Post
JA EN

Unlearning and Relearning in the AI Era — The 'Art of Letting Go' and 'Art of Relearning' for Adapting to Change

Posted
By
11 min read
Unlearning and Relearning in the AI Era — The 'Art of Letting Go' and 'Art of Relearning' for Adapting to Change
  • Target audience: Software engineers, developers using AI tools
  • Prerequisites: Basic experience using AI tools
  • Reading time: 12 minutes

Summary

“I’m good at learning new technologies, but it’s hard to let go of old habits” — many engineers probably feel this way.

In the AI era, the importance of unlearning (consciously letting go of old knowledge and habits) and relearning (learning anew in new contexts) is growing beyond just learning. As technological change accelerates, past success experiences can sometimes become obstacles rather than advantages.

This article explains the cognitive science background of unlearning and relearning, and examines what specifically to let go of and what to relearn in the AI era.

What Is Unlearning?

“Letting Go” Rather Than “Forgetting”

Unlearning is not simply forgetting information. It is the process of consciously letting go of old patterns and beliefs to become open to new approaches1.

flowchart TB
    subgraph Wrong["❌ Common Misconceptions"]
        direction LR
        W1["Unlearning = Forgetting"]
        W2["Erasing knowledge"]
        W3["Denying the past"]
    end

    subgraph Right["✅ Correct Understanding"]
        direction LR
        R1["Unlearning = Conscious letting go"]
        R2["Relativizing old patterns"]
        R3["Accepting new options"]
    end

In organizational learning research, unlearning is defined as “the process of discarding old knowledge and established routines and escaping from competency traps that block innovative activities”1.

The important point is that you don’t need to erase old knowledge itself. By relativizing old methods as “one option” rather than “the only right answer,” you become able to adopt new approaches.

Why “Just Learning New Things” Isn’t Enough

In cognitive psychology, a phenomenon called proactive interference is well known2. This is where previously learned information interferes with learning new information.

Proactive interference is a phenomenon where “previously learned but no longer valid information interferes with learning and retrieving new information”2. This differs from simple forgetting over time — it involves competition between old and new memories.

For example:

  • When learning a new programming language, syntax from previous languages interferes
  • When using a new framework, you’re pulled toward patterns from old frameworks
  • When adapting to new workflows, old habits get in the way

Simply “adding” new knowledge creates competition with old knowledge. By consciously “letting go” of old patterns, this interference can be reduced.

Why Is This Particularly Important in the AI Era?

Accelerating Technological Change

The World Economic Forum predicted that by 2025, 50% of all employees would need reskilling3. In five years, two-thirds of the skills currently important for today’s jobs will change, with one-third of essential skills composed of technology-related competencies.

For engineers, this means the importance of not only “learning new technologies” but also “not clinging to old methods.”

Past Success Patterns No Longer Work

With the advent of AI tools, many development practices previously considered “correct” are being forced into reconsideration.

Traditional Success PatternAI Era Change
“Write all code yourself”Productivity improvement through AI collaboration
“Time writing code = productivity”Design, verification, prompt creation also have value
“Deep mastery of specific technology”Cross-tool understanding and adaptability
“Years of experience = expertise”Speed of adapting to latest technology also matters

Adapting to these changes requires not just learning new skills but consciously reviewing old beliefs.

The Difference Between AI and Human Unlearning

Interestingly, 2025 Princeton University research revealed differences between human brain learning and AI learning4.

The human brain can efficiently learn new skills by modularly assembling “cognitive blocks” that are reusable across tasks. In contrast, current AI models tend to forget old skills when learning new ones (catastrophic forgetting).

“If you know how to bake bread, you can learn to bake a cake without learning to bake from scratch”4

This “compositionality” is a human strength, demonstrating the ability to build new skills while leveraging existing knowledge. Unlearning means letting go of old “fixed ideas” that obstruct this compositionality.

Cognitive Science Background — Cognitive Flexibility and Age

What Is Cognitive Flexibility?

Cognitive flexibility is the ability to switch thinking processes and adapt to diverse, dynamic situations5. It can be considered the cognitive function underlying unlearning.

This function is supported by networks of brain regions including the dorsolateral prefrontal cortex (dlPFC), dorsomedial prefrontal cortex (DMPFC), anterior insular cortex, and posterior parietal cortex5.

