How Humans Learn

Cognitive effort is not a bug—it’s the feature

Andrew Ellis

23 June, 2025

What You Just Experienced

Your Three Research Modes:

Research Mode	Description	How did it feel?
Without AI	Classical sources	Slower, effortful, uncertain
Single AI Prompt	One question	Fast, confident, comprehensive
Collaborative AI	Iterative questions	Engaging, interactive, refined

The Critical Question:

Which mode will you remember and transfer a week from now?

The immediate “feel” doesn’t predict long-term learning…

Why Schools Exist: The Knowledge Divide

🧬 Biologically Primary

We evolved to learn:

• Speaking & understanding
• Recognizing faces
  
• Reading social cues
• Basic problem solving

Natural discovery works!

🎓 Biologically Secondary

We didn’t evolve to learn:

• Reading & writing
• Mathematics & algebra
• Programming & analysis
• Scientific reasoning

Needs explicit teaching & struggle!

Primary knowledge can be learned through natural discovery, while academic skills require explicit instruction and structured practice.

To understand why AI disrupts learning so profoundly, we need to understand why schools exist in the first place.

How Expertise Actually Develops

The Brain’s Learning System:

Memory Type	Description	Characteristics	Example
🧠 Declarative Memory	“Knowing That”	• Facts and rules you hold consciously • Slow, effortful retrieval	“To solve \(3x + 5 = 20\), subtract \(5\) from both sides”
⚡ Procedural Memory	“Knowing How”	• Automatic “atomic thinking steps” • Fast, effortless execution	See \(3x + 5 = 20\) → instantly know \(x = 5\)

The journey: Facts → Thousands of practice cycles → Automatic procedures → Expertise

Why struggle matters: Each practice attempt strengthens the neural pathways that create expertise

Figure courtesy of Scott H Young

The Weak-to-Strong Methods Journey

How novices become experts through progressive skill building:

Stage	Description	Methods/Characteristics	Nature
🌱 Weak Methods (Novice)	General strategies when lacking knowledge	• Means-end analysis • Working backward • Trial and error • Surface analogies • Hill climbing	Slow, effortful, but essential for learning
🔄 Proceduralization	Patterns become procedures	• Repeated sequences chunk together • Still conscious but more fluid • Reduced cognitive load • Faster with fewer errors	The critical transition phase
⚡ Strong Methods (Expert)	Domain-specific automaticity	• Pattern recognition • Forward chaining • Compiled procedures • Deep structural understanding	Fast, accurate, unconscious

Why This Matters for AI Use

AI provides expert-level answers (strong methods) to novices who haven’t developed through weak methods first. This skips the essential struggle phase where real learning occurs.

When novices use expert tools, they miss building the foundational procedures that enable true understanding.

Productions: Your Brain’s Atomic Thinking Steps

How Complex Skills Are Built:

Productions = IF-THEN Rules

• IF see \(3x + 5 = 20\) THEN subtract 5 from both sides
• IF stuck on code THEN check syntax first
• IF need to remember phone number THEN repeat internally

Each one is an “atomic thinking step”

Production Competition

When you see \(3x + 5 = 20\):

• Production A: “Guess and check” (weak)
• Production B: “Subtract 5” (strong)
• Winner: Strongest fires automatically

Practice strengthens winning productions

Why this matters:
Complex expertise = thousands of these atomic steps compiled together. Offloading provides final answers without building the atomic steps.

Figure courtesy of Scott H Young

The AI Bypass Problem

Three Ways Learning Is Disrupted:

Issue	Problem	Consequence of offloading to AI
No Prediction Errors	Brain learns from gaps between “what I expect” and “what happens”	Eliminates this entirely
No Memory Formation	Information that isn’t actively processed isn’t stored	Provides answers without processing
No Procedural Development	Expertise requires thousands of “atomic thinking steps”	Skips this building process

Result: Offloading cognitive processes can lead to surface fluency without deep understanding and complete dependency on tools.

The Research Reality

What Studies Show About AI Use in Learning:

• 68% reduction in critical thinking among high AI users
• Surface learning only in programming students using ChatGPT
  
• Flynn Effect reversal: IQ scores declining since we stopped memorizing
• Students can’t solve basic problems without AI after repeated AI-assisted attempts

We abandoned proven learning methods just as neuroscience proved why they work!

Evidence-Based Guidelines

1. Embrace the Struggle
   • Confusion is a necessary stage of learning
   • Errors provide valuable prediction error signals
   • Difficulty signals brain growth

2. Practice Retrieval
   • Test yourself before checking answers
   • Explain concepts without notes
   • Teach others what you’re learning

3. Space Your Learning
   • Return to concepts over days/weeks
   • Allow forgetting between sessions
   • Relearn in different contexts

4. Vary Your Practice
   • Mix problem types
   • Change contexts
   • Seek novel applications

5. Protect Your Prediction Errors
   • Make attempts before seeking help
   • Notice when your expectations are wrong
   • Use errors as learning opportunities, not failures

Teaching That Matches How Brains Learn

1. Respect the Developmental Sequence

• Weak methods must come first—they’re not a bug, they’re a feature
• Proceduralization requires extensive practice with deliberative processes
• Compilation happens automatically after sufficient repetition

2. Calibrate Cognitive Load

• Novices need simplified, sequential presentation
• Experts can handle complex, parallel processing
• The same material must be presented differently based on expertise level

3. Protect the Struggle Window

• Allow sufficient time for independent struggle before AI consultation
• Document thinking process before seeking external help
• Multiple attempts required before accessing solutions

4. Scaffold, Don’t Substitute

• Use AI to reduce irrelevant cognitive load, not eliminate all effort
• Provide worked examples that show the process, not just answers
• Gradually fade support as competence develops

The Key Insight:

AI is an expert-level tool. Using it before building foundational skills through weak methods isn’t just ineffective—it actively prevents the very processes that create expertise.

Design learning experiences that match how brains actually develop competence.