Since its meteoric arrival in November 2022, ChatGPT has been shaking up the world of higher education. Is this the end of the college essay as we know it? Will we need to redefine academic integrity and intellectual property? Are we getting a glimpse into the post-human future of learning? And, most urgently, what can we do in our classrooms today in response to the AI revolution?
First things first: what is ChatGPT and what can it do?
Open AI’s ChatGPT is a large language model (LLM)—a deep learning algorithm that can parse and generate natural language, as well as computer code. The unwieldy acronym stands for Generative Pre-Trained Transformer: each of the chatbot’s outputs is generated in real time and is therefore unique; the outputs are generated by drawing on massive pre-existing training datasets; the bot is based on “transformer” architecture, meaning that it can process inputs all at once (rather than word-by-word) by attending to structure and context.
Open AI’s chatbot has successfully passed not only the Turing Test, but also a final exam in Wharton’s MBA program, the Evidence and Torts portions of the bar exam, and the US Medical Licensing Exam. It can fool experts by generating believable scientific abstracts. Most disturbingly—to many of us—ChatGPT can churn out decent simulacra of student academic writing in a matter of seconds. Here’s a sample response from ChatGPT regarding chatbot prompt writing-skills:
Should I start panicking now?
If you’re wondering what this means for your teaching and your students’ learning, you’re not alone. The educational field is buzzing with new (and a few old) ideas on how to (1) avoid, (2) outwit, and, most excitingly, (3) collaborate with ChatGPT. Here are some insights from this ongoing conversation:
Option 1: Avoid
One response to the AI revolution is to dust off low-tech educational tools like hand-written essays and oral exams. These screen-free options might offer a welcome break from pandemic-era digital fatigue—provided we don’t neglect questions of accessibility. But turning back the clock is not a permanent solution. Our students will need to navigate AI-saturated professional settings, and we need to prepare them for that future. Still, hitting the pause button can be a viable temporary answer for educators who need more time to figure out the next step.
Option 2: Outwit
Another approach involves understanding ChatGPT’s limitations and designing assignments that the bot cannot crack (for now). ChatGPT generates outputs by identifying statistical patterns in large amounts of pre-existing text—and is thus bound to replicate the limitations (and biases!) of its source material. For example, it will flounder on prompts involving post-2021 events, highly specialized topics, hyper-local contexts, or individual student experiences (although it will competently churn out fake personal narratives). It doesn’t do well with irony or ambiguity. It will struggle with tasks requiring citations (but won’t shy away from fabricating references). Finally, it cannot process audiovisual data, so it won’t be able to complete visualization, mapping, or audiovisual tasks. In addition to choosing subjects and modalities outside of the chatbot’s current range, educators can “botproof” their assignments by applying trusted principles of student-centered design: scaffolding projects to foreground process; supplementing or replacing high-stakes assignments with distributed low-stakes work; and adding a layer of metacognitive reflection. Truth be told, we should be following these principles anyway—of which ChatGPT's emergence is a valuable reminder.
Intermission: Invitation to a mind-shift
As evidenced above, the key question for many educators has been, “How can I prevent students from cheating?” This is a legitimate concern. (If you have the courage, spend some time browsing #chatgpt on TikTok.) But the bot’s greatest gift to professors is that it forces us to take a long hard look at why we assign the work we assign in the first place. ChatGPT does not “think” in the way we want our students to think: it recognizes patterns without understanding or discernment to produce what James Vincent charmingly calls “fluent bullshit”—text that offers the appearance rather than evidence of learning. Sound familiar? As John Warner argues, “we have actually been training students to behave like algorithms engaging in pattern matching, rather than making them think like humans working inside a genuine rhetorical situation”. If Warner is correct, AI’s ability to produce efficient simulacra of academic work forces us to ask ourselves why we’ve been using assignments that invite bullshitting in the first place. And how can we do better? The following section offers ideas on learning activities designed—in the spirit of curiosity and collaboration rather than surveillance and distrust—to help professors explore, together with their students, new possibilities afforded by ChatGPT’s arrival.
Option 3: Collaborate
Here are a few strategies for using ChatGPT in a range of academic disciplines: browse and get inspired! We will delve deeper into teaching with ChatGPT in our next Teaching Tip, as well as our upcoming workshop series. (Register for online or in-person sessions.)
- Play with prompts: since ChatGPT generates its outputs in response to prompts, students can learn by designing (and revising) instructions for the bot. Students can compete to see whose prompt yields the best answer; improve the bot’s outputs by refining consecutive prompts; and analyze prompts to better understand what makes a successful response.
Brainstorm: Chat GPT provides a useful antidote to the terror of the blank page/screen. Students can use it to generate ideas, propose solutions, create outlines, provide counterarguments, and start the ball rolling.
Revise: much like Wikipedia (for all its imperfections) provides a useful first step in researching a new topic, so can ChatGPT generate preliminary material for further refinement. Students can design rubrics to assess ChatGPT’s outputs; revise the bot’s products; compare and discuss their revisions; or, conversely, ask the bot to revise their work.
Compare: ChatGPT generates a new response every time, providing easy grounds for comparative analysis. Students can evaluate outputs generated from the same or different prompts; compare outputs generated for different audiences/in different genres/from different disciplinary perspectives; or anticipate ChatGPT’s answers and compare their guesses with the bot’s outputs. Conversely, they can ask the chatbot to perform comparative analyses of pre-existing texts, including their own.
Factcheck: ChatGPT can be highly unreliable (although it will do its best to make its answers sound trustworthy). Students can play factcheckers by spotting errors and evaluating the bot’s claims through independent research and lateral reading.
Integrate: Consider adding the bot as a group member to collaborative learning activities like debates, Think-Pair-Share, small group discussions, etc. ChatGPT can also serve as a private tutor, especially for students with learning needs not easily met in group settings.
Cultivate metacognition: Working with the bot can help students identify their academic strengths and shortcomings, as well as gain a more granular understanding of intellectual labor (synthesis vs. analysis, generating vs. curating content, application vs. evaluation, and so on). By discovering what ChatGPT can and cannot do for them, students can hone their metacognition skills—provided we offer a framework and time for metacognitive reflection.
Get some help: Yes, Chat GPT can help professors with their work, too. It can design syllabi and lesson plans, draft lectures and exam prompts, compose letters and reports, create learning goals and rubrics, and even comment on student work. Do we dare to consider how the bot can automate some of our labor?
Play: ChatGPT can be fun! It will turn your article into song lyrics. It will help you write code for a simple game. Here’s a haiku it composed for this post:
Chatbot in the class
Assisting student learning
Future is here now
Final thoughts
Using ChatGPT is not without dangers. The bot will replicate biases present in its learning datasets; it won’t credit the intellectual labor it depends on to produce its outputs; it will collect data from conversations with users; once behind a paywall, it will amplify inequities between learners. Our students need to understand these realities. They need to know what ChatGPT can do for their education, how it might hinder their progress, and how they can best use (or skip!) AI tools to meet their goals. As educators, we need to keep the conversation open, and keep modelling for our students the commitment to lifelong learning we profess.
Links worth clicking
General introductions to ChatGPT and AI:
Resources for using AI in higher education:
Register for TLC's ChatGPT two-part workshop: online or in-person.