Building AI-Resilient Classrooms Part 1: Measuring AI Over-Reliance
Most schools treat generative AI tools as a distraction or method of cheating. However, a 2025 study from researchers at Singapore Management University and the National University of Singapore found that overdependence on AI creates a “cognitive crutch” that replaces our ability to think, plan and solve problems independently over time.
Caption: Student using an AI chatbot to get advice about school, friendships and life decisions.
These researchers realized that it was easier to see the cognitive crutch in students than measuring it. Previously, researchers modified tests for social media and smartphone addictions, which failed to account for fundamental differences in engagement (i.e., passive engagement on social media vs. generative AI as a cognitive thought partner).
The result? The creation of the Generative AI Dependency Scale (GAIDS), a simple 11-question survey with each question evaluated on a 5-point Likert scale. Tested by over 1,300 participants across the United States and Singapore, GAIDS was found to be both reliable and consistent in measuring generative AI dependency.
GAIDS measures over-reliance across three core dimensions:
Cognitive Preoccupation: the degree to which generative AI is crucial in someone's thoughts, routines or decision-making processes. This dimension measures if the user is experiencing a “reflex” where AI is the default starting point for any task, problem, or decision. It can show up as compulsive prompting, surrendered control of decision-making, and/or intrusive thoughts.
Negative Consequences: the measurable decline in someone’s skills, self-efficacy and task performance as a result of dependency. This dimension measures what happens when the user continually delegates their cognitive load to AI. It can show up as eroded personal confidence, a decrease in quality of work and/or relying on AI to bypass the discomfort of beginning a task.
Withdrawal: the negative psychological effects experienced when access to generative AI tools is removed or restricted. Unlike other forms of physical withdrawal, AI withdrawal is based on the loss of cognitive scaffolding. It shows up as frustration, anxiety, feeling lost, and/or an inability to cope.
One important quality of GAIDS is its focus on skill loss, not screen time. GAIDS concentrates on the amount of cognitive load and agency transferred to these tools rather than the correlation between time spent using generative AI tools and AI overdependence. This shifts the conversation from how much AI is used to what for and when it is used. This allows educators and administrators to focus on building critical thinking and problem-solving with AI as a collaborative partner, not as the decision-maker.
The creation of GAIDS provides educators, school administrators, and leaders with a way to audit where AI shifts from collaborative partner to decision-maker. Beyond assessing dependency, GAIDS has implications for AI literacy, demonstrating that AI literacy is defined by evaluating models, problem-solving, and critical thinking, not simply the ability to produce outputs
