Last semester, one of my strongest students — sharp, thoughtful, the kind of kid who asks questions I haven’t considered — got flagged by our plagiarism detection software. The report came back with a high “AI-likelihood” score. I knew the work was hers. I’d watched her think through it in class, seen her drafts evolve, heard her argue her thesis out loud during peer review.
The software didn’t know any of that.
I overrode it. But I kept thinking: what happens to students whose teachers don’t know them as well? What happens to students whose writing style — shaped by multilingual backgrounds, by different cultural rhetorical traditions — reads as “unusual” to a model trained on a narrow slice of American academic prose?
That question brought me to Ruha Benjamin’s work. And it’s been sitting with me since.
Who Is Ruha Benjamin, and What Is the “New Jim Code”?
Ruha Benjamin is a Princeton professor of African American Studies and one of the most important voices on the intersection of race and technology. In her book Race After Technology, she introduced a concept she calls the New Jim Code: the way that digital tools — algorithms, automated systems, AI — can encode and perpetuate racial bias while appearing neutral, scientific, and objective.
The name is a deliberate echo of the “New Jim Crow,” Michelle Alexander’s term for how mass incarceration re-inscribed racial hierarchy after the Civil Rights era. Benjamin’s argument is that technology is doing something similar, but faster and at greater scale — and with the added cover of algorithmic legitimacy.
Her core claim: discriminatory design doesn’t require a discriminatory designer. The bias doesn’t have to be intentional. It can be baked in through training data that reflects existing inequalities, through optimization metrics that treat certain outcomes as normal, through the absence of diverse voices in the rooms where these systems get built.
Why This Matters in Education Right Now
You might be thinking: okay, this is relevant to facial recognition or hiring algorithms. But schools?
Yes. Schools. Increasingly and urgently, schools.
Plagiarism and AI detection tools are trained primarily on writing from dominant linguistic and cultural traditions. Students who write in non-standard dialects, who are multilingual, or who use rhetorical structures common in non-Western academic traditions are more likely to be flagged as suspicious — not because they’re doing anything wrong, but because the model doesn’t recognize their patterns as “normal.”
Adaptive learning platforms that place students into learning paths based on prior performance inherit whatever inequities existed in that prior performance. A student who tested lower in third grade due to unstable housing, or a pandemic school year, or a language barrier, gets routed into a slower track — and the algorithm reinforces that trajectory as if it were a natural fact about the student’s ability.
Automated grading systems, especially for writing, consistently score certain student populations lower — not because their work is weaker, but because the rubric was built from data that didn’t include them.
None of these systems were built by people trying to harm students. But intent doesn’t determine impact. Benjamin’s point is that neutrality is a fiction — and in technology, that fiction is dangerous because it stops us from asking the right questions.
What This Looks Like in an International School
I teach in Shanghai. My students are native Chinese speakers with near-native English proficiency. They write beautifully — but sometimes in ways that read as unexpected to Western AI tools. Long, carefully built arguments. Different paragraph structures. Rhetorical moves that are perfectly sophisticated but don’t match the template the detection software learned from.
I’ve had to override automated flags more than once this year.
And my situation is actually pretty good — I know my students, I have time to review, I have professional discretion. Think about students in large, under-resourced schools where teachers are managing 150 students per class and leaning heavily on these tools because they have to. A false flag becomes a disciplinary conversation. A biased learning path becomes a tracking decision. A skewed analytics report shapes a teacher’s perception of a student before they’ve even met.
Scale matters here. When we adopt AI tools school-wide, we’re not just automating a task. We’re systematizing a set of assumptions about what “good” looks like — and those assumptions have a history.
This Isn’t an Argument Against AI
Ruha Benjamin is not a technology pessimist. She’s a critical optimist. Her book ends with a call for “abolitionist tools” — tools designed with liberation in mind, built from the start with equity as a core value rather than an afterthought.
Her argument isn’t that we should abandon AI in education. It’s that we need to be much more honest about how these systems work, who they serve, and who they might harm.
That requires asking different questions when we adopt new tools:
- Who was in the room when this was built?
- What data was it trained on, and who is underrepresented in that data?
- What counts as “good performance” in this system, and where did that definition come from?
- Who gets flagged, and who doesn’t? Are those patterns random, or do they follow demographic lines?
- What human decision-making process exists when the algorithm gets it wrong?
These aren’t questions that require a computer science degree. They’re the same questions good teachers have always asked about curriculum, about texts, about which voices get heard in the classroom. We already know how to do this. We just need to extend the habit to our tools.
What I’m Actually Doing Differently
After sitting with Benjamin’s work, I’ve made a few concrete changes in how I use AI in my classroom.
I no longer accept AI detection flags without a conversation. If a tool flags a student’s work, that’s the beginning of a dialogue, not the end of one. I ask the student to walk me through their process. I look at their drafts. I treat the flag as a prompt for attention, not a verdict.
I’ve started asking vendors harder questions. When I evaluate a new EdTech tool, I now ask about training data, about bias testing, about what populations were included in development. Some companies answer honestly. Some get defensive. Both responses are informative.
I’ve introduced Benjamin’s ideas directly to my students. In an AP World History context, this fits naturally — we study how systems of power get built and maintained, how the language of science and objectivity has been used to justify inequality throughout history. The New Jim Code is just a contemporary example of a very old pattern.
The Takeaway for Teachers
This week, pick one AI tool you’re already using in your classroom — a plagiarism detector, an adaptive platform, an automated grading or feedback tool — and ask yourself: Do I know how this was built?
You don’t need to dig into code. Start with the company’s website. Look for a transparency report, a bias audit, a statement about training data. If you can’t find one, that’s worth noting.
Then ask yourself: Who in my class might this tool get wrong? Think about your multilingual students, your students with IEPs, your students who write in non-standard dialects or use rhetorical structures from different cultural traditions. Would this tool recognize the quality in their work? Would it flag them unfairly?
You don’t have to stop using the tool. But you should use it with your eyes open.
Benjamin’s insight isn’t that technology is racist. It’s that technology reflects the world that built it — and if we want our tools to serve all our students, we have to be the ones who insist on that. The algorithm won’t do it on its own.
David Jacobson teaches AP World History at Shanghai American School. He writes about AI, education, and the messy intersection of the two at shouldiuse.ai. Follow him on LinkedIn and Substack.
