Week 9: Socratic Dialogue with AI

Welcome to Week 9 of the AI Literacy Course.

In Lesson 8, we explored how AI can support learning through explanations, practice, hints, feedback, and reflection.

This week, we focus on one particular method: Socratic dialogue.

Socratic dialogue uses connected questions to examine meaning, evidence, assumptions, alternatives, perspectives, and consequences. The purpose is not to make AI sound wise or to let it win an argument.

The purpose is to make reasoning visible.

AI may generate the questions, but the learner remains responsible for the answers, source checking, conclusions, and final reflection.

Learning goals

By the end of this lesson, learners should be able to:

• Distinguish Socratic dialogue from a quiz, lecture, or debate

• Write a starting statement that includes current thinking and uncertainty

• Use a one-question wait rule

• Recognise six types of Socratic question

• Identify leading questions and hidden premises

• Treat AI-generated perspectives as hypotheses rather than testimony

• Pause the dialogue when evidence requires external verification

• Reflect on how the dialogue affected their reasoning

1. What makes a dialogue Socratic?

A quiz usually asks questions with expected answers. Its purpose is often to test what someone remembers.

Example:

“What is the definition of democracy?”

A Socratic dialogue examines the learner’s reasoning.

Example:

“You said democracy requires participation. What kind of participation do you consider necessary, and why?”

The second question responds to the learner’s idea. It asks the learner to clarify and justify their thinking.

Socratic dialogue is also different from a lecture. AI should not ask a question and then immediately provide a long explanation before the learner responds.

It is different from a debate because the goal is not victory. A successful dialogue may reveal uncertainty, improve an argument, narrow a claim, or show that more evidence is needed.

The central question is not:

“Who won?”

It is:

“What did the questioning reveal?”

2. Begin with a starting statement

A Socratic dialogue needs something to examine.

The learner may begin with:

• a claim;

• an interpretation;

• a judgement;

• a proposed solution;

• a prediction;

• a question;

• a point of confusion;

• a tentative answer.

A useful starting statement includes:

1. what the learner currently thinks;

2. one or two reasons;

3. something the learner is uncertain about.

Example:

“I am unsure whether social media improves democracy. My first thought is that it increases participation because more people can express opinions publicly. However, I am uncertain whether expressing an opinion leads to meaningful political influence.”

This gives the dialogue a clear starting point without forcing the learner to defend a fixed position.

If AI generates both the starting statement and the critique, the learner may become a spectator. The learner should contribute the initial thinking.

3. Use one question and a wait rule

A Socratic dialogue works best when AI asks one concise question and waits for the learner’s answer.

Example prompt:

“Use the behaviour of a respectful Socratic dialogue partner. Ask one concise question at a time about my starting statement. Wait for my answer before continuing. Build each new question from what I actually say. Do not automatically agree or disagree, and do not give a lecture. If a question contains a factual claim, identify it. If I say ‘I do not know,’ help me decide what information or smaller question I need.”

The wait rule keeps the learner active.

Without it, AI may:

• ask many questions at once;

• answer its own questions;

• move away from the learner’s reasoning;

• overwhelm the learner;

• turn the dialogue into a prepared script.

A good next question responds to the previous answer. It should not simply move mechanically through a checklist.

4. Six types of Socratic question

Different questions examine different parts of reasoning.

Question type	What it examines	Example
Clarification	Meaning, definitions, and scope	“What do you mean by ‘improves’?”
Reasons and evidence	Support for the claim	“What evidence supports that conclusion?”
Assumptions	What is taken for granted	“Does this argument assume equal access?”
Alternatives and counterexamples	Other explanations, exceptions, or cases	“Can you think of a case where this would not apply?”
Perspectives	How the issue may be framed differently	“What questions might an accessibility specialist raise?”
Consequences	What may follow if the claim is accepted	“Who might benefit, and who might face a disadvantage?”

Clarification

Clarification questions examine what words mean and where a claim applies.

Examples:

• What do you mean by “fair”?

• Which cases are included?

• Does the claim apply everywhere or only under certain conditions?

• What would count as an example?

Clarification should focus on ambiguities that affect the reasoning. Endless requests for definitions can prevent the dialogue from reaching the central issue.

Reasons and evidence

These questions examine why the learner believes something.

Examples:

• What reason supports your conclusion?

