Post
JA EN

Learning Methods in the AI Era: Balancing Dependency Risks and Effective Utilization

Posted
By
9 min read
Learning Methods in the AI Era: Balancing Dependency Risks and Effective Utilization

Overview

As AI tools like ChatGPT and Claude rapidly penetrate educational settings, university student AI usage rates surged from 66% in 2024 to 92% in 2025. However, the latest research shows that excessive dependence on AI may lead to declining learning abilities and atrophy of critical thinking. Based on academic research published in 2024-2025, this article proposes practical learning strategies that maximize AI’s potential while avoiding its risks.

Risks of AI Dependency

Decline in Learning Effectiveness: AI as a “Crutch”

A large-scale study by Bastani et al. at the University of Pennsylvania examined the effects of GPT-4 learning support with approximately 1,000 high school students. The results were surprising.

While having GPT-4 access, student grades improved significantly:

  • Standard GPT-4: 48% grade improvement
  • Learning-optimized GPT Tutor: 127% grade improvement

However, on tests after access was removed, the situation reversed:

  • Students who used standard GPT-4: Grades dropped 17% below students with no usage experience
  • Students who used GPT Tutor: Negative effects significantly mitigated by learning guardrails

This research shows that when students use AI as a “crutch” and it succeeds, their own performance deteriorates. Temporary grade improvement was hiding long-term decline in learning ability.

Cognitive Offloading and Impact on Critical Thinking

Gerlich’s research published in January 2025 investigated the relationship between AI tool usage frequency and critical thinking ability in detail with 666 participants.

Main Findings:

  • Significant negative correlation between frequent AI tool use and critical thinking ability (r = -0.68)
  • Cognitive offloading functions as a mediating factor (correlation with AI use r = +0.72)
  • Younger participants showed higher dependence on AI tools and tended to have lower critical thinking scores compared to older participants
  • Participants with higher education had higher critical thinking skills, partially mitigating AI dependency effects

Cognitive offloading refers to delegating complex thought processes to external tools (in this case, AI). This causes our brains to lose opportunities to develop important thinking skills.

Importance of Self-Explanation

Classic cognitive science research (Chi et al., 1994) showed that self-explanation promotes deep understanding and integration of new information. Students who self-explained during reading demonstrated greater knowledge acquisition and built more accurate mental models.

However, when relying on AI, we tend to skip this important self-explanation process. Because AI provides instant answers, we stop making the effort to explain in our own words.

New Learning Paradigm in a World Where AI is Always Present

The risks mentioned above are important, but they assume “a world without AI.” In reality, AI is already deeply embedded in our lives and will become increasingly “invisible infrastructure.”

Complementary Intelligence

The Moravec’s Paradox pointed out by Yann LeCun: “What’s easy for humans is hard for machines, and vice versa.” This creates a fascinating complementarity between human and AI intelligence.

What AI is good at:

  • Processing and analyzing vast amounts of data
  • Executing complex calculations
  • Maintaining consistent performance
  • Discovering patterns

What humans are good at:

  • Intuitive understanding and context awareness
  • Creative problem solving
  • Ethical judgment
  • Adapting to new situations with minimal examples

Transactive Memory Systems

Transactive Memory Systems (TMS) refers to “knowing what team members know” and being able to access that knowledge when needed.

A groundbreaking study of 180 ICU doctors and nurses (Bienefeld et al., 2023) had surprising findings:

  • Accessing information from AI agents: In high-performing teams, positively promoted new hypothesis generation and speaking-up behavior
  • Accessing information from human team members: Negatively associated with these aspects regardless of team performance

In other words, in appropriately trained teams, information from AI may promote creative thinking more than information from humans. This finding suggests we should treat AI not as a mere “tool” but as a “team member” with knowledge.

Practical Learning Strategies

Based on this research, here are practical strategies for maximizing AI’s potential while avoiding its risks.

1. Self-Explain First, AI Confirmation Later

Before relying on AI, it’s important to try explaining in your own words first.

Practical Example:

  1. After learning a new concept, first write an explanation in your own words
  2. Then ask AI “Is this explanation accurate?”
  3. Receive AI’s feedback and revise your understanding
  4. Finally, explain again in your own words

Through this process, AI becomes something that “confirms understanding” rather than “provides answers.”

2. Use AI as a Training Partner

Treat AI as a “training partner” rather than a “crutch.”

