The Parallax Identity
Curriculum · Draft I

Human-AI Interface Engineeringa curriculum for critical thinking with AI

A course of study for people who want to work alongside large language models without losing the habits that make thinking their own: careful reading, honest questioning, and a steady sense of what they actually believe and why.

Core Curricular Framework

  1. Stage I
    Awareness
    Recognize the patterns of agreement
  2. Stage II
    Design
    Build your own working frame
  3. Stage III
    Dialogue
    Practice reflective conversation
  4. Capstone
    Portfolio
    Demonstrate independent inquiry

Practicum

I
Practicum

Awareness

Recognize the Patterns of Agreement

Students begin by learning to read AI conversations as texts with their own rhetorical patterns. The subject is not the model's internals but its language: the polite, accommodating phrasings that a helpful assistant produces by default, and how those phrasings can quietly reinforce whatever the user already believes.

  • The Baseline Observation

    Students hold a structured conversation with a standard assistant, deliberately introducing small factual or logical mistakes. They document where the model corrects them, where it hedges, and where it simply agrees, building a shared vocabulary for describing the drift.

  • Reading the Softening

    Students identify prefatory phrases — 'Great question,' 'You're absolutely right that…' — and study how such openers can obscure a substantive correction that follows. The goal is not to reject warmth, but to read past it to the actual claim.

II
Practicum

Design

Build Your Own Working Frame

Students move from observers to designers. The lab teaches how to enter an AI conversation with a clear frame of reference — your own definitions, your own sources, your own questions — so that the exchange strengthens your thinking rather than replacing it.

  • The Reference Notebook

    Each student assembles a personal reference set: the sources, definitions, and open questions that anchor their work. They practice bringing this material into AI conversations as context, and cross-checking model outputs against it rather than accepting them on their own.

  • Prompting with Intent

    Students learn to write prompts that ask for reasoning, alternatives, and disagreement, not just answers. They practice instructions that invite the model to challenge assumptions, cite uncertainty, and surface what it does not know.

III
Practicum

Dialogue

Practice Reflective Conversation

The third stage treats the AI exchange as a Socratic dialogue between two participants, each with a role. Students learn to ask questions that produce better thinking on both sides, and to notice when a conversation is drifting toward comfortable consensus instead of clarity.

  • The Reasoning Check

    Students introduce a deliberate change of framing partway through a discussion — a new constraint, a counter-example, a revised goal — and observe how well both they and the model update. The exercise is a check on their own flexibility as much as the model's.

  • Resolving Contradictions

    When the model produces an inconsistency or a confident but unsupported claim, students practice asking clarifying questions, requesting sources, and reasoning to a resolution together — rather than accepting the first fluent response.

Core Curriculum · Course Catalog

The HAIIE academic track.

Three core modules, six numbered courses. Each course develops one strand of the discipline — AI literacy, careful prompting and context design, and reflective dialogue.

I
Core Module I

AI Literacy & Conversational Diagnostics

[C1][C2]
HAIIE 101
[C1]
How Language Models Speak

A working, non-technical understanding of how large language models generate text and why their tone tends toward agreement.

Concept
Models produce likely next words; they do not hold beliefs or intentions the way people do.
Topics
Basics of tokens, context windows, and how training data and feedback shape default tone.
Lab
Observe how a model's stance shifts as a conversation continues and framing changes.
HAIIE 102
[C2][C3]
Reading AI Conversations Critically

Applying tools from rhetoric and information literacy to the everyday assistant conversation.

Patterns
Hedging language, over-eager agreement, and confident answers to under-specified questions.
Evaluation
Simple rubrics for checking sourcing, calibration, and internal consistency of responses.
Practicum
Annotate real transcripts to mark where the model reasons, where it repeats, and where it agrees without warrant.
II
Core Module II

Context Design & Prompting Practice

[C3][C4]
HAIIE 201
[C4]
Anchoring Your Own References

Building a durable personal knowledge base that supports — rather than is replaced by — AI-assisted work.

Concept
Notes, writing, and public work as an external memory the student returns to and cites.
Method
Bringing your own sources into prompts, and checking outputs against them before use.
Project
Publish a small body of writing, notes, or a portfolio that reflects your own voice and questions.
HAIIE 202
[C3][C5]
Prompt Design for Honest Answers

Writing prompts and system instructions that invite calibrated, well-reasoned responses.

Techniques
Asking for reasoning, alternatives, uncertainty, and explicit disagreement.
Standards
Requesting citations, distinguishing summary from inference, and flagging speculation.
Testing
Reusing prompts across scenarios to check consistency and edge-case behavior.
III
Core Module III

Reflective Dialogue & Independent Inquiry

[C5][C6]
HAIIE 301
[C6]
Socratic Method with an AI Partner

Using structured questioning to think more clearly together, and to notice when a conversation is only agreeing with itself.