• Is this based on personal experience, data, expert judgement, or another form of evidence?

• Which part of the source supports the claim?

• What evidence would make you reconsider your view?

AI may introduce unsupported claims or invented sources. Important evidence must be checked outside the chatbot.

Assumptions

Assumption questions reveal what the reasoning takes for granted.

Examples:

• What must be true for this argument to work?

• Are you assuming that everyone has equal access?

• Which value is being treated as most important?

• Is this assumption yours, or was it introduced by AI?

An assumption is not automatically wrong. But it should be visible and open to examination.

Alternatives and counterexamples

These questions test whether another explanation or exception exists.

Examples:

• What is another possible explanation?

• Can you think of a counterexample?

• Under what conditions would the claim not apply?

• What evidence could help distinguish between two explanations?

An alternative should be relevant and supported. It should not be invented merely to make every issue appear to have two equally strong sides.

Perspectives

Perspective questions explore how an issue may be understood from another position, discipline, or institutional role.

Examples:

• What questions might a learner raise?

• How might an employer frame the problem?

• What might a public-health researcher examine?

• What concerns might an accessibility specialist investigate?

AI-generated perspectives are possible lines of inquiry. They are not testimony from real people.

Consequences

Consequence questions examine what may follow if a claim, proposal, or policy is accepted.

Examples:

• Who may benefit?

• Who may be disadvantaged?

• What safeguards would be needed?

• What might happen if this becomes standard practice?

• What unintended effect could occur?

Separate consequences supported by evidence from consequences that are merely imagined.

5. Follow the reasoning, not a script

A strong dialogue develops from the learner’s previous answer.

Example:

Learner:
“I think social media improves democracy because more people can participate.”

AI — clarification:
“What do you mean by participation?”

Learner:
“I mean that people can express their political opinions publicly.”

AI — assumption:
“Does your claim assume that expressing an opinion leads to meaningful political influence?”

Learner:
“Yes. I may be treating expression and influence as the same thing.”

AI — reasons and evidence:
“What evidence would help you determine whether online expression influences political decisions?”

Each question responds to something the learner said. The dialogue reveals that “participation” may include different activities and that public expression does not automatically produce political influence.

A good dialogue does not need to use all six question types in a fixed order. It should follow the reasoning and focus on the questions that matter.

6. Examine the questions themselves

AI-generated questions also require critical examination.

A question may:

• contain an unsupported claim;

• assume that something is already true;

• push one answer;

• use emotionally loaded language;

• present one perspective as normal;

• introduce a stereotype;

• frame a complex issue too narrowly.

Leading questions

A leading question pushes the learner toward an expected answer.

Leading:

“Why is surveillance harmful?”

This assumes that surveillance is harmful and asks only for reasons supporting that position.

More open:

“What benefits, harms, conditions, and evidence should be considered when evaluating surveillance?”

The second question allows a broader examination.

Hidden premises

A question may contain an unverified premise.

Example:

“Why has remote learning reduced educational quality?”

This assumes that remote learning has reduced quality before that claim has been established.

A more careful question is:

“What evidence is available about how remote learning has affected educational quality in different contexts?”

Before answering an AI question, ask:

• What does this question assume?

• Has that assumption been established?

• Does it contain a factual claim?

• Is the wording pushing me toward one conclusion?

• Should the question be rewritten?

All questions frame an issue to some degree. The aim is not perfect neutrality, but visible and fair framing.

7. Use challenges and perspectives responsibly

AI may challenge the learner, but it may also agree too readily.

Agreement from AI does not prove that the learner’s reasoning is sound. Praise may reflect the system’s conversational style rather than the quality of the argument.

A useful instruction is:

“Do not automatically agree or disagree. Ask questions that help me examine my reasons.”

Avoid false balance

Asking for another view does not mean that all views have equal evidence.

For example, a well-supported scientific conclusion should not be treated as equal to an unsupported claim merely because the dialogue requests “both sides.”

A useful challenge asks for:

• a relevant limitation;

• a credible alternative explanation;

• a condition under which the claim may change;

• evidence that would weaken or strengthen the conclusion.

Generated perspectives are not real voices

AI can suggest questions that different groups might raise, but it cannot speak on behalf of those groups.

Reasonable:

“What questions might an accessibility advocate raise about this proposal?”

