Artificial or Human Intelligence?

2025-09-02

AI Summary

This paper examines the nonlinear effects of large language models (LLMs) on learning motivation and human knowledge accumulation, addressing accuracy misjudgments arising from AI hallucinations and capability emergence. Method: We develop an educational economics incentive model to analyze discontinuities in student effort induced by AI capability thresholds—challenging the conventional assumption of continuous AI impacts in labor markets—and combine behavioral experiments with theoretical modeling. Contribution/Results: We find that students systematically overestimate AI reliability, leading to reduced effort and potential declines in aggregate human–AI collaborative knowledge. Innovatively, we propose a “proportion of AI-infeasible tasks” regulatory mechanism, demonstrating that strategically restricting AI usage domains can recalibrate student beliefs and restore optimal learning incentives. The study provides a theoretically grounded, cognitively bounded framework for AI governance in education, yielding actionable design principles rooted in incentive structures and epistemic constraints.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Philosophy and Ethics of AI: AI & Epistemology
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Artificial intelligence (AI) tools such as large language models (LLMs) are already altering student learning. Unlike previous technologies, LLMs can independently solve problems regardless of student understanding, yet are not always accurate (due to hallucination) and face sharp performance cutoffs (due to emergence). Access to these tools significantly alters a student's incentives to learn, potentially decreasing the sum knowledge of humans and AI. Additionally, the marginal benefit of learning changes depending on which side of the AI frontier a human is on, creating a discontinuous gap between those that know more than or less than AI. This contrasts with downstream models of AI's impact on the labor force which assume continuous ability. Finally, increasing the portion of assignments where AI cannot be used can counteract student mis-specification about AI accuracy, preventing underinvestment. A better understanding of how AI impacts learning and student incentives is crucial for educators to adapt to this new technology.

Problem

• AI tools alter student learning incentives and knowledge acquisition
• LLMs create discontinuous gaps in human-AI knowledge frontiers
• Educators need strategies to counteract AI's impact on learning

Innovation

• Uses large language models to solve problems independently
• Addresses AI hallucination and sharp performance cutoffs
• Adjusts assignments to counteract student AI mis-specification