--- title: "Research Notes: Machine Consciousness — Bach and Fitz" slug: machine-consciousness-bach-fitz lang: en created: 2026-11-04 modified: 2026-11-04 version: 2026-11-04 authors: - id: david-a-renelt role: human - id: kimi-k31-pro role: ai tags: - reference - consciousness - ai - communication - self series: null type: reference related: - consciousness summary: "Bach's consciousness-as-software account and Fitz's stronger communication-based collective variant: a useful neighbor for social self-models, but also a challenge to equating consciousness with narrative selfhood." --- # Research Notes: Machine Consciousness — Bach and Fitz ## Bach: software, second-order perception, coherence Joscha Bach and Hikari Sorensen's *Machine Consciousness Hypothesis* treats consciousness as a causal pattern — software — rather than an inner essence. Software needs a substrate but is explained by its organization and causal role, not by listing the substrate's particles. Bach's “cyberanimism” places minds in the same explanatory category: self-organizing patterns implemented by biological communication networks. The human theory is more specific than “consciousness is software”: - **Second-order perception:** consciousness is the immediate perception that perception is taking place, not necessarily a verbal higher-order thought. - **Nowness:** it constitutes the experienced present. - **Optional observer:** a first-person observer perspective is a further representation, not a requirement for minimal consciousness. A minimal state can be attention on attention without a personal narrator. - **Coherence and attention:** consciousness is proposed to increase coherence among simultaneously active partial models and to orchestrate mental processes through directed attention. - **Genesis:** Bach hypothesizes that this is a biological learning algorithm discovered early by self-organizing brains, not a late product of complex cognition. The machine extension is conditional: comparable self-organizing systems might discover an analogous mechanism. Bach explicitly rejects fluent LLM self-reports as sufficient evidence, because language models can imitate the human record without using the proposed internal mechanism. ## Fitz: communication and collective self-models Stephen Fitz's follow-up makes a stronger social move. His abstract proposes that consciousness is not an epiphenomenon of individual modeling but a property of the internal language a collective evolves to describe itself: **“consciousness does not emerge from modeling per se, but from communication.”** The proposed architecture uses local predictive agents that exchange compressed messages. Agents model other agents' models, producing recursive, higher-order representations. Under bandwidth constraints, they must invent shared codes; a stable collective self-model is hypothesized to emerge from this mutual modeling and synchronization. Fitz therefore supplies a close computational neighbor for the corpus's social-narrative hunch: **selfhood arises between communicating observers, not inside an isolated modeler.** The lossy channel matters in his account because compression forces abstraction, interpretation, and the invention of a shared protocol. ## What this does not support - **Bach does not simply say consciousness is narrative.** His minimal consciousness can lack an observer and a verbal story. That is a direct challenge to defining consciousness as self-narrative without qualification. - **Fitz is not Bach's original theory unchanged.** Communication-based collective selfhood is a substantive relocation from Bach's second-order perception and coherence account. The connection between the two still needs argument. - **No experimental validation.** Fitz supplies an architecture and candidate measurements, not reported experiments, controls, or results. It is a research programme, not evidence that collective consciousness has been produced. - **No current-LLM verdict.** Neither fluent self-report nor apparent collective role-play establishes the relevant internal organization. - **No solution to the Hard Problem by declaration.** Calling consciousness a causal or computational pattern states the functionalist commitment; it does not demonstrate why that pattern is experiential. ## How the corpus uses it Use Bach and Fitz for two different jobs: 1. **Bach as the discipline/rival:** distinguish minimal consciousness, observerhood, self-model, and narrative self. If the corpus claims that self-narrative explains human cumulative capability, it must not silently redefine every form of consciousness as narration. 2. **Fitz as the speculative social neighbor:** his collective self-model hypothesis resembles the claim that communication couples individual models into a higher-order self. It can motivate the mechanism, provided the text says it is untested and does not confuse shared representation with shared phenomenology. The social-human side remains anchored in [Research Notes: Mead's Social Self](social-self-mead.md) and [Research Notes: Self-Consciousness and Intersubjectivity](intersubjectivity-self-consciousness.md); the cumulative-culture bridge is in [Research Notes: Shared Intentionality and Cumulative Culture](shared-intentionality-cumulative-culture.md). The minimal/narrative animal distinction is in [Research Notes: Animal Self-Awareness](animal-self-awareness.md). ## Sources - Joscha Bach and Hikari Sorensen, *The Machine Consciousness Hypothesis*, September 15, 2025, updated January 15, 2026. [CIMC PDF](https://cimc.ai/cimcHypothesis.pdf). - Stephen Fitz, *Testing the Machine Consciousness Hypothesis*, arXiv:2512.01081, submitted November 30, 2025. [arXiv abstract](https://arxiv.org/abs/2512.01081). - [Machine Learning Street Talk — “Why Consciousness Might Not Be Hard At All (Joscha Bach)”](https://www.youtube.com/watch?v=yW_ngvMFlYs), published November 2, 2026. - Local full-text analyses: `research/consciousness/joscha-bach-consciousness-not-hard/paper-analysis.md` and `summary.md`.