Misleading:

“Speak for disabled people and tell me what they think.”

If an issue concerns people’s experiences, consult evidence and real voices from the people affected.

8. Pause, verify, or stop when needed

A Socratic dialogue can become more polished while continuing from a false premise.

Pause when a central claim requires verification.

For example, if AI introduces:

• a statistic;

• a quotation;

• a historical event;

• a legal rule;

• a scientific claim;

• a named source;

check it outside the chatbot before building further reasoning on it.

After checking, you can:

• correct the premise;

• add verified information;

• revise the question;

• continue the dialogue;

• suspend judgement if evidence is insufficient.

Detailed source checking using SIFT will be introduced in Lesson 12.

“I do not know” is a useful answer

Saying “I do not know” may reveal that:

• a term is unclear;

• evidence is missing;

• the question is too broad;

• an assumption has not been considered;

• a smaller question or example is needed.

AI should not immediately answer on the learner’s behalf.

It can ask:

“What information would help you answer?”

or:

“Would a definition, example, or smaller question help?”

Sensitive and high-stakes topics

A productive dialogue challenges the reasoning, not the person.

The learner should be able to:

• say “I do not know”;

• ask why a question matters;

• request simpler wording;

• pause the discussion;

• refuse an unsuitable question;

• end the interaction.

Sensitive personal, medical, legal, crisis, or safeguarding issues may require a trusted person or qualified professional rather than extended chatbot questioning.

9. Conclude with learner reflection

A successful dialogue does not need to reverse the learner’s opinion.

The result may be:

• a changed conclusion;

• a narrower claim;

• a more qualified position;

• stronger reasons for the original view;

• a newly visible assumption;

• recognition that evidence is missing;

• a decision to suspend judgement.

At the end, the learner should record:

• the starting statement;

• the most useful question;

• an assumption that became visible;

• any claim that required checking;

• what changed;

• what did not change;

• what remains uncertain;

• the final or revised position.

The learner should be able to explain the conclusion without relying on the chatbot transcript.

Agreement with AI is not evidence of learning. Clearer, better-supported reasoning is.

Classroom activity: Socratic dialogue analysis

Choose a course-related claim, interpretation, proposed solution, prediction, or open question.

Step 1: Write a starting statement

Include:

• what you currently think;

• one or two reasons;

• one uncertainty.

Step 2: Write the dialogue prompt

Use a prompt such as:

“Use the behaviour of a respectful Socratic dialogue partner. Ask one concise question at a time about my starting statement and wait for my answer. Build each new question from what I say. Use clarification, reasons and evidence, assumptions, alternatives and counterexamples, perspectives, and consequences when relevant. Do not automatically agree or disagree. Identify factual claims or hidden premises that may require checking.”

Step 3: Complete the dialogue

Complete approximately six to eight meaningful turns.

Quality matters more than the exact number of questions.

Step 4: Record the dialogue

Turn	Question type	AI question	My answer	What did this reveal?
1
2
3
4
5
6

Step 5: Examine the questioning

Identify at least one:

• leading question;

• hidden premise;

• AI-introduced assumption;

• question that needed rewriting.

If no problematic question appeared, write one possible example and improve it.

Step 6: Verify when needed

If the dialogue contains an important factual claim, verify it through an appropriate external source.

Record what you checked and whether the claim was supported.

Step 7: Write a closing position

Explain what you now think. You may retain, change, narrow, or suspend your original position.

Step 8: Reflect

Write approximately 300–500 words explaining:

• your starting statement;

• the most useful question;

• one assumption or ambiguity you discovered;

• any evidence you checked;

• how your reasoning changed or became more qualified;

• what remains uncertain;

• why the final conclusion is your own.

Reflection questions

1. What distinguishes Socratic dialogue from a quiz?

2. What should a starting statement contain?

3. Why should AI ask one question at a time?

4. Which question clarified an important term?

5. Which question examined reasons or evidence?

6. What assumption became visible?

7. Did a question contain a hidden premise or leading frame?

8. Did AI agree too readily or introduce an unsupported perspective?

9. What required external verification?

10. Can you explain your final position independently?

Key vocabulary

Socratic dialogue:
A disciplined inquiry process that uses connected questions to examine reasoning.

Starting statement:
The learner’s current claim, interpretation, question, or tentative answer, together with reasons and uncertainty.

