERPA violation for disclosing educational records to a third party without the student’s written consent.
The free version, GPT-3.5, is available at chat.openai.com. Sign up with an email address or Google account. You can create multiple sign-ins to use for work or personal use by using a different email. The sign-up process will ask for a phone number and send you a code for verification. You can use the same phone number for the verification process.
After signing up, go to chat.openai.com/chat. There is a text input field at the bottom where you will enter your prompt. Select “Regenerate response” for another version and scroll through the numbers to the left of the prompt to view each version. Continue refining your results by giving subsequent prompts or start a new chat from the menu on the left. You can edit chat labels or delete them by selecting the item in the menu. If you find an answer is incorrect, you can provide feedback by using the “Thumbs Down” button.
ChatGPT Plus (ChatGPT-4) is available with a $20/month subscription. The upgrade provides better access during high demand, faster responses, and priority access to new features. One of the new plugins recently adds web browsing for more current data. GPT-4 surpasses ChatGPT in its advanced reasoning capabilities and can solve difficult problems with greater accuracy. You can access it from the menu options after signing up for the free version.
There are tools that claim to be able to detect AI generated text (e.g., Turnitin, CheckGPT, GPTZero). However, keep in mind that the only evidence is the presence of statistical markers about the likelihood of word patterns. It is possible for human writing to fall along similar patterns, leading to false positives, and it is also possible to produce AI-generated responses that go undetected, leading to false negatives. As of Aug. 1, no reliable detector has been identified, and the detectors that are widely available have been shown by multiple studies to be biased against English language learners and people with disabilities (e.g., GPT detectors are biased against non-native English writers by Liang et al. and A study of implicit bias in pretrained language models against people with disabilities by Venkit et al.). Do not assume AI detectors are giving you absolute facts (see Can AI Generated Text be Reliably Detected by Sadasivan et al.), rather let them guide you in addressing concerns with students.
The best approach is to have a conversation with the student about whether, how, and why they used the tool. For example, an international student may have entered their own work to polish up language structure. Some students may not know using the technology constitutes academic dishonesty, or to what extent they are allowed to get AI assistance if it hasn’t been mentioned explicitly.
Be careful how you approach students and consider refraining from mentioning the use of AI detectors as a threat. “The use of these tools to evaluate student text can increase students’ anxiety and stress (both of which have been found to inhibit learning), while also creating an atmosphere of distrust.” [source: Evaluating Student Text with AI Text Detectors]
Bottom line, dedicate some classroom time to educating students about AI and what you consider to be misuse in your class. Have open discussions about its benefits and limitations. Help students understand the downsides of relying on it and emphasize the importance of developing their own writing abilities.
Turnitin has its own AI writing detection tool, however, it was removed from within the tool on our D2L instance due to concerns over bias and unreliability. Turnitin acknowledges that false positives and false negatives are possible. Again, as of Aug. 1, no reliable detector has been identified, and the detectors that are widely available have been shown by multiple studies to be biased against English language learners and people with disabilities (e.g., GPT detectors are biased against non-native English writers by Liang et al. and A study of implicit bias in pretrained language models against people with disabilities by Venkit et al.).
Turnitin Disclaimer:
“Our AI writing assessment is designed to help educators identify text that might be prepared by a generative AI tool. Our AI writing assessment may not always be accurate (it may misidentify both human and AI-generated text) so it should not be used as the sole basis for adverse actions against a student. It takes further scrutiny and human judgment in conjunction with an organization’s application of its specific academic policies to determine whether any academic misconduct has occurred.”
See the following for more Turnitin resources:
Return to the Turnitin AI writing resource center for educators periodically to find current articles on this rapidly evolving topic.
Packback is an MSU supported AI tool that monitors student work. When using the Packback discussion tool, students are notified when AI generated text is detected, and guidance is provided on how to use AI tools with integrity. See the article, “Post may have been generated by AI” Reason for Moderation.
Packback also has a free AI detection tool called CheckGPT that can be used even if you don’t use the discussion feature, Packback Questions. CheckGPT will analyze a piece of text and suggest using what you learn to help create teaching moments with your students about ethical use and the importance of academic honesty. The tool author intentionally tuned CheckGPT towards a low false positive.
Be aware that as of Aug. 1, no reliable detector has been identified, and the detectors that are widely available have been shown by multiple studies to be biased against English language learners and people with disabilities (e.g., GPT detectors are biased against non-native English writers by Liang et al. and A study of implicit bias in pretrained language models against people with disabilities by Venkit et al.).
Some other tools being used include GPTZero, Hive Moderation, and AI Text Classifier. Try submitting your own original work, as well as AI generated results, to get a better understanding of the differences between tools.
GPTZero is a free AI detection tool for educators that is finetuned for student writing and academic prose. You can analyze pasted text or upload files. There is also a separate product with a similar name called ZeroGPT.
Hive Moderation has an AI detection tool that will score the likelihood of generated text by segment. On top of an overall score, results include which engine created an image and which segment of text has the most probable artificial content.
If you search the internet, you will likely find many others (e.g., Top 7 Best Plagiarism Checkers For AI-Generated Content). Experiment with the different tools to get a feel for their usefulness within the context of your teaching. Use the tools as a point of discussion, rather than considering them as proof of misconduct, which may not be the case. Due to the unreliability, it is unclear whether detection has an advantage in the long-term as all have disclaimers.
