New research from BYU maps 8 distinct modes of human-AI interaction, ranked by cognitive agency. The framework pulls from Bloom's taxonomy, Chi's ICAP model, and Vygotsky's ZPD theory to predict long-term skill retention vs. atrophy.
The gradient:
Passivity tier (modes 1-2): Oracle mode = treating AI as authoritative answer machine. Production mode = offloading creation with minimal verification. Both map to Bloom's "remembering" level. Short-term gains, long-term skill decay.
Partnership tier (modes 3-4): Tutor mode = scaffolded learning within ZPD. Collaborative Problem-Solver = distributed cognition where human directs, AI executes. Shared cognitive load.
Agency tier (modes 5-8): Verification Agent = reactivating epistemic vigilance, cross-referencing claims. Critical Challenger = adversarial reasoning, forcing AI to defend positions. Creative Expander = human-directed divergent exploration.
The core finding: interaction mode determines whether AI amplifies or atrophies your thinking. Lower modes optimize for immediate output but train dependency. Higher modes force metacognition and preserve skill transfer when AI is removed.
The framework suggests most users default to modes 1-2 because fluent AI output bypasses our evolved skepticism of communicated information. The cost is invisible until you try to solve problems without the tool.
Practical implication: if you're using AI for production work, deliberately shift up the gradient periodically. Verify outputs, challenge reasoning, reframe problems yourself. Otherwise you're training a cognitive dependency that compounds.
The gradient:
Passivity tier (modes 1-2): Oracle mode = treating AI as authoritative answer machine. Production mode = offloading creation with minimal verification. Both map to Bloom's "remembering" level. Short-term gains, long-term skill decay.
Partnership tier (modes 3-4): Tutor mode = scaffolded learning within ZPD. Collaborative Problem-Solver = distributed cognition where human directs, AI executes. Shared cognitive load.
Agency tier (modes 5-8): Verification Agent = reactivating epistemic vigilance, cross-referencing claims. Critical Challenger = adversarial reasoning, forcing AI to defend positions. Creative Expander = human-directed divergent exploration.
The core finding: interaction mode determines whether AI amplifies or atrophies your thinking. Lower modes optimize for immediate output but train dependency. Higher modes force metacognition and preserve skill transfer when AI is removed.
The framework suggests most users default to modes 1-2 because fluent AI output bypasses our evolved skepticism of communicated information. The cost is invisible until you try to solve problems without the tool.
Practical implication: if you're using AI for production work, deliberately shift up the gradient periodically. Verify outputs, challenge reasoning, reframe problems yourself. Otherwise you're training a cognitive dependency that compounds.