Project Documentation · Master Reference

Irreducibly Human: What AI Can and Can't Do

Complete project reference — series architecture, course inventory, production pipeline, and open action items. Internal working document for orientation and handoff.

Version 1.0  |  March 2026  |  Reviewed by Dev the Dev
01

Thesis and framing

Core argument

Machines are superhuman at pattern recognition, fact retrieval, and syntactic correctness. Everything else is irreducibly human. Here is the curriculum for everything else.

The series addresses an educational failure that predates AI and is now impossible to ignore: curricula that optimized for fact retrieval, arithmetic speed, and syntactic correctness — the exact capacities where machines are now superhuman. The arrival of AI did not create this failure. It exposed it.

The curriculum was built for a world in which arithmetic speed and fact retrieval were genuinely valuable human capacities. That world is gone. The mistake was not malicious. Institutional inertia is not stupidity — it is the normal lag between a changed environment and a changed response. Schools change slowly because they were built to transmit what is known, not to respond to what is new. That feature is now a bug.

The forklift argument: The intelligent response to a forklift is not to practice lifting heavier objects. It is to learn to operate the machine, understand what it can and cannot lift, and — most crucially — develop the judgment to know what needs lifting in the first place. The series does not teach students to compete with AI. It teaches them to supply the reasoning AI tools require humans to provide — reasoning no tool can supply on their behalf.

The one-sentence version

A curriculum series, production pipeline, and measurement infrastructure for the cognitive capacities the AI era most urgently requires humans to develop — demonstrated by the method used to build it.

The series is called Irreducibly Human: What AI Can and Can't Do. Previously titled "The Human Half: What AI Can't Do." The rename reflects a sharpening of the thesis: the series is not only about limitation, it is about the specific human capacities that AI augments rather than replaces — and how to develop them deliberately.

02

The seven-tier taxonomy

The taxonomy organizes human intelligence into seven tiers by what machines can and cannot do. This is not an academic classification — it is a triage. Where machines are strongest, training humans to compete directly is malpractice. Where machines are weakest (Tiers 4, 5, 7), education needs to rebuild from scratch.

The taxonomy extends Howard Gardner's multiple intelligences framework, which was built before machines became capable and did not need to ask which intelligences technology endangered. Gardner's framework is a framework of individuals. These tiers add the supervisory layer (Tier 4), the causal layer (Tier 5), and the collective layer (Tier 6) — none of which any individual-focused framework can account for.

Tier 1

Pattern & Association

Linguistic intelligence, logical-mathematical reasoning, musical pattern recognition, spatial reasoning, naturalistic categorization, encyclopedic recall and associative lookup.

AI capability: Superhuman. Teach humans to use these tools, not replicate them.
Course: AI Literacy, Fluency, and Trust (Book 1)
Tier 2

Embodied & Sensorimotor

Bodily-kinesthetic intelligence (proprioception, physical skill, tool use), spatial-causal prediction of physical consequences, ecological intelligence. The knowledge that lives in the hands, the auditory and tactile feedback through a tool.

AI capability: Weak / emerging. Humans hold the general case. Robotics improving in narrow domains.
Not a course — addressed via Teacher's Guide companion (Companion 1). Nobody stopped teaching woodshop because a language model can describe joinery.
Tier 3

Social & Personal

Interpersonal intelligence (reading others, social navigation), intrapersonal intelligence, emotional regulation, cultural intelligence, pedagogical intelligence, moral and ethical reasoning under genuine stakes.

AI capability: Simulates — does not feel. No stakes, no skin in the game. The danger is not that the output is wrong. The danger is that the capacity atrophies in the person who stops exercising it.
Courses: Ethical Play (Book 4) + AImagineering (Book 3)
Tier 4

Metacognitive & Supervisory

Plausibility auditing (knowing an answer is wrong before recomputing), problem formulation (deciding what is worth solving), tool orchestration (knowing which tool to use, when, and whether to trust it), interpretive judgment (what does this result mean in this context), executive integration. Gardner never named these. This is the gap.

