Posted by: Joseph Van Harken - Innovator in Residence | Categories: Emerging Tech, AI, AI Collective, Data Science, Higher Ed Pedagogy, Cognitive Science
If you feel like the conversations surrounding Artificial Intelligence are moving at a breakneck speed, you aren’t alone. In higher education, we constantly monitor how rapidly large language models evolve and redefine classroom, research, and workplace dynamics.
Lately, a monumental shift has occurred. With the rollout of next-generation frontier models, such as OpenAI 5.5 and Anthropic's Opus 4.7, the traditional concept of "prompt engineering" is rapidly becoming obsolete. What used to be a highly sought-after technical skill is now just table stakes.
In a recent strategy analysis, tech analyst Nate B. Jones argued that treating AI as a junior assistant, where you must painstakingly spell out every minor step of a task, is a 2025 mindset. In 2026, today's models are easily 100 times more powerful than their predecessors in how they handle data, utilize tools, and navigate long-context windows.
To unlock their full potential for research, administration, and coursework, we have to change our mental model: AI is no longer an intern; it is a senior partner on your team.
The Digital Native Dilemma: Confronting the "Brain Rot" Trend
This evolution isn’t just a trend among tech influencers; it directly addresses a growing crisis in the learning community. Recently, news feeds have been flooded with alarming headlines regarding Gen Z and early cognitive decline, a phenomenon colloquially dubbed the "brain rot epidemic."
The data backing these concerns is stark:
- The Self-Reporting Spike: A seminal 2025 study from the Yale School of Medicine found that self-reported cognitive issues among young adults (ages 18 to 34) nearly doubled over the last decade, leaping from 5.1% to 9.7%.
- The Student Concern: According to a March 2026 report by the RAND Corporation’s American Youth Panel, student reliance on AI for homework jumped to 62% by the end of 2025. Crucially, 67% of those students explicitly expressed concern that overusing AI would actively harm their own critical thinking skills.
Cognitive scientists refer to this trap as "cognitive offloading", the habit of outsourcing the actual act of thinking to an external tool. A joint May 2026 study conducted by researchers at CMU, Oxford, MIT, and UCLA revealed that when users rely on generative AI for direct answers, their independent problem-solving capabilities drop dramatically. When the AI assistant was removed after just 10 minutes of use, the test-takers' problem-solving accuracy plummeted by 20% compared to a control group, and their likelihood to simply give up and skip hard questions doubled.
Looking closer, the researchers discovered a critical silver lining: those who did not ask for direct answers, but instead used the AI for hints, iterations, and conceptual clarifications, experienced zero cognitive drop-off. The tool itself isn’t causing the decline; our passive, transactional relationship to it is.
The Pedagogical Fix: From Prompt Engineering to Problem Formulation
To reverse this trend of "cognitive stunting," academic institutions like MIT Sloan are actively shifting university vocabulary away from prompt engineering toward Problem Formulation.
While prompt engineering focuses on the specific syntax used to coax a surface-level response from a machine, Problem Formulation is a higher-order cognitive skill. It focuses on identifying a problem, discovering its core focus, setting its scope, and defining its boundaries.
This mirrors the core philosophy behind AI pioneer Andrew Ng's 2026 Frameworks, which emphasize shifting AI's role from an automated "answer generator" into an authentic "thought partner." By forcing students and professionals to intellectually frame a problem before touching the keyboard, we transform AI from a crutch that causes cognitive decline into a barbell that builds cognitive resilience.
How to Guide a "Senior Partner": Three Core Principles
To bridge the gap between classroom pedagogy and practical workflows, we can adopt Nate Jones’s AI Question Method, transforming our interactions with modern models into a rigorous exercise in critical thinking:
-
The Flashlight Approach (Define Focus & Boundaries)
When collaborating with a senior colleague, you don't give them a completely blank slate, nor do you micromanage them. You give them a vector. Your instructions should act like a flashlight: providing a brilliant, clear focus at the center, while acknowledging the softer edges of the problem space.
Instead of asking a flat, open question like "Help me research marketing automation," provide a thesis and hard guardrails:
"I have a thesis that our student engagement data is dropping because our onboarding emails are improperly segmented. I want you to investigate this dataset with that angle in mind. Note: Ignore the data from the 2024 pilot program, as those metrics are skewed."
This forces the user to define the parameters of the problem first, giving the AI a clear target while leaving it room to explore, synthesize, and push back on assumptions.
-
Multi-Layered Synthesis over Flat Evaluation
When undertaking complex knowledge work, such as drafting a comprehensive project proposal, a syllabus, or a research paper, stop asking the AI to just generate text based on flat, automated templates. Construct multi-layered questions that force the model to contend with opposing variables and synthesize a solution.
For example, if you are designing a new digital campus experience, don't just ask for a feature list. Ask: "How do we weave a high-tech immersive experience into our library infrastructure while ensuring it remains fully accessible to users who have never interacted with AR/VR technologies?" By prompting the AI to wrestle with the friction between the technology's capability and the human experience, the user engages in active synthesis rather than passive consumption.
-
Anchor Intent Across Data Ecosystems
Today’s enterprise AI tools allow us to point models toward entire directories of diverse files, Excel sheets, meeting transcripts, PDFs, and repositories all at once. However, a common pitfall is that the AI might hyper-focus on just one file and ignore the broader context.
To counter this, your inquiries must intentionally weave your data artifacts together with your qualitative opinions. Frame your question in a way that forces the model to look across the entire data ecosystem:
"Looking across the student support transcripts, the quantitative retention spreadsheets, and our department's original mission document, how does our current trajectory align with our goals? Challenge my perspective if the data suggests our current strategy is failing."
The Innovation Takeaway for GVSU
As we integrate these advanced AI capabilities into our academic and operational ecosystems at GVSU, our success won't be defined by mechanical "cheatsheets." It will be defined by our curiosity, critical thinking, and problem formulation skills.
If we treat AI like a capable, senior researcher, give it high-leverage context, and ask questions that demand deep synthesis, it becomes an incredibly transformative collaborator for faculty, staff, and students alike. We aren't outsourcing our brains to the machine; we are leveling up our abilities to manage them.
How are you shifting your workflows to keep your critical thinking sharp as AI models become more autonomous? Let us know in the comments below, or reach out to the IT Innovation and Research team to collaborate on your next project!
References & Further Learning
- Watch: "Opus 4.7 and OpenAI 5.5 Made Your Prompting Style Obsolete" by Nate B. Jones.
- Watch: "Full AI Prompting Course with Andrew Ng (2026)" for updated strategies on AI thought-partnership.
- Read: Fast Company / Yale School of Medicine (2026). "Gen Z reports early cognitive decline. Here's what to know about the brain rot epidemic."
- Read: RAND Corporation Research Report (March 2026). "More Students Use AI for Homework, and More Believe It Harms Critical Thinking."
- Read: Carnegie Mellon University, Oxford, MIT, & UCLA Joint Study (May 2026). "AI Assistance Reduces Persistence and Hurts Independent Performance."
- Read: MIT Sloan Teaching & Learning Technologies. "Effective Prompts for AI: The Essentials."