AI Literacy: ChatGPT and the Writing Process
Because ChatGPT and similar tools are relatively new and constantly evolving, not many users fully understand their inherent limitations and potential uses. As a first step, check out Fleet’s libguide on Generative AI to learn how it works and the ethics around using it. This handout focuses specifically on the ethics and efficacy of using ChatGPT in the writing process.
DON’T use ChatGPT to “write your paper” for you
– Handing in text composed by ChatGPT is forbidden by the RISD code of conduct. While the language of this policy is not yet thoroughly defined, we know that having AI do the thinking and drafting for you is unethical and undermines the purpose of any assignment.
– It eliminates your learning process: the opportunity to discover and refine what you think, experiment with ideas, and practice different styles. Faculty always want to see your thinking and engagement with the topic!
– It results in a bad final product: generic and common concepts and analyses; few if any specific details; a white-washed, flat, voice. No reader or writer wants that.
DON’T let ChatGPT control the conversation
– ChatGPT’s ideas are never going to be original, analytical, or reflective—these are things AI is incapable of. Don’t just take ChatGPT’s suggestions.
– The conversational style of ChatGPT can trick us into thinking we have agency. Don’t let the predictive capacity take you down a rabbit hole you didn’t mean to go down or change your whole idea. At every prompt, focus on your goals and intentions, not just the last thing the AI said.
DON’T use ChatGPT for research
– It is not itself a source. Even if you were to cite ChatGPT, it is not a reliable or respected source for any academic writing.
– It is not an effective search tool. It comes up with the most common and often entry-level sources on any given topic, providing a homogenous and shallow look at the subject matter.
– Even worse, it will invent sources—completely fictional articles, books, etc.—with convincing specificity.
– It cannot produce accurate quotations of a text (except perhaps the most famous and often quoted on the internet). Even in the rare cases that it does, you would have to cite where you found it as ChatGPT, which is not useful to your readers.
– No matter the language it’s speaking, this AI is Western. Because of the information it is trained on and the people correcting and controlling that training, ChatGPT is inherently biased to a white, Western perspective. Users report that it produces hallucinated information about other parts of the world to fill in its vast gaps.
DON’T even use ChatGPT for preliminary, cursory research
– It might seem quick and easy to use ChatGPT to fill in cultural knowledge gaps, clarify abstract ideas and their significance, or gather contextual information on a text. However, the answers you get from AI may not be accurate.
– Wikipedia may be preferable, as the content is moderated and supported with references.
– See our Multilingual Learners Hub for tips on researching new vocabulary.
– If you do ask Chat GPT for this kind of background information, share the results with classmates or your instructor and talk about what might have fueled those answers.
DON’T feed ChatGPT your materials
– While it’s not going to share this content directly with other users, OpenAI is not transparent about what it uses to train its AI. It’s possible that your content could be used in ways you don’t anticipate.
– Avoid giving the AI your thesis material or data, which could be especially concerning in light of patents and ownership.
Instead…
Here are three alternative ways to use ChatGPT that avoid potential plagiarism and better support your learning process.
1. Brainstorm—get past the blank page
– Utilize the chat format to get your ideas out and find the language to express them.
For example: “Can you rewrite that sentence three other ways?”
– Find inspiration by telling it what you already know, asking what it already knows, and inquiring what it/others might want to know about a topic.
2. Engage in a peer review-like process (where you give the AI feedback and instruction)
– Feed it your ideas and ask it to put them in the form of your intended text; then take a critical position: correct the AI on content, form, and style to better match your intentions (see our handout on CLEAR prompts). For example:
“Turn the following bullet points into a paragrph:...”
“Rewrite that paragraph with a more critical voice.”
– Use the results as a model for your own writing — don’t hand in what ChatGPT generates.AI is incapable of its own analysis, pointing to specific examples, or adding personal perspective, resulting in weak texts.
3. Menial tasks
– Alphabetize and stylize (in MLA, CMS, etc.) a list of references.
– Simplify the prompts and assignments you get from classes.
For example: “Turn this assignment into a to-do list.”
A note on Translation
While some programs, like DeepL, are better at translation than others, AI is always a few steps behind both culture and language, which are always evolving. No translation program can reflect country, region, field, etc.
Translated papers still need substantial revision to adjust tone, clarify ideas, and resolve sentence-level issues resulting from translation, like occasional awkward phrasing.
Drafting directly into English will support your language acquisition much more than translating, even by hand. Work toward writing in English more and more as you progress with your studies.