AI capability: Poor. The intelligences that oversee all the others. Tier 1 without Tier 4 is a very efficient way to be confidently wrong at scale.
Courses: Conducting AI (Book 5) + AImagineering (Book 3)
Tier 5

Causal & Counterfactual

Interventional reasoning (Pearl Rung 2 — if I do X, what happens?), counterfactual reasoning (Pearl Rung 3 — what would have happened?), causal model formulation (constructing the causal graph itself). Current AI is superhuman at Rung 1 (observation) and nearly absent at Rungs 2 and 3.

AI capability: Absent for formulation; weak for intervention; very weak for counterfactual.
Course: Causal Reasoning (Book 2)
Tier 6

Collective & Distributed

Collective intelligence (groups solving what no individual could — science, markets, democracy), Ubuntu intelligence ("I am because we are"), collaborative synthesis (teams whose output exceeds summed individual ability). Not a property of any individual — emergent from systems of people in relationship. LLMs are a lossy compression of collective human intelligence: they reflect the record, not the practice that generated it.

AI capability: Absent by definition. The algorithm has access to the literature. It is absent from the practice that generates new knowledge by definition.
Course: The Collective (Book 6)
Tier 7

Existential & Wisdom

Existential intelligence (grappling with meaning, mortality, cosmos), phronesis — Aristotelian practical wisdom (knowing when and how to apply what you know, and when not to; requires stakes and the possibility of loss), narrative identity (making a coherent self from lived experience over time).

AI capability: Absent. Requires being alive, mortal, and situated in time. An algorithm has no stakes. It cannot commit because it cannot lose.
Not a standalone course — the horizon the series points toward. Addressed via AI Sherpa (Companion 2) and the AImagineering Commit stage.
The Gardner trap: Gardner named multiple intelligences and cracked something open. Forty years later there is still no validated assessment for intrapersonal intelligence. The framework became vocabulary. It never became a curriculum. The Irreducibly Human series is explicitly Stage 1 (Name it) of a three-stage sequence: Name → Teach → Measure. Stages 2 and 3 require the measurement infrastructure currently in development in collaboration with the Center for Curriculum Redesign.
03

Core series — six books

#Full titleTaglineTier(s)
Book 1 AI Literacy, Fluency, and Trust The entry point. How to operate the cognitive forklift without being replaced by it. Tier 1
Book 2 Causal Reasoning The identification layer. What causes what, and why no algorithm can answer that for you. Tier 5
Book 3 AImagineering Post-AI design thinking. One week on ideation. The rest on the judgment that makes it matter. Tiers 3, 4, 7 partial
Book 4 Ethical Play Build a game that makes a player feel moral weight. Survive the AI audit. Prove the ethics are in the mechanics. Tier 3 deeply
Book 5 Conducting AI The five supervisory capacities no algorithm possesses. Hear the wrong note. Choose the piece. Direct the sections. Tier 4 fully
Book 6 The Collective Intelligence that cannot be possessed. Only accomplished. Together. Tier 6

Companion books

#TitleReader and focusTier addressed
Companion 1 A Teacher's Guide to AI for Embodied Learning 15 domain chapters: lab science, woodshop, PE, studio art, music, dance, nursing simulation, surgical training, architecture, culinary, theater, physical therapy, early childhood, special education, trades. Each self-contained. Each answering: where does AI help the teacher, and where must the teacher protect what only the body can learn. Tier 2 — via the teacher, not the student
Companion 2 The AI Sherpa: A Practitioner's Guide for Experiential Learning Co-op coordinators, clinical placement directors, study abroad advisors, anyone who sends students into the world to learn. Experience maps, living journals, the Sherpa before/during/after. Phronesis cannot be taught in a classroom but it can be scaffolded in the field. Tier 7 — via experience; the horizon the series points toward

