Using AI tools for writing tasks enhances learning and skill development, contrary to widespread concerns that such tools would lead to skill atrophy.

Objective: The study aimed to investigate whether using AI tools for writing improves skill development over time or hinders learning, challenging the common assumption that AI assistance promotes immediate productivity at the expense of long-term human capital development.

Methods: The researchers conducted three pre-registered experiments involving American adults on the Prolific platform. In Study 1, forecasters predicted the effects of AI use on learning. In Studies 2 and 3, participants completed a baseline writing test (revising a poorly written cover letter), learned evidence-based writing principles, and were randomly assigned to different conditions: practicing with an AI writing tool, practicing without AI, seeing only an AI example (Study 3), or no practice (Study 2). Participants completed a writing test without AI assistance immediately after and one day later. GPT-4o rated cover letters on five writing principles, and human raters assessed which letters would more likely secure job interviews.

Key Findings:

• Contrary to forecaster predictions (65% expected manual practice would be better), participants who practiced writing with AI showed greater improvement in their independent writing skills compared to those who practiced without AI.
• Surprisingly, merely viewing an AI-generated example (without editing it) was as effective for learning as practicing with the AI tool, and both conditions outperformed manual practice.
• AI practice improved learning despite requiring less effort - participants using AI spent less time, logged fewer keystrokes, and reported less subjective effort during practice.
• Learning benefits from AI assistance persisted at least one day later, with participants who practiced with AI or saw AI examples continuing to outperform those who practiced manually.
• AI-assisted learning was equally effective across all participant subgroups, regardless of prior AI experience, age, gender, education, or baseline writing skill.
• Participants who used AI tools showed no evidence of developing an "illusion of mastery" - they did not overestimate their abilities or seek less feedback than other groups.

Implications: The study challenges widespread concerns about AI tools reducing learning, demonstrating that by providing high-quality, personalized examples, AI can actually accelerate skill acquisition rather than hinder it. This aligns with research on learning through worked examples and imitation. The findings suggest that AI tools could democratize skill development by making high-quality, just-in-time, personalized examples available to learners who may not have access to expert human teachers or mentors.

Limitations: The study focused on short-term skill development (one day later) in the specific domain of writing cover letters. The experimental setting also explicitly incentivized learning rather than just immediate performance, which might differ from real-world scenarios where immediate performance is often prioritized. Additionally, participants only interacted with the AI tool once, whereas repeated use in real-world settings might yield different outcomes.

Future Directions: Future research should explore whether these learning benefits extend to domains beyond writing, particularly those where solutions aren't as visually informative as writing samples (e.g., mathematics, programming). Researchers should also investigate what interaction strategies with AI maximize learning, how the benefits change with repeated use over longer periods, and whether there are diminishing returns to learning from AI examples over time.

Title and Authors: "Learning not cheating: AI assistance can enhance rather than hinder skill development" by Benjamin Lira, Todd Rogers, Daniel G. Goldstein, Lyle Ungar, and Angela L. Duckworth.

Published On: February 25, 2025

Published By: Not peer-reviewed preprint

The research provides compelling evidence against the assumption that AI tools necessarily lead to skill atrophy. Instead, it suggests that when designed properly, AI can serve as an effective teaching tool by providing tailored examples that help users internalize principles of effective writing. This counters the prevailing narrative in educational settings where many schools have banned or restricted AI tools out of concerns they will undermine learning. The authors highlight that decades before AI, coach John Wooden declared that the four laws of learning are explanation, demonstration, imitation, and repetition - and modern generative AI excels at providing the demonstration component of this learning process.