USC Built a Five-Level AI Framework for Students. They Forgot the Ground Floor.
by BrainStream CEO Steve Alcorn
USC’s Sol Price School of Public Policy just published a detailed case study about integrating AI into their curriculum. Their instructional designer, Minh Trinh, built a framework that scaffolds students from basic AI use (summarizing readings) through increasingly sophisticated engagement—using AI as a debate partner, a mentor that asks questions instead of giving answers, and eventually having students manage and grade AI-generated drafts.
It’s smart work. The framework is sound, the progression is thoughtful, and the emphasis on critical thinking over passive consumption is exactly right. They even require students to write “AI usage reflections” explaining how the tools shaped their thinking. That turns a cheating concern into a learning moment.
But there’s a gap in the framework that no one at USC seems to have noticed.
Every level of their AI integration assumes the student has already done the reading. The AI summarizes a text—but did the student read it first, or did they skip straight to the summary? The AI debates a policy position—but is the student arguing from knowledge of the source material, or from whatever the AI fed them two minutes ago? The mentor chatbot asks guiding questions—but guiding toward what, if the student never engaged with the assigned content?
This is the accountability problem, and it’s the one thing AI integration frameworks consistently ignore.
You can build the most sophisticated AI scaffolding in the world. You can require reflections, mandate source verification, design rubrics that reward critical thinking. None of it matters if the student never actually read the book.
Trinh’s framework has five levels of AI engagement. Every one of them assumes a foundation that isn’t verified. It’s an impressive building with no ground floor.
That’s the problem BrainStream solves. Before any AI-assisted analysis, before any debate or draft management or mentor chatbot interaction, BrainStream confirms that the student engaged with the actual assigned material—the specific text the teacher chose, not a summary, not a paraphrase, not an AI-generated substitute. The system tests comprehension of the real content in real time.
USC’s approach and BrainStream’s approach aren’t in competition. They’re sequential. Accountability comes first. Then the AI scaffolding has something real to build on. Without that foundation, even the best framework is asking students to think critically about material they may never have encountered.
The question isn’t whether AI belongs in education. It does. The question is whether we’re verifying that students bring something to the conversation before we hand them a thought partner.