Tier coverage map

TierNameAI capabilityCoverage
Tier 1Pattern & AssociationSuperhumanBook 1: AI Literacy, Fluency, and Trust
Tier 2Embodied & SensorimotorWeak / emergingCompanion 1: Teacher's Guide — not a standalone course
Tier 3Social & PersonalSimulates, doesn't feelBook 4: Ethical Play + Book 3: AImagineering
Tier 4Metacognitive & SupervisoryPoorBook 5: Conducting AI + Book 3: AImagineering
Tier 5Causal & CounterfactualWeak to absentBook 2: Causal Reasoning
Tier 6Collective & DistributedAbsent by definitionBook 6: The Collective
Tier 7Existential & WisdomAbsent — no stakesCompanion 2: AI Sherpa + AImagineering Commit stage
04

Book inventory and status

BookStatusNotes
Book 1 — AI Literacy, Fluency, and Trust Exists Full syllabus complete. Tic TOC prompt written.
Book 2 — Causal Reasoning Near complete Syllabus complete and submitted. Book approximately 2 weeks to completion. Syllabus serves as the Tic TOC outline.
Book 3 — AImagineering Architecture done Architecture complete. Needs Tic TOC run (~2 hours). Prompt not yet run.
Book 4 — Ethical Play Architecture done Architecture complete. Tic TOC prompt written. Full syllabus produced.
Book 5 — Conducting AI Prompt ready Tic TOC prompt written. Not yet run.
Book 6 — The Collective Partial Research base complete. Tic TOC prompt not yet written. Most novel course in the series — no comparable course exists elsewhere.
Companion 1 — Teacher's Guide Prompt ready Tic TOC prompt written. Full syllabus produced.
Companion 2 — AI Sherpa TOC complete Full 18-chapter TOC compiled. 15 open questions in register. Two are blocking: Chapter 16 legal review (OQ-006) and Chapter 18 dual-use document policy (OQ-007).

Tic TOC prompt status

BookPrompt status
Book 1 — AI Literacy, Fluency, and TrustWritten
Book 2 — Causal ReasoningWritten — syllabus is the outline
Book 3 — AImagineeringNot yet run — needs ~2 hours
Book 4 — Ethical PlayWritten
Book 5 — Conducting AIWritten
Book 6 — The CollectiveNot yet writtenUrgent
Companion 1 — Teacher's GuideWritten
Companion 2 — AI SherpaWritten — TOC compiled
05

Production pipeline

A complete infrastructure exists for producing and validating technical content. The series is, deliberately, demonstrated by the method used to build it.

Bookie the Bookmaker
Chapter drafting. Primary production engine for book content.
Tic TOC
Textbook architecture consultant. Full command library for publication-ready tables of contents. Generates complete chapter-by-chapter documentation with learning outcomes, prerequisite maps, and sequencing logic.
Popper
Assertion verification. Scans thousands of claims, flags suspect ones for expert review. Named after Karl Popper.
Figure Architect
Identifies high-assertion zones, generates figures and labels to reduce assertion density and improve legibility.
Zelda
Game design document consultant. Adapted for the Ethical Play course's GDD production workflow.
CAZE
Custom case study generator for any domain. Generates instructor version with full analysis. Used across all six courses.
CRITIQ
Peer review and paper development protocol. Used in capstone and research-track assessment.
SOCRIT
Socratic prompt evaluation using the Paul-Elder framework. Used to evaluate discussion quality.
Subby
Substack writing assistant. Supports the Theorist.ai publishing pipeline.
DevProof
Developer portfolio verification. Measures intent and decision patterns, not just syntax.
Medhavy
Adaptive learning platform with multi-armed bandit pedagogy engine (Thompson Sampling). Five learning modes. White-label architecture.
The Boyle System
AI-assisted documentary infrastructure for scientific reproducibility. MVAL protocol. Piloted across 150+ Humanitarians AI Fellows. Mentor meeting gap-review time: 60% → 20%.

