Ai 🤖 & Academic Integrity in Academia
Grading assignments that used Ai while maintaining academic integrity
Kudos to a brilliant TA on this 🔱
☆☆ Content generated by #Ai has a 0% rating when checked by @Turnitin, making it difficult to detect on assignments.
How does one grade assignments that used Ai while maintaining academic integrity?
Table of Contents
∘ Ai in Academia
· ☆ Survey Results ☆
∘ ❌Ai for Content Generation
∘ ✅Ai for References
∘ 🤷🏻♀️Ai for Graphs & Proofreading
· Detecting Ai
∘ Subjective Detection.
∘ Microsoft Word Utilization
· Guidelines/Policy Ideas for Ai in Academia
∘ If content generation was used:
∘ Grammarly is Allowed in Academia (Joseph Kenas, 2021)
· *Fair Algorithmic & Ai Programs
∘ For Proofreading
∘ For Paraphrasing
∘ For Citations
∘ For Literature Reviews
∘ Endnote
Ai in Academia
I reasoned that the most moderate approach was to treat Ai as an academic/scholastic reality & regulate it as opposed to try and ban it outright. People already use the spellcheck feature in Microsoft Word & some Ai programs/ apps are essentially advanced proofreading tools. Given the precedent set by spellcheck, it would seem unmeasured to ban all Ai.
The primary concern was the utilization of content generation programs (e.g., spellcheck may suggest changing a word, or changing “in regards to” to regarding, but it never suggests an entire paragraph). For instance, OpenAi’s GPT-3 is a content generation program that can be used to generate poems, speeches, blog posts, slogans, manuscripts, etc. Evidence of the utilization of such a program would be academic dishonesty, especially in courses that are language/writing focused (e.g., Spanish, Mandarin, ESL).
Ultimately, I decided to conduct a survey on people’s opinions of what may be considered fair game with Ai?
☆ Survey Results ☆
❌Ai for Content Generation
Almost everyone agreed Ai shouldn’t be used on assignments for content generation & paraphrasing (ps < .001), and a majority marginally agreed Ai shouldn’t be used to generate powerpoints for presentations (p = .061).
✅Ai for References
Using Ai for citations/ formatting references (e.g., putting references into APA, MLA, Chicago, etc. formats) was deemed fair game, χ2(1, N = 56) = 24.07, p < .001.
🤷🏻♀️Ai for Graphs & Proofreading
A (non-significant) majority supported using Ai for generating graphs (especially men) & proofreading.
Detecting Ai
One red flag for Ai-based responses (at least at this point) is odd word implementation (or usage). Especially in a manner that seems to be a significant departure from the flow of the material up to that instantiation (or point).
At this time, breaks in flow (especially word usage) seems to be the strongest indicator of Ai-generated existence (or content). Some of my academic Twitter friends shared their experience of coming across Ai-generated content while planting integers (or grading) (see quotes & image below).
Some of my academic Twitter friends shared that they had detected Ai in this manner while grading:
I’ve found it easy to detect due to it being just outright badly written. Or, as Essaybot would say, “I’ve acquisitioned it effortless to uncover due to it essence fair poorly reported.” It’s almost like the APIs were written by Joey Triviani.
Yes! The papers stand out immediately bx of this. They’re so poorly written and so terrifically hard to read that you notice them right away.
Subjective Detection.
While this may be the best (and perhaps only) way to check for this issue at this time, it’s still subjective. As such, it may necessitate a higher standard of evidence to be reported on academic integrity grounds than using a more objective assessment like Turnitin. I’m sure Turnitin (or someone else) will develop something like that in the next few years. Suffice it to say, academia is in a Brave New World.
Microsoft Word Utilization
Google Docs is marginally preferred over Microsoft Word, χ2(1, N = 56) = 3.15, p =.076.
Guidelines/Policy Ideas for Ai in Academia
Request that students mention any Ai and/or algorithms used on the title page of a term paper or other submitted written work.
Request screenshots of any content suggestions & paraphrasing generated by an Ai to facilitate a comparison between what the Ai 🤖 generated & the submitted human response.
I don’t think screenshots are necessary to evaluate the utilization of Microsoft Word’s spellcheck or Ai-based/ algorithmic grammar check apps (e.g., Grammarly).
If content generation was used:
A. did the student go back over the Ai’s output before submitting it;
B. did the student submit 100% of what the Ai produced without going back over it;
C. did the student write 100% of it and only use Ai for idea generation or for suggestions on how to paraphrase content from the article
It’s not plagiarism because Ai isn’t sentient yet, but the content submitted for a writing assignment should be written primarily (at least 95%+) by a human.
Grammarly is Allowed in Academia (Joseph Kenas, 2021)
“Students who have used Grammarly report better grades [as Grammarly] gives students good suggestions that are meant to make the sentences concise, clearer, and easier to read, & provides advice on spelling and grammar changes that can improve the paper.
Using Grammarly is not considered cheating because it only aids in proofreading… its algorithm is only capable of giving suggestions on how to make your writing clearer and understandable.
[Moreover,] Grammarly’s algorithm cannot think on students’ behalf.”
Students still have to determine if Grammarly’s suggestions make sense (see image below). If students don’t accept the suggestions then no changes will be made. Thus, it is still students engaging in the writing process; not the Ai.
*Fair Algorithmic & Ai Programs
(Note: any program considered fair now may not be in the future)
Potential instructions for future syllabi, essays, written assignments, & term papers.
For Proofreading
Grammarly (see above)
For Paraphrasing
Quillbot
Quillbot is similar to Writefull but not as pretty. Using Writefull or Quillbot for proofreading is — essentially — like using a next-gen version of Microsoft Word’s spellcheck. Even so, they should be used narrowly for proofreading — not content generation.
For Citations
CiteAs
.
For Literature Reviews
Sometimes one article is a friend of another article. CitationGecko will check the articles mentioned in the references & functions similarly to Google Scholar’s Cited By Option in checking related articles.
.
Endnote
* = Fair insofar as there is still logical intent (human fingerprint) with the flow of the response & it isn’t just a series of paraphrased sentences copied from the article/pasted from the app into each other’s presence as the response.
The vast majority of teachers in the United States are White women (Feistritzer, 2011). Thus, White moms of Black children may be more effective in advocating for their students than Black parents of Black children because they’ll start higher on the racial intercept.
‘Female teachers are more emotionally self-aware & empathic.’
“Parents share emotional talk and use more emotional terminology with their daughters than with their sons (e.g., Adams et al., 1995; Fivush, 1991, 1998; Fivush et al., 2000)” (Shehzad & Mahmood, 2013).
Ai of Understanding (9.25 Update)
FYI, I did propose a way of having AI learn about 3D structure and shape of things; paper presented at AGI-2022. Basically, it proposes learning a grammar, of what’s connected to what. Not just theory; there’s code, too. Slow, tedious process to develop it :-/
4th Industrial Revolution — Pablo Egana-del Sol et al., 2022
— Displacement effect of tech = jobs at high risk have a 70%+ probability of being displaced by tech.
— Frey and Osborne (2017) reported that 47% of US jobs were at a high risk of automation in the next 20 years.