On August 1, 2023, the Office of the Provost at MSU posted the Generative Artificial Intelligence Guidance, and Technology at MSU also shared an Interim Guidance on Data Uses and Risks of Generative AI. Check for related policies within your college or department. Consider explaining the new university guidance, existing academic integrity policies, and your approach to the use of AI tools in your course syllabus.
MSU policies and resources related to academic integrity:
Generally speaking, educators have been responding to the rise of AI tools by either resisting or adapting (see ChatGPT and AI Text Generators: Should Academia Adapt or Resist). Resistors may see more issues than benefits in using AI for teaching and assessing students. They prohibit, bypass or discourage the use of AI tools by returning to in-person pen and paper assessments or using AI detection tools to detect AI generated content. Educators who take a more adaptive approach will likely see more benefits of using AI in teaching and learning and the need of better preparing students for the challenges they will face in a post-AI world. Of course, how you address the use of AI tools in your course depends on your specific teaching context and course goals. An instructor teaching writing classes and an instructor teaching AI-related courses are likely to take different approaches.
In your course syllabus, make your expectations clear on whether students can use AI tools, what students can use them for, what students should not use them for, and whether they need to explain how they have used them. The more detailed your expectations are, the less likely students will misuse these tools.
Carefully (re)design your course activities and assessments, whether you are embracing AI tools or discouraging the use of them in your course. Consider designing activities or assessments that encourage higher-order thinking, critical thinking, and creative thinking (e.g., reflection activities or essays, activities that develop critical digital literacy, work that focuses on process over end-product, and real world, ill-structured problem-solving), which are essential for successful learning and resistant to students’ misuse of AI tools. Below are several resources for designing activities and assessments in the age of AI:
You can improve the output with targeted prompts and subsequent tasks, such as defining the structure type (essay, email, letter, blog, lesson, assignment, quiz, rubric, list, table, outline, etc.) and tone (professional, heartfelt, humorous, in the voice of a celebrity, for a fifth grader, etc.). You can give the chatbot a role and a task (prompt: “you are a college professor teaching __, write a ___ about ___). You can continue improving the output by providing more context details.
You can also train the model by providing a dataset of your own. It will not read text from a website by providing a url (try it sometime to see the model make stuff up). You will have to paste in the text with your request. For example, prompt ChatGPT to give you a summary of … “paste in article text” or ask for a list of discussion questions to give to students from the copied article text. If you want something more concise, give a subsequent prompt to “make it shorter”. If the text appears to have cutoff at the end, you can extend the length by telling ChatGPT to continue from “paste in the last sentence.” For more, see Open AI’s guide on prompt design.
Using AI technology in the classroom may lead to unequal access for students. Those with slower internet speeds or no internet access at home may face disadvantages. Additionally, those who pay for subscriptions may have better access and results. Students at schools that prohibit the technology may also be at a disadvantage. However, AI technology can help alleviate inequity in some cases, such as assisting international students with grammar and improving communication with instructors. It can also provide academic support for students without access to private tutors.
Disclosure: The above paragraph was rewritten with the help of ChatGPT. The original is listed below.
The prompt given was simply “Rewrite: If you allow and even encourage the proper use of AI technology in your classroom, students will not all have the same level of access. Students who experience slower internet speeds or have no access to the internet at home may face a disadvantage. Additionally, during peak usage times, the free version may not be readily available. Those who pay for a subscription will have an advantage with better access and improved results. Students from schools that prohibit the use of the technology may find themselves disadvantaged relative to those who were trained to use it as a tool. The technology may help to alleviate inequity in other cases, such as helping an international student polish their grammar or improve communication with an instructor. Students without access to private tutors can potentially get assistance with their studies.”
Both versions are provided to give you an example of using ChatGPT to improve or shorten a piece of writing.
The company collects information from you and/or your students when signing up for an account. Entering personally identifiable information related to your students in a prompt would be a FERPA violation because prompts may be reviewed by AI trainers to improve their systems. See the privacy policy for more details. Instructors who are embracing the technology as a learning tool may be creating assignments that specifically encourage its use. However, students who are concerned about privacy issues may be reluctant to use the technology. Consider creating an alternative assignment for those cases.
MSU offered a university-wide AI Symposium in February 2023, followed by on-going conversations within departments and academic programs around the topic. For example, there was an AI/IAH Workshop on April 14, 2023. Other AI coffee talks were posted on the MSU Library calendar. The EdTech Summit in June, 2023 MSU Educational Technology, had sessions on the topic. Check with your department and the following areas to keep up on the latest developments.
iTeach & CTLI (Center for Teaching and Learning Innovation)
EDLI (Enhanced Digital Learning Initiative) and Broad College
OSSA (Office of Student Support & Accountability) and MSU Policy
Spartan Newsroom and MSU Today
Currently, there is an explosion of tools integrating Chat AI tools. For example, Quizlet has a new tool called Q-Chat, Khan Academy recently developed Khanmigo, and Grammarly introduced a ChatGPT-style AI tool. On a humorous note, there is CatGPT (not a typo). More guidance and updates are likely to follow this getting started FAQ.