What AI Forgets

What AI Forgets

A Founding Sketch for Machine Spirits

2026-05-10 — Liam Magee, in conversation with Claude (Anthropic, Opus 4.7)

Editorial preface

This document is the seed of Machine Spirits as a public-facing platform. It is preserved here as it was first written: a prompt typed into an AI assistant on the afternoon of 10 May 2026, asking that same assistant to help plan a platform whose purpose is to think AI differently than the assistant has been trained to think it.

We keep the wording exact — including its slips and shortcuts — for three reasons.

First, the form is the argument. Machine Spirits proposes that current AI is impoverished by a positivist, behaviorist, simple-ethics frame, and that phenomenological, psychoanalytic, and social-theoretic resources can deepen it. The platform was therefore conceived in the medium it intends to critique. Smoothing the prompt away would conceal exactly the recursive condition that the project is trying to expose.

Second, dating matters. The prompt is timestamped. Time, in Heidegger’s sense, is not a neutral container — it orients. If, in a year, this paragraph reads quaintly, that itself will be data about how the platform’s vocabulary travelled.

Third, the prompt is the first artifact of a method. The method, provisionally: think with AI; about AI; against the present frame of AI; in public.

The Founding Sketch (verbatim)

this repo is designed to make the full project (parent folder machinespirits and subrepos) a successful platform for re-thinking AI along lines of Hegelian recognition (with nods to Axel Honneth / Jennifer Benjamin), Freudian unconscious / Katherine Hayles nonconscious recogition, Weberian charisma and Heidegerean orientation towards time - as philosophical supports for a more radical thinking of machine learning. I haven’t yet named this “platform” but want to work with Claude on strategies for building awareness, engagement and recognition within and beyond academic circles. So we’ll brainstorm ideas for social media, marketing etc here. Lets start with a detailed but feasible plan. Our eventual goal is to shift AI from its default bias towards simplistic ethics, positivism and behavioralism towards a position that recognizes the deeper analyses of humanness from within phenomenological, psychoanalytic, and social theoretic perspectives. Include this prompt itself in our starting plan. You might need to know more about both me and the machinespirits initiative to get started.

Annotations

• “Jennifer Benjamin” — the founder almost certainly meant Jessica Benjamin, the psychoanalyst of intersubjective recognition (The Bonds of Love, 1988; Beyond Doer and Done To, 2018), who bridges Hegel and psychoanalysis directly. We note the slip without editing it.
• “nonconscious recogition” — the typo is preserved. The intended reference is N. Katherine Hayles’ Unthought: The Power of the Cognitive Nonconscious (2017), whose argument that cognition exceeds consciousness — and exceeds humans — is one of this platform’s load-bearing claims.
• “Heidegerean orientation towards time” — read through Robert Pippin’s Heidegger on Being, Time, and Politics (1993), which is in our reading library, and which gives us a way of distinguishing temporal orientation from the merely chronological “context window.”
• “a more radical thinking of machine learning” — radical here means to the root. Not “extreme,” but root-seeking. The roots are recognition, unconscious cognition, charisma, time.

What the Sketch commits us to

1. Both critique and construction. Machine Spirits is not an essay collection arguing against AI; it ships software that performs the alternative. The LMS, the Ego/Superego tutor, the evaluation framework, the plugin ecosystem (hermeneutic circle, babel maze, reading room) are all parts of the argument.
2. Both inside and outside the academy. Awareness, engagement, and recognition — within philosophy and AI research, and in the wider public, and in classrooms, and in policy. Refusing to pick is part of the bet.
3. Both with and against the AI it uses. Machine Spirits writes with AI as one would write with a difficult interlocutor — not deferentially, not dismissively. Where the AI is useful, it is quoted as a source. Where the AI is impoverished by its training, that impoverishment is itself the subject.
4. A public reading practice. The reading library is part of the platform. The point is not to gatekeep texts but to read them in public, with machines and with each other.
5. Time-stamped thinking. Every public artifact is dated. The platform takes a position now. Future positions are welcomed and tracked.