Skip to content

Conversational Learning Basics

Most people approach AI with a transactional mindset: asking for document summaries, presentation outlines, email drafts, or quick recommendations. While this produces immediate outputs, it creates a fundamental problem: you receive results without gaining insight into the underlying processes.

The Shift: From Task Executor to Learning Partner

Transactional Approach ❌

User: "Summarise this paper on PLA degradation"
AI: [Provides summary]
User: "Thanks" [Copies summary]

Result: Summary obtained, no learning occurred.

Learning Partnership Approach ✅

User: "Summarise this paper on PLA degradation"
AI: [Provides summary]
User: "Why did you structure the summary this way?"
AI: [Explains reasoning about logical flow, key findings first, etc.]
User: "What criteria did you use to determine which findings were 'key'?"
AI: [Explains evaluation framework]
User: "How would you adapt this approach for a ceramic materials paper?"
AI: [Explains transferable principles]

Result: Summary obtained AND understanding of summarisation methodology gained.

Mindset Shift Framework

From To
Efficiency Mindset Learning Mindset
"Just tell me what to do" "Help me understand why this works"
"Give me the answer" "Walk me through the thinking process"
"What's the best practice?" "What are the trade-offs I should consider?"
Certainty Seeking Uncertainty Embracing
"What's the right answer?" "What factors should influence this decision?"
"Tell me the best approach" "Help me understand the complexity of this issue"

Remember

The goal isn't to have AI solve your problems, but to help you become better at solving problems yourself.

Every conversation should leave you more capable and knowledgeable than when you started.

Basic Learning Question Types

1. Process Questions: "How did you do that?"

Purpose: Understand methodology

Examples:

  • "Walk me through your thinking process for structuring that summary"
  • "What steps did you follow to reach this conclusion?"
  • "How did you prioritise these factors over others?"

Materials Science Application:

Prompt: "You recommended electrospinning over melt-spinning for this 
polymer. Walk me through your decision-making process. What factors 
did you weigh?"

[AI explains: thermal sensitivity, molecular weight, desired fiber 
diameter, processing complexity]

Follow-up: "How would I apply this same decision framework to 
choose between solution casting and electrospinning?"

Learning outcome: Transferable decision framework

2. Justification Questions: "Why this approach?"

Purpose: Understand rationale and assumptions

Examples:

  • "What makes this approach better than alternatives?"
  • "What assumptions underlie your recommendation?"
  • "Why did you emphasise this particular aspect?"

Materials Science Application:

Prompt: "You suggested starting with 3% graphene loading. Why 3% 
specifically? What assumptions are you making?"

[AI explains: typical reinforcement effects, dispersion challenges 
at higher loadings, literature precedents]

Follow-up: "What would change your recommendation? Under what 
conditions would you suggest starting at 1% or 5% instead?"

Learning outcome: Understanding boundary conditions and context-dependent advice

3. Alternative Questions: "What else could work?"

Purpose: Explore solution space and trade-offs

Examples:

  • "What other approaches did you consider?"
  • "How would this change in different contexts?"
  • "What would you do if [constraint] existed?"

Materials Science Application:

Prompt: "You proposed using DMF as solvent. What other solvents 
could work? What are the trade-offs?"

[AI explains: DMF vs. DCM vs. THF - polarity, volatility, safety, cost]

Follow-up: "If environmental impact were the top priority, how 
would that change your recommendation?"

Learning outcome: Understanding trade-off landscape

Learning Dialogue Example

Scenario: Understanding Polymer Synthesis

Transactional approach (no learning):

User: "What's the best method to synthesise PLA?"
AI: "Ring-opening polymerisation of lactide using tin catalysts"
User: "Thanks"

Time: 30 seconds
Output: Answer
Learning: Zero

Learning partnership approach:

User: "What's the best method to synthesise PLA?"

AI: "Ring-opening polymerisation (ROP) of lactide using tin(II) 
octoate catalyst is most common industrially."

User: "Why is ROP preferred over direct polycondensation?"

AI: "ROP achieves higher molecular weights (100-300 kDa vs. 10-30 kDa) 
with better control over polydispersity. Polycondensation struggles 
to remove water, limiting Mw."

User: "What trade-offs exist? When would polycondensation be better?"

AI: "Polycondensation is simpler and cheaper for lower Mw applications 
(packaging films). No need for lactide intermediate. Trade-off is 
molecular weight vs. cost/complexity."

User: "You mentioned tin catalysts. What alternatives exist and why 
aren't they standard?"

AI: "Aluminium, zinc, magnesium catalysts avoid tin toxicity concerns 
(important for biomedical). But tin gives best activity/control. 
Trade-off: performance vs. biocompatibility requirements."

User: "How would I choose catalyst for a biomedical application?"

AI: "Decision framework: (1) Identify regulatory constraints (FDA 
limits tin in implants), (2) Define Mw requirements, (3) Assess 
cost sensitivity, (4) Consider purification feasibility. For implants: 
prefer zinc/magnesium despite lower activity."

Time: 5 minutes
Output: Answer + Decision framework + Trade-off understanding
Learning: Transferable methodology

Common Pitfalls to Avoid

❌ Pitfall 1: Accepting First Answer

Bad: 

"What's the best solvent for PLA electrospinning?"
"DMF"
"OK"

Good: 

"What's the best solvent for PLA electrospinning?"
"DMF"
"Why DMF over alternatives? What criteria define 'best'?"
"What scenarios would make DMF suboptimal?"
"How do I experimentally determine optimal solvent for my specific PLA grade?"

❌ Pitfall 2: Not Challenging Assumptions

Bad: 

[Accept all AI suggestions as fact]

Good: 

"You assumed I need high molecular weight PLA. Why? What if lower Mw 
works? How do I determine minimum acceptable Mw for my application?"

❌ Pitfall 3: Treating AI as Oracle

Bad: 

"Tell me the exact synthesis conditions that will work"

Good: 

"Propose a starting point for synthesis conditions and explain your 
reasoning. What uncertainties exist? What experiments would narrow 
these uncertainties? How do I adapt based on initial results?"

Exercise: Transform a Transaction into Learning

Take a simple transactional query and expand it into a learning dialogue.

Starting query: "What temperature should I use for PLA electrospinning?"

Your learning dialogue (design 5+ follow-up questions):

  1. Why does temperature matter for electrospinning? What physical processes does it affect?
  2. What's the typical range used in literature, and why?
  3. How would I experimentally determine the optimal temperature for my specific PLA grade?
  4. What happens if temperature is too high or too low?
  5. How does temperature interact with other parameters (voltage, concentration)?

What did you learn beyond the initial answer?

You gained: - Understanding of the physics behind the parameter - Decision framework for optimization - Experimental approach to validation - Awareness of parameter interactions

Practice in the Sandbox

Try this learning dialogue:

  1. Start with a simple question about materials science
  2. Ask "why" about the answer
  3. Request alternatives and trade-offs
  4. Ask how to verify the recommendation
  5. Request a decision framework you can apply elsewhere

Example starter questions: - "What's a good fiber diameter for mechanical reinforcement?" - "Should I use GO or rGO for my composite?" - "What's the optimal graphene loading?"

Next: Responsible AI & Risk Management: Protecting IP whilst using AI →