Deployed work

WorkStatusImpact data
Cancer Biology and Therapeutics Deployed 38-chapter textbook written in approximately 1 month. In production in an NIH program. Positive cohort feedback.
The Boyle System Piloted Mentor meeting gap-review time: 60% → 20%. 150+ Humanitarians AI Fellows.
Humanitarians AI (501(c)(3)) Active 150+ Fellows. Applied AI research in humanitarian contexts.
Causal Reasoning (Book 2) Submitted Syllabus submitted for course approval. Course 1 in the series proposal.
06

To do

Immediate — before March 31

Urgent
Send the short course-proposal email. Attach Course 1 syllabus (Causal Reasoning). Describe Course 2 (AImagineering) in 2 sentences. Do not over-explain.
Urgent
Run AImagineering (Book 3) through Tic TOC. Approximately 2 hours. Unblocks Course 2 syllabus submission and the March 31 deadline.
High
Submit Course 2 (AImagineering) syllabus for course approval.
High
Write the 2-page series vision document. One document serves the institutional conversation, the Microsoft conversation, and the Fadel collaboration framing simultaneously.
High
Disclose tool use to the ISTE contact before submitting to EdSurge or EL Magazine. The disclosure needs to happen before submission, not after.

Book production

High
Write the Tic TOC prompt for The Collective (Book 6). Research base is complete. Most novel course in the series — no comparable course exists elsewhere.
Normal
Run Conducting AI (Book 5) through Tic TOC. Prompt is written.
Normal
Write The Economist follow-up piece. The exchange — accusation, defense, partial concession, deeper reading — has been identified as strong public Substack material. Has not been written.
Normal
Document the production status of the Voodoo Economics game. A complete GDD exists. Its relationship to the Ethical Play course is clear. The game's own production status has not been discussed.
Normal
Explore and document the hedge fund theory. Mentioned as developed for a hedge fund and potentially applicable to education. Not explored in any recorded conversation.

AI Sherpa — blocking items

Critical
OQ-006: Initiate legal review for Chapter 16 mandatory reporting language. Clinical domain chapter. Liability risk. Template cannot ship before review. Long lead time — start now.
Critical
OQ-007: Design the employee consent and ownership framework for Chapter 18 dual-use refiguration document. The tension between employee developmental ownership and program talent-development data must be resolved before production.
Urgent
OQ-012: Create character continuity sheets for Characters B–E. These characters appear across multiple chapters and contributors. Required before first contributor assignment. Lead author task — not delegable.
Urgent
OQ-013: Write the "Getting Started" minimum viable deployment note for Chapter 9. Two pages. Required for the publisher proposal.
Normal
OQ-010: Platform-agnostic language audit. Named AI products in the main text tie shelf life to vendor. Audit before final manuscript submission.
Normal
OQ-015: Negotiate rapid edition cycle rights before contract signing. AI capability claims will age quickly. Rapid cycle rights protect accuracy over time.

Institutional and structural

High
Formalize the conversation with the experiential learning office. They initiated contact about an AI Sherpa. The conversation has not been formalized.
Normal
Document what is being built for the Center for Curriculum Redesign collaboration specifically. Named as a collaborator; specifics not recorded. Stages 2 and 3 (Teach → Measure) depend on this.
07

New session setup

Documents to paste into a new session

  • This document
  • The seven-tier taxonomy (full table version from "Knowing Enough to Distrust the Machine")
  • Course 1 syllabus (Causal Reasoning — complete)
  • AImagineering architecture document (context prompt for new Claude session)
  • The Medhavy white-label architecture document
  • The Boyle System technical documentation

Most likely immediate tasks

  • Draft the short course-proposal email (attach Course 1 syllabus, two sentences on Course 2)
  • Run AImagineering through Tic TOC
  • Write the 2-page series vision document
  • Write the Tic TOC prompt for The Collective (Book 6)
08

Style notes

Prefers pushback over validation.
Works fast and builds functional things — prioritize drafts over frameworks.
"I build shit" — the person who makes other people's ideas tangible.
When Nik says "notes to me" — do not critique as finished writing.
The audience for course proposals is skeptical engineering faculty.