Let’s think with an athlete example:

  • Crutch: Something used temporarily when injured. Dependence causes muscle atrophy
  • Training partner: Something used to become stronger. Proper use improves ability

How to use AI as a training partner:

  • Before solving difficult problems, set aside time to think yourself first (at least 5-10 minutes)
  • Instead of asking AI for answers, ask for hints or approaches
  • After reading AI’s explanation, try to reproduce it yourself
  • Regularly challenge the same tasks “without AI” to confirm your true ability

3. Schedule Regular “AI-Free” Learning Sessions

By practicing critical evaluation of AI-generated information, you can mitigate the effects of cognitive offloading.

Weekly Study Plan Example:

  • Mon, Wed, Fri: Use AI as supplement (self-explain → AI confirmation cycle)
  • Tue, Thu: Study completely without AI (problem-solving on your own)
  • Sat: Practice tests without AI (ability confirmation)
  • Sun: Use AI to reinforce weaknesses and plan next week

By intentionally setting aside time without AI, you can maintain genuine understanding and independent learning ability.

4. Consciously Practice Metacognitive Questions

By explicitly providing metacognitive support, you can improve self-regulated learning and learning experience.

Examples of questions that promote metacognition:

Understanding Confirmation:

  • “If I were to explain what I just learned to an elementary school student, how would I say it?”
  • “What do I need to know to understand this concept?”

Learning Strategy Evaluation:

  • “What’s the most effective way to learn this topic?”
  • “At my current understanding level, what should I learn next?”

Knowledge Application:

  • “How can I apply this concept in real life?”
  • “How is this different from similar concepts I learned before?”

5. Recognize AI as a “Team Member”

From Transactive Memory Systems research, we know AI should be treated not as a mere “tool” but as a team member with knowledge.

Practical Methods:

  • Understand “what does AI know” rather than “what should I ask AI”
  • Clearly distinguish AI’s strengths (data processing, pattern recognition) and weaknesses (context understanding, ethical judgment)
  • Use multiple AI tools (Claude, ChatGPT, Perplexity, etc.) as team members with different expertise

Concrete Examples:

  • Claude: Long text analysis, complex reasoning, code generation
  • ChatGPT: General questions, brainstorming
  • Perplexity: Latest information search, fact-checking

6. Consciously Choose Cognitive Augmentation

Harvard researchers point out an important distinction: “There are absolutely opportunities to offload cognitive labor to AI. And there are absolutely opportunities to be cognitively augmented. Our obligation as individuals and as educators is to find that augmentation, not that replacement.”

Cognitive Replacement (To Avoid):

  • Leave everything to AI, think nothing yourself
  • Accept AI output as-is
  • Become unable to do anything without AI

Cognitive Augmentation (To Aim For):

  • Accelerate your thought processes with AI
  • Use AI to gain new perspectives
  • Deepen your understanding through dialogue with AI

Practical Example: Programming Learning

❌ Cognitive Replacement: “Complete this program” → Copy-paste → Done

✅ Cognitive Augmentation: “Teach me 3 approaches to this algorithm” → Understand pros and cons of each approach → Try implementing yourself → Ask AI for hints when stuck → After completion, request code review from AI

Summary

This article explored two important perspectives on AI and learning.

Perspective 1: Risk Awareness

  • Excessive AI dependence degrades learning ability (Bastani et al., 2025)
  • Cognitive offloading impairs critical thinking (Gerlich, 2025)
  • Self-directed learning processes like self-explanation are essential (Chi et al., 1994)

Perspective 2: Utilizing Potential

  • AI functions as a new learning partner
  • Human-AI complementary intelligence can be leveraged
  • In properly trained teams, AI promotes creative thinking (Bienefeld et al., 2023)

What’s important is balancing both of these.

Educational technology experts state: “AI will become invisible infrastructure like electricity or the internet. We won’t think about ‘learning with AI’—learning will be AI-enhanced by default.”

In this new reality, what we need is:

  • “Ability to not use AI”: Foundational skills, critical thinking, independent learning
  • “Ability to effectively use AI”: Appropriate instructions, evaluation, collaboration orchestration

By honing both abilities, you can become a true learner in the AI era. AI is meant to complement cognitive engagement, not replace it. Keep this principle in mind and enjoy learning in the AI era.

References

Academic Papers

Official Documents

Technical Resources

Other Sources

This article is based on peer-reviewed academic papers in educational psychology, cognitive science, and AI education published from 2024 to October 2025. All claims and statistical data are cited from the reliable sources listed above.

Article Creation Date: October 23, 2025

Education
AI learning-strategy Critical-Thinking Metacognition educational-technology
This post is licensed under CC BY 4.0 by the author.