AI is coming for critical thinking. Teachers are fighting back.

On a sweltering May day in lower Manhattan, about 50 teachers gathered in an air-conditioned room at the AFT’s National Academy for AI Instruction and scrutinized two side-by-side photos of Pomskies—a dog breed that is half Pomeranian, half husky—and debated which one was real and which was created by artificial intelligence.

In their groups, they looked for clues. Did the dog on the left have crooked legs? Why didn’t the throw pillow it was sitting on crease under the dog’s weight? Was there visible dust in the air? The exercise was designed to test their AI-detection skills: One dog was the beloved pet of someone in the room. The other had been generated from a prompt.

There were no obvious giveaways. The teachers were split.

It turned out the seemingly weightless dog with the deceptively crooked legs was real. She belonged to Maria Elena Guzman, an instructor at the academy, where educators were participating in a training session called “The Thinking Classroom: Navigating Critical Thinking and Cognitive Offloading in the Age of AI.”

“Pomskies are smaller than you think,” Guzman said, explaining the throw pillow.

The exercise underscored the mountain range of challenges teachers are facing right now: How do educators prepare students for a world where seeing isn’t believing? How do they ensure students can use a tool that future jobs will likely require, and teach them to question the veracity of its output at the same time? And how do they design assignments and lesson plans that require students to think and do the work themselves?

Those questions are why the AFT, alongside its New York City affiliate, the United Federation of Teachers, partnered with Microsoft, OpenAI and Anthropic to launch the academy: an educator-led effort to help teachers understand how to navigate AI and protect classrooms as places where students still read deeply, write clearly, reason through problems and learn to think for themselves.

The struggle is the point

Teaching students the value in learning to read, write and reason—even when AI can do those things in seconds—is what brought UFT member Sam Chaney, an eighth-grade English and language arts teacher in Brooklyn, N.Y., to the session at the academy.

“The thing that concerns me the most about AI is the degree to which our students see it as a way to cognitively offload,” he says, referring to the concept of delegating mental tasks to something external. Not so long ago, cognitive offloading meant writing down appointments on a calendar or using a calculator to balance a bank account. In the age of generative AI, it can mean using ChatGPT or Claude to write essays, puzzle through logic and math problems, or even turn long texts into podcasts so reading isn’t required at all.

That is where educators like Chaney see the risk. Some forms of cognitive offloading, like a calendar, help manage information. Others, like using AI for reading, writing and reasoning, allow students to bypass the critical-thinking muscles school is meant to build. And that can lead to atrophy.

“The thing I tell my students most often,” Chaney says, “the one thing I hope they learn from me, is what Frederick Douglass said best: ‘If there is no struggle, there is no progress.’ Learning is often a struggle, and that’s OK. That’s part of the point. The struggle is necessary to learn.”

Preparing students for an AI world—whether we want one or not

The irony of AI is that it does not eliminate the need for critical thinking. It increases it, and demands that students are sharper, faster and equipped with a more varied set of tools.

For example, academy instructor and UFT member John Holleran says, AI always appears confident in its answers. The problem is that it can also be wrong, biased, flat-out fake or all three at once.

To demonstrate his point, Holleran noted that in a response about student mental health data, an AI chatbot cited “a recent Harvard study.” When Holleran pushed it to be specific, the chatbot had to admit the study didn’t exist. That puts educators in the position of instilling in their students the concept that even if it walks like a duck, looks like a duck and quacks like a duck, you still have to prove it’s a duck.

In the case of the chatbot’s nonexistent Harvard study, Holleran was dealing with an answer provided by a prompt, which is an example of AI we “opt in” to, said Seth Reznik, a director with Microsoft and frequent academy instructor. The world is full of AI we don’t opt in to—and can’t opt out of—like mis- and disinformation campaigns.

In New Hampshire’s 2024 primary, for example, voters received robocalls what sounded like then-President Joe Biden’s voice telling them not to vote. Election officials proved it was an AI-generated imitation, fabricated as part of a larger voter-suppression campaign.

As a result, whether they want to or not, students need to learn to function in a world where deception is the point. It is an inherently cynical starting point that runs the risk of eroding a foundation for hope. So what is a teacher left with?

Teach students that there is trustworthy information, they just need the skills to parse it out, Reznik says. “Demand provenance,” he says, “not just plausibility. Before we ask, ‘Is this real?’ we have to ask, ‘Where did this come from?’ Separate doubt from dismissal. Uncertainty is a reason to investigate, not disengage.”

Adapting classroom teaching to outsmart AI

Another layer of irony is that AI doesn’t make teachers less necessary, despite the appearance of a robot billed as a teacher at the White House. It makes teachers more necessary—but it also changes the way classroom teaching will look from here on out.

For example, the session suggested adding steps to lesson-planning that analyze the risk of cognitive offloading so teachers can redesign accordingly. It equipped teachers with methods to use in the redesign: Socratic seminars that require students to work through a problem in real time through student-led discussion and open-ended questions. Error analysis, where students are given the wrong answer and shift from producing an answer to evaluating one. Peer teaching, where students study a topic and teach the concept to another small group. Short, in-class, handwritten writing assignments. Requiring students to show their reasoning, not just the answer.

Phillip Perry, a high school English and language arts special education teacher in the Bronx, noted that tools like these are why he came to the academy.

“Are my students using AI? Yes, and they’re using it when I don’t want them to,” Perry says. “Right now, the only move I have is for assignments to be handwritten, but I’m looking forward to more professional development, where I can get some more instructional moves to subvert AI and really get students thinking.”

The stakes, he says, could not be higher.

“I’m worried that my students’ ability to think, particularly their ability to imagine, is being robbed from them,” he says. “A child who cannot imagine also cannot have vision, and they need vision in order to have a future. If they can’t think about what they want their future to look like because they haven’t strengthened those muscles, they can’t build toward something bigger, something better.”

[Melanie Boyer]