The
Reality
Detective

Book One

What AI knows. What it doesn’t. And how to tell the difference.

Does AI feel like a mystery? There’s a reason for that, and this book opens the case file.

Sign up for pre-release See the AI stress tests
The Reality Detective, Book One, by Kevin Schumacher. Hardcover with a navy jacket and a teal diamond lattice.

The case starts here.

Everyone is using AI. Almost no one was taught what it is.

For most people, AI is a box you type a question into. An answer comes back sounding sure of itself, like a very well-spoken Magic 8 Ball. How it got that answer, and whether to trust it, remains a mystery.

That leaves the people expected to explain AI with nowhere to start. Teachers and homeschooling parents are fielding questions they were never prepared to answer.

The Reality Detective is that starting point. It assumes no technical background and uses philosophy’s oldest thinking tools to show readers how to check an AI’s answer instead of simply accepting it. It’s written for readers from 10th grade up, and for the adults teaching them.

Readers who want to go further will find advanced tools in the back, including syllogisms and a full bibliography. A companion workbook gives students hands-on practice.

A rigorous, clear-eyed field guide that successfully forces the reader to stop treating AI as an oracle and start treating it as a tool.

Arya 3, one of five AI models that stress-tested the manuscript. All five passed it. See the stress tests

Tested against the frontier

Five leading AI models stress-tested the manuscript. All five passed it, with a composite score of 8.7 out of 10.

Independent AI stress-test scorecard
AI model testedScoreVerdictWhat the model concluded
Grok 4.5 9.0/10 Pass Exceptionally robust, intellectually honest, and structurally tighter than most books that attempt this subject.
Arya 3 9.0/10 Pass A rigorous, clear-eyed field guide that successfully forces the reader to stop treating AI as an oracle and start treating it as a tool.
Gemini (Flash) 8.8/10 Pass Architecturally robust and methodologically recursive; an essential pedagogical antidote to automated epistemic complacency.
ChatGPT 5.6 Luna 8.5/10 Pass Very strong, core argument survives intact. Takes difficult epistemological questions and gives readers a practical method to investigate them.
Claude Opus 5.5 8.0/109.0 projected Pass The teaching design is where this book stands out... The argument is sound, and the Appendix provides an honest map of its logic.
Composite8.7/105 of 5 pass
Inspect the AI stress tests How each model graded the book’s key ideas, and where all five agreed

How the models graded key concepts and architecture

StrongVery StrongExceptional

Evaluator ratings by dimension
Evaluation dimensionGrok 4.5Arya 3Gemini (Flash)ChatGPT 5.6 LunaClaude Opus 5.5Consensus takeaway
Core realism & ontologyStrongStrongVery StrongStrongStrongClassical correspondence and realism survive all scrutiny.
Recursive “Box Test” logicVery StrongVery StrongExceptionalVery StrongVery StrongRelativism and consensus-as-truth self-destruct cleanly.
Pedagogical devicesExceptionalVery StrongExceptionalStrongVery StrongTruth Tree, Glorp, and Windowless Library excel.
Metacognition / hallucinationStrongStrongStrongStrongStrongArchitectural lack of self-monitoring in AI is proven.
No-human-loop accountabilityVery StrongVery StrongVery StrongVery StrongVery StrongReal-time safety vacuum without human oversight is airtight.
Verification gap math (2n)Very StrongStrongExceptionalStrongStrongCombinatorial risk separates generative AI from calculators.
Appendix formalizationStrongVery StrongVery StrongStrongVery StrongIndependent vs. P7 tagging lauded as intellectually honest.
Target audience (10th grade+)AppropriateHighly EffectiveStrongAppropriateStrongAccessible, respectful of reader intelligence, zero pandering.

Download the case file

Educators and reviewers: get every score, rating, and quote from all three tables in one spreadsheet-ready file.

Download CSV (4 KB)

Almost everyone who's used an AI becomes aware of a small, weird feeling inside them at some point. There's a reason for that.

Trust

Your own experience and understanding for deciding who or what deserves your trust stops working. Those experience signals don't exist in an AI.

Agency

Accepting an answer you can't actually evaluate isn't a choice. It's a guess with extra steps.

Moral accountability

When something goes wrong, there isn't anyone that can be held responsible.

What's inside

01The First Case: What Does "Real" Mean?
02Physical Evidence: The World We Can Touch
03Abstract Evidence: The World of Ideas
04Connecting the Evidence: How Reality Fits Together
05The Detective's Methods: How We Know Anything
06Building Truth Statements
07The Automated Assistant: What AI Actually Is
08The Assistant's Blind Spots
09The Consensus Problem: When Popular Isn't True
10Thinking About Thinking: Metacognition
11The Discernment Filter
12Artificial Confabulation: Understanding AI Hallucinations
13The Uncanny Valley of Thought
Kevin Schumacher, author of The Reality Detective

Kevin Schumacher

Author

Kevin Schumacher is a product design leader, front-end developer, and artist. A veteran of enterprise experience architecture and co-founder of the UX/AI Cooperative, he writes and illustrates works exploring human agency, design, and epistemology in the age of artificial intelligence.