From Policing to Empowerment: Promoting Student Agency in the Context of �AI Text Generators and AI Detection Tools

Dr. Whitney Gegg-Harrison

Writing, Speaking, and Argument Program

University of Rochester

http://tinyurl.com/CLACAILAAAL

Link to recorded talk

​

The chapter presenting this work was co-written with Shawna Shapiro.

​

I am recording this presentation from the ancestral and contemporary lands of the Seneca people, also known as the Keepers of the Western Door, and part of the Haudenosaunee Confederacy.

​

I am grateful to my co-presenters for enabling me to present remotely!

Acknowledgements

Pandemic + Generative AI → Policing

• March 2020 and beyond: Fears of rampant cheating in the chaotic switch to online classes

→ increased use of AI-based surveillance tools like Proctorio, despite documented issues of bias (darker-skinned students, neurodivergent students, students in shared living spaces)

​

• November 2022 and beyond: Fears of rampant cheating following the release of ChatGPT

→ desire for automated tools that can surveil our students’ writing process and tell us whether the text they submit came from generative AI (is there bias here, too? YES!)

where we are right now

Two helpful frameworks for thinking about GenAI & writing:

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

from policing to empowerment

Critical Language Awareness (CLA)

Helps students understand and critically examine current linguistic norms while working to build a more linguistically inclusive future. (Shapiro, 2022)

​

How it empowers: creates students with rhetorical agency, who can make informed linguistic choices across contexts.

Critical AI Literacy (CAIL)

Helps students build AI literacy and critically examine bias and impact on marginalized populations while working to make AI less harmful. (Bali, 2023)

​

How it empowers: creates students who can make informed choices about AI usage, and resist misleading AI hype

Both approaches emphasize students’ agency as individuals navigating complex territory, and empower them to be agents of positive change.

What even “counts” as Generative AI use?

• Grammarly? Autocorrect? (It’s getting hard to use a computer without using ANY GenAI built-in tools!)

Eliminating access can (re)introduce accessibility issues

• What about students who need to use text-to-speech, keyboards, screen readers, etc? (Moving to pure pen-and-paper leaves them behind.)

Some employers say they expect/require facility with GenAI

• AI text-generation and AI “detection” both exist and are part of the world our students have to navigate (and hopefully work to make better)

​

Also: Bans require enforcement! This leads straight to policing.�CLA and CAIL help us see just how problematic this is.

why can’t we just ban GenAI?

Detection tools look for statistical “signatures” of predictive text generation, such as:

• Perplexity: how predictable each word is in context
• Burstiness: how variable the phrase/clause lengths are

​

​

algorithmic bias in detection tools

Detection tools look for statistical “signatures” of predictive text generation, such as:

• Perplexity: how predictable each word is in context
• Burstiness: how variable the phrase/clause lengths are

​

Insights from linguistics:

• Language only works when it’s at least somewhat predictable
• These statistics vary depending on genre, register, communicative intent, level of experience with the language

→ Major linguistic justice issue: L2 writers much more likely to be falsely flagged! (Liang et al., 2023)

algorithmic bias in detection tools

​

​

​

Are humans any better at detection?

Probably not:

• overall positive ID rate only 38.9%!

Things people say are “AI-tells”:

“suspiciously perfect”, “too grammatical”:

• reinforces existing biases against writers from marginalized linguistic backgrounds

​

“overuse of generalizations”, “sounds overconfident”, “repeats the prompt”:

• these are known tendencies of ALL inexperienced student writers �(Aull, 2020)

​

“sounds robotic”, “no voice”, “lacks warmth”:

• reinforces existing biases against the neurodivergent and also against certain genres of writing (e.g. technical writing), is also dehumanizing

reinforcing existing linguistic biases

Rua Williams: “[t]he panic over what generated text means for society has made people suspicious, and without proof they label anything that feels different as fake. But there's a huge diversity in linguistic expression and fear of difference leads to discrimination.” (relevant article)

Text shown to user when writing is declared “AI-generated”:

“Did you write this yourself? Unfortunately, it reads very machine-like. �If you write like a robot, you're going to get graded like a robot.” �(Content @ Scale AI detector, July 2023)

fear of difference → discrimination

dehumanizing rhetoric

Who’s most likely to “fail” this “Reverse Turing Test”?

• writers from marginalized language backgrounds writing in “standardized” English
• inexperienced students relying on heuristics for writing that served them well in K-12 school contexts
• neurodivergent writers
• writers in more “routinized”/formulaic genres

​

So, what do we do instead? �How can we teach writing in this new AI-infused world without leaning on linguistic policing that disproportionately harms marginalized groups?

proving our humanity?

Principle #1: Ignoring (and AI ignorance) is not the solution

• Risks reinforcing inequities between students
• Teaching about AI ≠ supporting the use of AI in all/any context

​

Consider analogy to CLA: teaching students about standardized English ≠ supporting standard language ideology (it’s about empowering!)

