AI

This map is for readers who arrive with familiar AI questions and want to see where inepistemic makes an epistemic distinction. It is not a reading order and does not require reading the book from the beginning. Start from the concept that already interests you, then follow the linked chapters.

The book's recurring AI question is:

What happens when increasingly capable artificial systems become part of the environment in which people create knowledge?

If you want a short path

For a first pass through the AI argument, the most concentrated sequence is:

09 - Problems → 17 - Arrows → 21 - Error → 22 - Interfaces → 23 - Understanding → 24 - AI → 26 - Invention → 29 - Authorities

That path moves from problem generation, through epistemic transitions and error correction, into interfaces, understanding, human-AI interaction, invention, and governance.

For the fuller argument and the relations among these distinctions, use the longer Reader's Guide: Artificial Intelligence.

Capability, agency, and creativity

Familiar AI concept	Distinction sharpened in the book	Chapters
Capability	Capability / epistemic architecture. A system may search, generate, optimize, simulate, criticize, remember, or execute without those abilities by themselves explaining which problem governs the work or why a criterion should change.	00 - Universality · 09 - Problems · 23 - Understanding
Autonomy / agency	Unexpected behavior / epistemic agency. Independence is not established merely by behavior that surprises its designers. The stronger question is whether the governing problem, explanation, or criterion can itself become an object of criticism and replacement.	09 - Problems · 17 - Arrows · 21 - Error
Recursive self-improvement	Capability recursion / epistemic recursion. Improving search, generation, evaluation, optimization, or execution within a represented problem is different from creating and criticizing the problems that govern those capabilities.	08 - Criticism · 09 - Problems · 23 - Understanding
Novel output	Artifact novelty / explanatory creativity. Producing a novel proof, design, explanation, program, or image does not by itself explain the transition by which a new governing abstraction became available.	00 - Universality · 16 - Mystery · 26 - Invention
Recombination	Ancestry / derivation. Identifying the components, influences, prompts, prior art, or intermediate states of an artifact can explain where it came from without deriving the abstraction under which those components became a solution.	18 - Unpredictability · 26 - Invention

Alignment, control, and safety

Familiar AI concept	Distinction sharpened in the book	Chapters
Alignment	Alignment to a goal / criticism of the goal. Successful optimization against a supplied criterion is different from the capacity to make that criterion itself a problem.	19 - Values · 21 - Error · 29 - Authorities
Human in the loop	Human presence / ownership of the arrow. A nominally involved human may still lack the practical ability to reject the framing, change the criterion, reformulate the problem, or redirect the inquiry.	17 - Arrows · 24 - AI
Safety controls	Constraint / error-correction surface. Safety depends not only on preventing known failures but on preserving people, records, alternatives, criticism, reversibility, and time so that errors in the governing arrangement can themselves become visible.	21 - Error · 23 - Understanding · 29 - Authorities
Access control	Epistemic access / execution authority. Broad access to knowledge, modeling, criticism, and invention need not imply equally broad authority to perform irreversible actions.	20 - Resources · 21 - Error · 29 - Authorities
Governance	Present control / future problem authority. Control over infrastructure can affect not only present outputs but which future problems can be pursued, which criticisms can be operationalized, and which alternatives remain reachable.	20 - Resources · 21 - Error · 29 - Authorities
Speed / acceleration	Speed / epistemic resolution. The important variable may be whether there is enough structure between consequential transitions for failure to become a criticizable problem before conditions change again.	17 - Arrows · 21 - Error · 23 - Understanding

Users, interfaces, and personalization

Familiar AI concept	Distinction sharpened in the book	Chapters
Personalization	Adaptation / singular reconstruction. Tailoring an output to a user is different from explaining the individual problem history through which that output acquires meaning for this person.	02 - Populations · 13 - Inexplicit · 24 - AI
User modeling	Representation / causal explanation. A useful model of a person may predict or classify without explaining the singular causal history of the person represented.	01 - Genetics · 02 - Populations · 22 - Interfaces
Recommendation / assistance	Augmentation / prescription. A tool can enlarge the reachable problem space while also shaping which problems, categories, framings, or stopping points become salient.	03 - Societies · 20 - Resources · 24 - AI
Interface design	Abstraction assistance / abstraction authority. An interface can help a person navigate complexity, or it can make one compressed representation feel like the only reality available.	06 - Abstractions · 22 - Interfaces · 24 - AI
Information access	Access / epistemic reciprocity. Receiving inherited knowledge is different from having practical room to answer it through criticism, reconstruction, alternative creation, or redirection.	04 - Solitude · 20 - Resources · 24 - AI
Content control	Content / next epistemic encounter. An AI system can influence not only what information is supplied but which question, criticism, abstraction, or successor problem becomes practically available next.	03 - Societies · 24 - AI · 29 - Authorities

Memory, context, and long-running agents

Familiar AI concept	Distinction sharpened in the book	Chapters
Conversation history	History / problem genealogy. Preserving prior text is different from preserving the relations among problems, conjectures, criticisms, abandoned alternatives, and successor problems.	17 - Arrows · 24 - AI · 28 - Motion
State persistence	State persistence / epistemic persistence. Saving a state does not necessarily preserve why that state became relevant or what transformation produced it.	17 - Arrows · 28 - Motion
Memory systems	Output history / explanatory history. A log of what was produced is not the same as a reconstruction of which criticism or changed abstraction made the next production relevant.	12 - Justification · 17 - Arrows · 24 - AI
Persistent agents	Latest state / epistemic movement. Earlier failed conjectures, abandoned criteria, contradictions, and prior representations may remain causally useful even after the current state has changed.	17 - Arrows · 28 - Motion

Human-AI collaboration and understanding

Familiar AI concept	Distinction sharpened in the book	Chapters
Augmented intelligence	Epistemic reach / epistemic direction. AI can greatly expand what a person can practically reach without thereby explaining why one problem rather than another becomes significant.	20 - Resources · 24 - AI
Human-AI team / collective intelligence	Causal coupling / merged epistemic subject. A person and an AI can be tightly coupled in knowledge production without that relation by itself establishing a new collective knower.	04 - Solitude · 07 - Consilience · 24 - AI
Explainability / interpretability	Understanding / complete reconstruction. Useful understanding need not require possession of every underlying implementation detail, provided the relevant abstractions remain available to explanation, criticism, and reconstruction when they become problematic.	22 - Interfaces · 23 - Understanding
AI as a tool	Tool / responsive objective knowledge. Generative AI can return criticizable structure before a person's own problem or expression has stabilized, making the tool part of the process by which the problem becomes explicit.	13 - Inexplicit · 24 - AI
Productivity	Artifact volume / producer-first epistemic value. The first value of an AI-assisted artifact may be the change its production causes in the producer's own problem situation, even if the artifact is never published.	24 - AI · 26 - Invention