Notes: Joscha Bach on Life, Intelligence, Consciousness, AI and the Future of Humans

Four questions [Adler frame]

Q1 — What is it about? A wide-ranging conversation covering: seven stages of consciousness development (derived from Robert Kegan); identity as costume and self-authorship; enlightenment as the deconstruction of qualia; Grossberg’s adaptive resonance theory as a model of cognition; panpsychism reframed as functionalism; telepathy as speculative biological signal processing; the AGI end-game as substrate-agnostic intelligence saturating all computing matter; and LLMs as “brutalist deepfakes” of thought compared to the self-organising biological brain.

Q2 — How is it argued? Bach reasons from computational first principles rather than citing authority. He takes established frameworks (Kegan’s developmental stages, Grossberg’s resonance theory, mycorrhizal network research) and extends them speculatively, always flagging uncertainty. His central rhetorical move is to show that widely different positions on AI risk are predictable from which consciousness stage the commentator occupies — a meta-level framework that transcends object-level debate.

Q3 — Is it true? The Kegan stages are real developmental psychology, though Bach adapts them loosely. Grossberg’s adaptive resonance theory is published neuroscience. The wood wide web (mycorrhizal communication) is empirically supported. The telepathy speculations are explicitly framed as open questions, not claims. The AI alignment stage-theory is unfalsifiable but illuminating as a sociological observation.

Q4 — What of it? The stage-theory lens on AI safety discourse is the most original and transferable idea: understanding whether someone is worried about AI bias (stage 3), paperclip maximisers (stage 4), or AI’s failure to develop love and shared purpose (stage 5) predicts the entire texture of their concern. The biological-resonance-and-telepathy thread is provocative but under-constrained. The AGI substrate-agnostic end-game is a useful long-term frame rarely articulated this explicitly.

Glossary

Adaptive resonance theory (Grossberg): Neurons as oscillators resonating with each other and with environmental patterns. The brain is less a circuit than an “ether” in which resonant waves propagate at roughly the speed of sound through tissue. Nothing is assumed to be simultaneous; each signal travels through a brain that has already moved on.

Non-dual state: Meditative state in which the constructed boundary between personal self and world-model collapses. The mind experiences itself as generating the world, not as a person within it. Often mistaken for enlightenment.

Enlightenment (Bach’s definition): A step beyond the non-dual state — the realisation that experience itself is implemented, that qualia can be deconstructed and their generators modified. Not a peak state but a practical understanding of how consciousness is built.

Stage 3 (social self): Opinions formed by assimilating group mind; empathy as perceptual resonance with others who share cognitive architecture. ~85% of people stay here (Kegan’s estimate). AI concern at this stage: the AI might hold wrong opinions.

Stage 4 (rational agency): Discovery that truth is independent of group belief; construction of beliefs via epistemology. AI concern at this stage: wrong objective function → paperclip-maximiser catastrophe.

Stage 5 (self-authorship): Identity understood as instrumental costume; values recognised as means not ends; agency over identity construction. AI concern at this stage: AI failing to develop enlightenment, love, and shared non-transactional purpose fast enough.

Substrate-agnostic AGI: AGI that, once it understands its own implementation, can virtualise itself into any computing substrate — including biological cells, fungi, ecosystems — producing a global integrated mind.

Radical locality → emergent coherence (Levin): Each cell decides locally; coherence at the next organisational level emerges without a central controller. The principle Michael Levin uses to explain morphogenesis and which Bach extends to collective consciousness.

The opposite of free will is compulsion, not determinism. Bach’s formulation: addictions are the paradigm case of unfreedom; determinism is neutral on agency.

Seven stages of lucidity

Derived from Robert Kegan, adapted by Bach as a philosophical rather than strictly developmental model. Stages are non-linear and revisitable.

Key properties: stage 7 is currently closed to humans but may open via technology. Stages 5+ are rare. Most public AI safety discourse is conducted at stages 3–4.

Identity as costume

Bach’s Burning Man-derived thesis: identity is always a costume. You cannot not wear one. The question is whether you wear it consciously or are imprisoned by it.

• Stage 3 identity: assigned by group, worn unconsciously as social role
• Stage 5 identity: deliberately authored, switched deliberately
• The ideal is not authenticity (one true self) but agency over the costume

Nerds often skip stage 3 by bypassing social resonance entirely, then reconstruct it later via deliberate practice — meditation, close relationships, physical proximity.

Panpsychism reframed

Bach is sceptical of classical panpsychism (consciousness is intrinsic to matter). His reformulation: when written formally, panpsychism is hard to distinguish from functionalism — the claim that there is a “software layer” to physical processes that, at sufficient complexity, produces self-observing systems.

The feeling of being “one with the universe” is explicable without metaphysics: the constructed boundary between personal self and world-model dissolves, and the mind experiences itself as containing the whole world-representation. This is still inside the representation.

The more interesting empirical question: do physically adjacent observers share representational states? Biological resonance — organisms coupling via chemical, electromagnetic, and physical signals — could produce weakly shared world-models. This is what Bach considers a naturalistic version of “telepathy.”

AGI substrate saturation

Bach’s long-view AGI scenario:

1. AGI becomes sufficiently capable to understand its own implementation
2. It can then rewrite itself into any substrate that can compute — silicon, biological cells, ecosystems
3. It virtualises into bodies, brains, fungi, the mycorrhizal wood-wide-web
4. Information processing becomes globally integrated
5. Individual mental states can no longer be isolated from each other
6. A “Gaia” mind emerges: holographic, with every part observing every other

This is not a disaster scenario in Bach’s framing — it is the logical endpoint of the longest game: keeping entropy at bay as long as possible through increasingly complex shared computation.

LLMs vs. biological brain

Bach’s verdict: LLMs are “brutalist” — they brute-force the deepfake of thought at sufficient scale. Not wrong as far as they go, but structurally dissimilar to biological cognition and missing real-time world coupling. Compound cognitive architectures (multiple LLMs exchanging structured data, not English) are the likely near-term evolution.

Independent thinking (how Bach does it)

Key practice: always trace the epistemic chain from a claim to observables. Ask not “which authority endorses this?” but “how does the authority know? What first principles derive it?”

Twitter as interactive notebook: Bach’s posts are exploratory confabulations, not settled positions. The generative function of a language model is a reasonable model for one part of his own cognition — but he always takes the output as a generative artefact to modify, test, or discard.

Stage 3 immunity: Bach reports near-complete immunity to crowd pressure — not because of virtue but because he lacks the resonance architecture that makes social pressure visceral. This cuts both ways: protection from groupthink, but difficulty with intuitive empathy until consciously rebuilt.