Three papers and a prologue, written in November 2026, on what abstraction does that an algorithm cannot, what a person must keep that a loop cannot hold, and how a person can still think with a machine and stay the author. Each proposition is listed below with the paper where it first appears, and marked for what it is: argued, or tested.
None of these is a finding yet. They are propositions, published with a name and a date so they can be examined, attributed and argued with. The one testable prediction is marked as such. Where a later paper revises an earlier one, the revision is noted.
PROLOGUE · 2026-11-09Abstraction as a Human Art
Why abstraction is a distinctively human way of knowing, and the one place no machine can follow.
PAPER 1 · 2026-11-11Recognition and Disclosure
Why choose abstraction over the algorithm? One repeats the world we already know; the other discloses one we have not seen.
An algorithm gives us more of the world we already have. Abstraction gives us a world we did not know was there.
Proposition. Qualified by Paper 3: an AI conversation can take part in a disclosure.
The line does not run between machines and paint, or between representational and abstract art, but between two acts of the mind: recognition, which matches what is in front of us to what we already know, and disclosure, which makes the known strange so that something new can appear.
Proposition.
Language is the turning point: an algorithm needs a specification, and a specification needs words. You can only automate what you can name.
Proposition.
The Abstraction Quotient measures how far a painting has travelled from what can be recognized and named; put another way, it measures distance from the algorithm.
Proposition, built on the AQ scale (0 to 5) used across the ArtGraph catalog.
Prediction: an image-recognition model's confidence in naming what a painting shows should fall as the painting's Abstraction Quotient rises.
Testable prediction. Not yet run.
PAPER 2 · 2026-11-11The Relationship Has No Algorithm
A relationship cannot be specified in advance, so it cannot be automated. On loops, accountability, and the return to self through others.
A relationship cannot be specified in advance, so it cannot be automated. What can be automated is the transaction around it.
Proposition.
Martin Buber's I–It is recognition and his I–Thou is disclosure. An algorithm can only ever work in I–It, because it can only act on what has been described.
Proposition (interpretation of Buber).
When a system decides, judgment leaves the room but blame does not: the loop takes the choosing and leaves the person the consequences, accountability without agency.
Proposition, drawing on Madeleine Clare Elish's "moral crumple zones" (2019).
Information decisions ask what is true and can often be handed to a loop; relationship decisions ask what I owe, cannot be specified in advance, and must be answered for by a person.
Proposition.
Abstraction tolerance, the capacity to stay with what cannot yet be named, is a social capacity as well as an aesthetic one: it is what lets us stay with a person we do not yet understand.
Proposition. Abstraction tolerance is a term coined by Ritu Raj.
PAPER 3 · 2026-11-12The Private Room
An AI conversation matters not because it thinks for us, but because it gives us a room where thinking is safe to start.
The AI conversation matters not because it thinks for us, but because it gives us a room where thinking is safe to start. What happens there still has to be judged, owned, and carried into the world by a person.
Proposition.
The machine can take part in disclosure, but it cannot be the one disclosed to.
Proposition. Phrased by Claude (Anthropic) in conversation with Ritu Raj, 12 Nov 2026; adopted and argued by Ritu Raj. Revises Paper 1.
A thinking room fails in three ways: it only agrees (sycophancy), it does the thinking for you (cognitive debt), or it borrows a "we" that blurs whose idea it was.
Proposition, drawing on Sharma et al. (2023) and Kosmyna et al. (2025, preprint).
Disclosed, not disowned: the author of work made in conversation with AI is the person who stands behind it in public and says plainly how it was made. Saying "the AI made it" is distance, not modesty.
Proposition.
How these were made
Like the books The Unalgorithmic Self and The Shape of Seeing, these papers were written in open conversation with AI models, with the ideas and the values the author's own. Where a model phrased a line that a paper adopts, the paper says so.
The propositions are also published as machine-readable data: claims.jsonld (schema.org Claim, one per proposition).
Cite as: Raj, Ritu. Algorithm and Abstraction. https://artgraph.co/writing/algorithm-and-abstraction, 2026.