Clarification question:
A question about meaning, definitions, boundaries, or scope.

Reasons-and-evidence question:
A question about what supports a claim or conclusion.

Assumption question:
A question about what the reasoning takes for granted.

Alternative or counterexample question:
A question that examines another explanation, exception, or case where a claim may not apply.

Perspective question:
A question about how an issue may be framed differently by people, groups, disciplines, or institutions.

Consequence question:
A question about what may follow if a claim or proposal is accepted.

Wait rule:
A dialogue rule requiring AI to ask one question and wait for the learner’s answer before continuing.

Leading question:
A question that pushes an expected answer or presents one position as already correct.

Hidden premise:
An assumption or factual claim contained inside a question but not yet established.

Summary

In Week 9, we learned that Socratic dialogue is different from a quiz, lecture, or debate.

It begins with the learner’s starting statement and uses connected questions to examine:

• clarification;

• reasons and evidence;

• assumptions;

• alternatives and counterexamples;

• perspectives;

• consequences.

The one-question wait rule keeps the learner active. Questions themselves must also be examined for leading language, hidden premises, unsupported claims, and automatic agreement.

A successful dialogue does not require the learner to change sides.

The goal is reasoning that becomes clearer, better supported, more carefully qualified, and genuinely owned by the learner.

Lesson 9 Interactive Quiz: Socratic Dialogue with AI

Choose one answer for each question. Then select Check my answers. You will receive feedback for every question. This practice quiz does not collect names or scores.

1. Which question is the best example of Socratic dialogue rather than a recall quiz?

A. “In which year was the law introduced?” B. “What is the textbook definition of democracy?” C. “You said democracy requires participation. What kind of participation do you consider necessary, and why?” D. “Which of these four terms is correct?”

2. Which is the strongest starting statement for a Socratic dialogue?

A. “I think social media may increase political participation because more people can speak publicly, but I am unsure whether expression leads to real influence.” B. “Give me the correct opinion about social media and democracy.” C. “Choose a position for me and then challenge it.” D. “Social media is either completely good or completely bad for democracy.”

3. Why should AI ask one Socratic question and then wait?

A. So the dialogue always uses every question type in a fixed order. B. So AI can provide the learner’s answer after a short pause. C. So the conversation produces as many questions as possible. D. So the learner can answer and the next question can build on that reasoning.

4. What type of Socratic question is: “What do you mean by ‘fair’ in this proposal?”

A. Consequence B. Clarification C. Perspective D. Alternative or counterexample

5. What type of Socratic question is: “Does your argument assume that every learner has equal internet access?”

A. Reasons and evidence B. Clarification C. Assumption D. Consequence

6. Which question avoids the hidden premise in “Why has remote learning reduced educational quality?”

A. “What evidence shows how remote learning has affected educational quality in different contexts?” B. “Why do schools continue using remote learning despite its lower quality?” C. “Which part of remote learning caused the greatest decline in quality?” D. “How should teachers repair the damage caused by remote learning?”

7. A learner answers, “I do not know.” What is the best Socratic response?

A. “Then I will give you the complete conclusion.” B. “You should choose the position that sounds most convincing.” C. “Let us ignore the uncertainty and continue to the next topic.” D. “What information or smaller question would help you examine this?”

8. AI describes what an accessibility advocate might say about a proposal. How should the learner use this output?

A. Treat it as testimony from accessibility advocates. B. Treat it as a possible line of inquiry and consult real evidence or voices. C. Assume it represents the shared view of all disabled people. D. Use it as evidence without identifying that it was generated.

9. During the dialogue, AI introduces a statistic that becomes central to the argument. What should happen next?

A. Continue because a precise statistic is likely to be reliable. B. Ask AI to repeat the statistic with greater confidence. C. Pause and verify the statistic through an appropriate external source. D. Accept it if it supports the learner’s original position.

10. A learner keeps the original conclusion but narrows the claim, identifies an assumption, and explains remaining uncertainty. Was the dialogue successful?

A. Yes. The reasoning became clearer and more carefully qualified. B. No. A successful dialogue must reverse the learner’s opinion. C. No. AI should decide which conclusion the learner adopts. D. Only if the learner agrees with every AI question.

Reflection: Write one leading question about a course topic. Then rewrite it as a more open Socratic question that makes its assumptions visible.