The Fascinating Difficulty of What Seems Easy: Complicating AI Outputs

Brent Duckor and Carrie Holmberg recently shared an interesting concept from their forthcoming book AI and Deeper Learning: Promises, Paradoxes and Evolving Practices. The framework describes an “interpassive-interactive learning tradeoff” in AI-mediated education. The core concern is that AI can create the appearance of cognitive engagement while bypassing some of the developmental processes through which learning reorganizes understanding.

The framework reminded me of Slavoj Žižek’s discussion of interpassivity through the example of the sitcom laugh track. For Žižek, the laugh track does more than “laugh for us.” It pre-structures the affective relation to the scene in advance. The apparatus partially performs enjoyment on behalf of the viewer before they actively and reflectively engage the show. One can sit passively through the show without laughing at all and nevertheless experience the sense that the enjoyment has already taken place.

Something similar may be happening in AI-mediated cognition. AI systems do more than provide answers or complete tasks. As the laugh track does with laughter, LLMs can situate thought preemptively by generating rhetorically persuasive outputs before students encounter the ambiguity, contradiction, estrangement, or conceptual difficulty that would otherwise provoke interpretive labor.

Many teachers worry that AI removes the difficulty and struggle of learning. But that concern often assumes that AI outputs effectively perform the work. I would complicate this: AI can make weak arguments feel complete before students encounter the contradictions, ambiguities, and missing relations that would otherwise force them to question, complicate, and develop their thinking.

As an English teacher, I often argued against the idea that popular texts were pedagogically “easy.” Conventional wisdom is that "literature" presents the necessary difficulty that requires students to wrestle with understanding and their own assumptions. Yet, I turned this argument on its head: In many ways, popular texts were actually harder to teach precisely because they appeared so natural, familiar, and self-evident. Students had to learn how to denaturalize what already felt obvious. It is very difficult to complicate what seems easy.

AI-generated discourse intensifies this problem dramatically. These systems can produce rhetorically sophisticated outputs that appear conceptually developed while still omitting or failing to establish the basic elements of effective argument.

This is why working effectively with AI can become extraordinarily demanding even for advanced scholars. AI literacy may increasingly require capacities many curricula do not explicitly cultivate: The ability to more independently generate ambiguity where discourse appears resolved, to denaturalize rhetorically persuasive language, and to create the complexity necessary for learning.