Changes with Age

Unfortunately, cognitive flexibility tends to decline with age. According to the frontal aging hypothesis, the prefrontal cortex decreases by an average of about 5% per decade, particularly in older adults6.

flowchart TB
    subgraph Timeline["Cognitive Flexibility by Age"]
        direction TB
        T1["20s-30s<br>High flexibility<br>Easy to accept new approaches"]
        T2["40s<br>Still high flexibility<br>Optimal time for unlearning"]
        T3["50s and beyond<br>Declining flexibility<br>Integration with existing knowledge is effective"]
    end

    T1 --> T2
    T2 --> T3

2021 research showed that decline in working memory and inhibitory control begins at the relatively early stage of 30-40 years old7.

This is not bad news, but motivation to start countermeasures early. By establishing unlearning habits while cognitive flexibility is high, you can maintain future adaptability.

Proactive Interference Occurs Regardless of Age

On the other hand, research shows that proactive interference itself occurs regardless of age2. Both younger and older adults are similarly affected by past learning interfering with new learning.

In other words, the need for unlearning exists regardless of age. However, unlearning is “easier” while cognitive flexibility is high.

What Engineers Should Unlearn

Let’s specifically list the beliefs and patterns that engineers should particularly review in the AI era.

1. “Code Should Be Written by Yourself”

Old belief: Great engineers write all code themselves. Copy-pasting or code generation is “cheating.”

Direction to reconsider: Collaboration with AI is a new form of expertise. What matters is not “who wrote it” but “does it work correctly,” “is it maintainable,” and “do you understand the intent.”

Unlearning point: Let go of attachment to “writing with your own hands” and revalue “efficiently producing correct code.”

2. “Time Writing Code = Productivity”

Old belief: The longer you’re typing, the more work you’re doing.

Direction to reconsider: Design, review, testing, documentation, and appropriate prompt creation for AI are all valuable work.

Unlearning point: Update your evaluation criteria from “typing time” to “value creation time.”

3. “Should Understand Everything”

Old belief: Great engineers know all the details of the system.

Direction to reconsider: In an era where AI complements details, the human role is shifting to grasping the big picture and making decisions.

Unlearning point: Let go of insistence on “knowing everything” and emphasize the ability to “access the right information when needed.”

4. “Experience Is Correct”

Old belief: Years of experience always lead to correct judgment.

Direction to reconsider: Experience is valuable, but when the technical environment changes, rules of thumb may no longer apply. Judgment based on data and current conditions is important.

Unlearning point: Put “it used to be this way” thinking on hold and develop the habit of confirming “how is it now.”

5. “New Things Are Unstable”

Old belief: New technologies and tools should be avoided until they mature.

Direction to reconsider: At the pace of AI-era change, things can become obsolete while you’re waiting. The stance of evaluating new technologies while managing risk is important.

Unlearning point: Reconsider the automatic reaction of “new = dangerous” and evaluate fairly.

Relearning — What to Learn After Letting Go

Unlearning is not the goal but preparation for relearning. After letting go of old patterns, what should you newly acquire?

The Learning-Unlearning-Relearning Cycle

In organizational learning research, “learning-unlearning-relearning” is understood as continuous steps1.

flowchart TB
    subgraph Cycle["Continuous Adaptation Cycle"]
        direction TB
        L["Learning<br>Acquire new knowledge and skills"]
        U["Unlearning<br>Let go of old patterns"]
        R["Relearning<br>Relearn in new contexts"]
    end

    L --> U
    U --> R
    R --> L

Skills to Relearn in the AI Era

Unlearning TargetRelearning Target
“Write everything yourself”AI output evaluation, verification, editing skills
“Code time = productivity”Effective prompt design skills
“Deep mastery of specific technology”Adaptability across multiple tools
“Should understand everything”Balance of appropriate delegation and verification

AI Utilization as “Scaffolding”

Cognitive science research demonstrates the importance of using AI as a “scaffold” rather than a “substitute”8.

  • Scaffold: Temporary, adaptive, empowering — aims to strengthen internal abilities and gradually reduce dependence on technology
  • Substitute: Permanent, dependent — technology assumes responsibility, reducing intrinsic skills

The goal of relearning is to improve your own skills while utilizing AI as scaffolding.

Practical Methods — Making Unlearning a Habit

1. Regular “Assumption Inventory”

Quarterly, or when major technological changes occur, write out what you take for “granted” and review it.