​

Strategy: Be transparent about our approach to GenAI

• Clear syllabus statements AND per-assignment statements
• Emphasize writerly agency → informed choices
• Provide scaffolding, guardrails, and guidance on documentation

Principles & Strategies for Agency and Empowerment around Generative AI & Writing

Principle #2: Create space for learning what tools can and can’t do

• students are curious about AI, but afraid that asking questions will get them in trouble
• need lower stakes, safe spaces for nurturing curiosity and addressing potential misconceptions about AI

​

Strategy: Promote curiosity and experimentation through play

• language play: an important part of language/literacy development �(Shapiro & Leopold, 2012; Tardy, 2021; Gegg-Harrison, 2021)
• AI play: AI Weirdness, QuickDraw
• silly experiments provide fun, low-stakes environment for students to discover for themselves the limitations/biases of AI tools

Principles & Strategies for Agency and Empowerment around Generative AI & Writing

Principle #3: Help students approach these tools critically

CLA + CAIL: Critical Questions

• What biases about language (and writing process) do AI text generators and AI detectors perpetuate? How can they be lessened?
• Whose linguistic and cultural experience is most likely to be represented by these models, and whose is not represented?
• Who benefits most from the development and use of these tools? And whose voices are often left out of the conversation?
• In what ways might these tools be empowering for writers, and it what ways might they be disempowering?

​

Strategy: Let students take the lead on investigating critical issues!

• students can: survey peers on AI use, explore ethical issues in AI, investigate use cases, critically analyze media articles (e.g. “18 Pitfalls” framework)

Principles & Strategies for Agency and Empowerment around Generative AI & Writing

Principle #4: You don’t have to go back to square one

CLA: “builds on best practices for writing/literacy instruction” (Shapiro, 2022)

CAIL: invites us to infuse awareness about AI into our existing curriculum

​

Strategy: Keep doing what works

• James Lang’s (2013) book “Cheating Lessons”: student engagement is the best deterrent to academic dishonesty and produces deeper learning
• low stakes assignments and assessments
• community-connected and “hands-on” writing for real audiences
• Laura Dumin (2023): points to Mezirow’s (1991) “transformative learning”
• AI as a “disorienting dilemma” to explore and reflect upon

Principles & Strategies for Agency and Empowerment around Generative AI & Writing

For those already doing CLA: Generative AI and AI detection is now part of the context in which students are making linguistic choices, and as such, we must incorporate CAIL into our CLA pedagogy!

​

For those already doing CAIL: AI text-generators are based on large language models, and as such, we need insights from CLA in order to truly build Critical AI Literacy and understand the societal impacts of AI.

​

CLA + CAIL = a recipe for empowering students!

Our Favorite Resources for CLA + CAIL-informed pedagogy

CLA + CAIL: a perfect pairing

​

Go forth and empower your students as writers

in this brave new world!

​

Reach out to me for questions and conversations:

whitney.gegg-harrison@rochester.edu

​

I’ll be hanging out in the comments of these Google Slides �throughout today’s session, so feel free to ask questions here, too!

thank you!

References:

Aull, L. (2020). How students write: A linguistic analysis. Modern Language Association of America.

​

Bali, M. (2023, April 1). What I Mean When I Say Critical AI Literacy. Reflecting Allowed. https://blog.mahabali.me/educational-technology-2/what-i-mean-when-i-say-critical-ai-literacy/

​

Casal, J. E., & Kessler, M. (2023). Can linguists distinguish between ChatGPT/AI and human writing?: A study of research ethics and academic publishing. Research Methods in Applied Linguistics, 2(3), 100068. https://doi.org/10.1016/j.rmal.2023.100068

​

Dumin, L. (2023, October 13). AI in Higher Ed: Using What We Already Know About Good Teaching Practices. EdSurge. https://www.edsurge.com/news/2023-10-13-ai-in-higher-ed-using-what-we-already-know-about-good-teaching-practices

​

Gegg-Harrison, W. (2021). Encouraging playful, productive curiosity about language in the writing classroom. Journal of Teaching Writing, 36(1), 159–195.

​

Lang, J. (2013). Cheating Lessons: Learning from Academic Dishonesty. Harvard University Press.

Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7). https://doi.org/10.1016/j.patter.2023.100779

​

Shapiro, S. (2022). Cultivating Critical Language Awareness in the Writing Classroom. Routledge.

Shapiro, S., & Leopold, L. (2012). A critical role for role-playing pedagogy. TESL Canada Journal, 29(2), 120.

​

Tardy, C. M. (2021). The potential power of play in second language academic writing. Journal of Second Language Writing, 53, 100833. https://doi.org/10.1016/j.jslw.2021.100833

references & resources

Title