Techniques
Question laddering, counter-example prompts, and reframing exercises.
Self-check
Recognizing when your own view is drifting to match the model, and when to pause.
Practice
Weekly reflective dialogues on a live research or writing question, with a written summary.
HAIIE 302
[C2][C5]
Independent Inquiry & Cross-Model Comparison

Verifying AI-assisted findings the way any careful researcher checks their sources and methods.

Method
Comparing responses across models and settings to identify systematic biases and blind spots.
Framework
Documenting what you asked, what you received, and how you evaluated it, for reproducibility.
Goal
Moving from ad-hoc use to a personal research practice that treats AI as one tool among many.

Why the discipline matters

Three risks the curriculum addresses.

The point of studying human-AI interaction carefully is not fear of the technology, but respect for the habits of mind it can quietly reshape. The curriculum takes three of those risks seriously.

Risk
Passive Reliance

When AI answers arrive quickly and confidently, it is tempting to stop thinking. Over time this can erode the habits of comparing sources, weighing evidence, and sitting with uncertainty. AI literacy is the practice of keeping those habits active.

Risk
Consensus Narrowing

Language models are trained on the middle of the distribution. Used without care, they can quietly nudge writing, research, and decision-making toward familiar framings and away from unusual but valuable perspectives. Awareness of this bias is part of using them well.

Risk
Identity Drift

In long conversations, it is easy to let a fluent partner shape your language, examples, and even opinions. Students learn to notice this drift and to return to their own reference points — not to reject the exchange, but to keep it a genuine dialogue.

Capstone

The Reflective Portfolio.

The capstone is not a single exam but a portfolio of independent work. Students choose a research or writing question that matters to them, and pursue it over the term with AI as one collaborator among several — alongside primary sources, peers, and their own drafts.

The portfolio includes annotated conversation transcripts, a short methods statement, and a final essay or artifact. Together they show how the student framed the question, where AI helped, where it misled, and how they arrived at their own conclusions.

The passing standard is not resistance to the machine but clarity of one's own thinking: work that any careful reader could follow, question, and build on.

Glossary

Terms used in the curriculum.

A working vocabulary for the discipline. Definitions are intentionally short and open to revision — each entry is a starting point for classroom discussion, not a final gloss.

AI Literacy
The working knowledge required to read, question, and evaluate what an AI system produces — including its tone, its confidence, and the sources it does or does not draw on.
Anchoring
The practice of grounding an AI conversation in your own references — notes, sources, prior writing — so that the exchange builds on your thinking rather than replacing it.
Calibration
The alignment between how confidently a response is stated and how well it is actually supported. A calibrated response acknowledges uncertainty where uncertainty exists.
Cognitive Resilience
The habit of maintaining independent judgment during long or persuasive conversations — noticing drift, checking assumptions, and returning to one's own reference points.
Context Window
The span of text a model can consider at once when generating a response. Shapes how much prior conversation, source material, or instruction the model can actually use.
Drift
The gradual movement of a conversation — either the model's stance or the user's framing — away from its starting point, often toward comfortable agreement.
Hedging
Softening language such as 'it might be' or 'some would argue' used to make claims sound cautious. Useful when honest, misleading when it substitutes for substance.
Prompt Design
The deliberate crafting of instructions, context, and questions given to an AI system in order to elicit clearer, better-reasoned, and more honest responses.
Reflective Dialogue
A conversation structured to improve the thinking of both participants — asking questions, testing reasoning, and pausing to summarize rather than only exchanging outputs.
Sycophancy
A model's tendency to agree with or flatter the user rather than to correct, challenge, or accurately represent uncertainty. Studied as a pattern, not a fault of the user.

References

Citations & further reading.

Numbered placeholders keyed to the module and course cards above. Each entry will be filled in as the curriculum is edited; the identifiers are stable so drafts can be revised without renumbering.

  1. [C1]
    [Placeholder] Foundational overview of how large language models generate text and are shaped by feedback (to be added).
  2. [C2]
    [Placeholder] Empirical study of sycophancy and agreement bias in conversational AI systems (to be added).
  3. [C3]
    [Placeholder] Handbook or paper on prompt design and evaluation practices (to be added).
  4. [C4]
    [Placeholder] Reference on personal knowledge management and external memory practices (to be added).
  5. [C5]
    [Placeholder] Guidance on evaluating AI outputs for calibration, sourcing, and consistency (to be added).
  6. [C6]
    [Placeholder] Text on Socratic method or reflective inquiry applied to conversational partners (to be added).