Questions:

  • When was this assumption formed?
  • Is this assumption still valid in the current technical environment?
  • If this assumption were wrong, what would change?

2. Intentionally Try “The Opposite Position”

Try what you normally avoid in a small scope.

Examples:

  • If you’re “anti-AI reliance,” try intensively using AI tools for a week
  • If you’re “AI dependent,” try working without AI for a day
  • If you’re “TypeScript-only,” try making a small project in a different language

The purpose is not “finding the right answer” but relativizing your assumptions.

3. Intentional Exposure to Different Perspectives

Consciously read content from people with different backgrounds or opinions than yours.

  • Blogs from engineers younger/older than you
  • Communities using different tech stacks
  • Opinions from both AI advocates and skeptics

4. The Habit of Re-asking “Why”

When you think “this is correct,” ask yourself “why do I think so.”

  • Because you actually verified it?
  • Because someone told you?
  • Because it’s always been that way?

5. Countermeasures for Proactive Interference

Research indicates the following methods for reducing proactive interference2:

  • Spaced repetition: Leave intervals between learning sessions
  • Context change: Practice in different environments or situations
  • Inserting tests: Release from proactive interference occurs by inserting tests between learning episodes

When learning new technology, creating context that clearly distinguishes from old methods is effective to prevent “confusion.”

Summary

In the AI era, not just learning but unlearning (consciously letting go) and relearning (learning anew in new contexts) are becoming important.

Key points of unlearning:

  • Not mere forgetting, but relativizing old patterns as “one option”
  • Necessary to reduce proactive interference (old knowledge blocking new learning)
  • Effective to establish as a habit while cognitive flexibility is high

Beliefs to reconsider in the AI era:

  • “Code should be written by yourself” → AI collaboration is also expertise
  • “Code time = productivity” → Design, verification, prompts also have value
  • “Should understand everything” → Balance of appropriate delegation and verification
  • “Experience is correct” → Judgment based on data and current conditions
  • “New things are unstable” → Fair evaluation

Direction of relearning:

  • AI output evaluation and verification skills
  • Effective prompt design
  • Adaptability across multiple tools
  • The stance of utilizing AI as “scaffolding”

In an era of rapid change, what to let go of is as important as what to learn. Unlearning is not weakness but an expression of adaptability.

See other articles related to this theme:

References

References are listed in order corresponding to citation numbers in the main text.

Additional References (Not Numbered in Main Text)

On Citation Accuracy: The research cited in this article has been verified through the following methods:

  • Confirmation via academic databases (PubMed, Google Scholar, etc.)
  • Information verification on official journal websites
  • Cross-verification through multiple independent sources

  1. Organizational unlearning as a process: What we know, what we don’t know, what we should know - Management Review Quarterly (2024). [Reliability: High] (Peer-reviewed) ↩︎ ↩︎2 ↩︎3

  2. How Proactive Interference during New Associative Learning Impacts General and Specific Memory in Young and Old - Journal of Cognitive Neuroscience (2020). [Reliability: High] (Peer-reviewed) ↩︎ ↩︎2 ↩︎3 ↩︎4

  3. Reskilling and Upskilling the Future-ready Workforce for Industry 4.0 and Beyond - PMC, citing World Economic Forum prediction (2022). [Reliability: High] ↩︎

  4. Scientists uncover the brain’s hidden learning blocks - ScienceDaily, Princeton University research (2025). [Reliability: Medium-High] ↩︎ ↩︎2

  5. How to Improve Cognitive Flexibility: Evidence From Noninvasive Neuromodulation Techniques - PMC (2025). [Reliability: High] (Peer-reviewed) ↩︎ ↩︎2

  6. Differential aging of the brain: patterns, cognitive correlates and modifiers - Raz, N. & Rodrigue, K.M., Neuroscience and Biobehavioral Reviews (2006). [Reliability: High] (Peer-reviewed) ↩︎

  7. The developmental trajectories of executive function from adolescence to old age - Scientific Reports (2021). [Reliability: High] (Peer-reviewed) ↩︎

  8. Cognitive Atrophy Paradox of AI–Human Interaction: From Cognitive Growth and Atrophy to Balance - MDPI Information (2025). [Reliability: High] (Peer-reviewed) ↩︎

AI・Technology
AI Career Learning Skills Cognitive-Science
This post is licensed under CC BY 4.0 by the author.