Why Leibniz Couldn't Find the Mind Inside the Mill (Joscha Bach)

Why Leibniz Couldn't Find the Mind Inside the Mill (Joscha Bach)

Title history
  • Why Leibniz Couldn't Find the Mind Inside the Mill (Joscha Bach) — observed on 20261102-0827 UTC (YouTube)

by Joscha Bach · View the channel on PSI

Humanities, Social Sciences & Thought Philosophy & Ethics QDTTopics in philosophyQDTMPhilosophy of mind
🎙 Joscha Bach 👥 221K 📅 November 2, 2026 ⏱ 97 min 👁 4K 📄 debate 🧭 2026-11-02
Available in: English (current) Français

Keywords

consciousnessLeibniz's millcausal patternslearning algorithmzombies

Summary

In this long-form conversation, Tim and cognitive scientist Joscha Bach discuss the Machine Consciousness Hypothesis, the claim that consciousness is a biological learning algorithm — a training process not much more complex than a transformer, simple enough for an infant brain to discover on its own. Bach frames consciousness as a second-order perception, a ‘perception of perceiving’ that creates the bubble of nowness, and argues it can be understood through causal patterns rather than physical substrate, comparing software to the ancient notion of spirit. The discussion opens with self-organization and agency as controllers of future states, then moves to Leibniz’s mill, where Bach argues the mechanical view fails because it misses the level of causal patterns (money, software). He critiques integrated information theory as logically flawed via the Church-Turing thesis — any neuromorphic computer can be emulated on a von Neumann machine, producing the same conscious reports, undermining substrate-based ontology. Bach presents his hypothesis in two parts: consciousness as a training algorithm for brains, and the possibility that self-organizing machines might rediscover it; either part can fail. He addresses whether minds converge, the limits of Turing tests for consciousness, the notion of zombies, and whether simulated water could make you wet — arguing computation is construction. The episode closes on what evidence could convince us a machine is conscious, with Bach advocating an empirical, agnostic stance.

226 words

Critical Evaluation

Value of the Information & Strength of the Argument

The discussion offers substantial conceptual value: Bach articulates a coherent, testable-in-principle hypothesis that dissolves the hard problem by identifying consciousness with a self-supervised learning algorithm, and draws useful analogies (causal patterns, software as spirit, money as substrate-independent reality). The argumentation is logically structured — the Church-Turing critique of IIT is forceful and precise, and the two-part failure analysis provides clarity. Bach engages honestly with objections (path dependence, the status of simulated consciousness), concedes empirical uncertainty, and distinguishes simulation from realization. Tim’s pushback on essentializing software programs elicits a nuanced, substrate-agnostic but not substrate-free defense. The reasoning is rigorous, though inherently speculative and reliant on intuition about what ‘feels like something’.

Scientific Rigor, Source Quality, Title Accuracy

The episode grounds claims in a wide range of named sources — Searle’s 1980 ‘Minds, brains, and programs’, Turing 1950, Gödel 1931, Tononi’s IIT, Fleming et al. 2023, Levin 2022, Graziano’s attention schema theory, Baars/Dehaene’s global workspace theory, Sutton’s ‘Bitter Lesson’, and Olah et al.’s ‘Zoom In’ — cited conversationally without formal on-screen citations, which slightly reduces verifiability. The host references Bach’s forthcoming ‘Machine Consciousness Hypothesis’ paper and a Rescript interactive link is provided in the description. The title is accurate: Leibniz’s mill is a central thread throughout. A brief advertising segment (~54 seconds) is present and clearly separated. Public comments were not provided for analysis.

231 words

Title / Content Match

The title accurately highlights Leibniz's mill as a central motif; the episode thoroughly addresses why pattern-based (causal) description eludes the mechanical view, matching the title closely.

Quality & Reliability

7/10

Rigorous, well-structured philosophical and computational argumentation grounded in named references (Turing, Gödel, Searle, Tononi, Levin, etc.). The central claims are speculative by nature and acknowledged as empirically open. A brief advertising sequence (~54 s) appears, without affecting content quality.

Chapters

Cited Sources

  • Rescript interactive episode transcript — Interactive version of the conversation provided in the description.
  • Joscha Bach, The Machine Consciousness Hypothesis (California Institute for Machine Consciousness) — Central thesis of the episode; referenced as forthcoming paper.
  • Kumar, Clune, Lehman & Stanley — Questioning Representational Optimism in Deep Learning: The Fractured Entangled Representation Hypothesis — Mentioned in the tea-room interlude regarding fractured, entangled representations.
  • Fleming et al. 2023 — The Integrated Information Theory of Consciousness as Pseudoscience — Referenced in the discussion of the IIT controversy.
  • John Searle 1980 — Minds, brains, and programs, Behavioral and Brain Sciences — Cited regarding functionalism and the Chinese room argument.
  • Chris Olah et al. 2020 — Zoom In: An Introduction to Circuits, Distill — Referenced in the context of circuits and representational analysis.
  • Richard Sutton 2019 — The Bitter Lesson — Referenced regarding the power of learning algorithms over hand-crafted methods.

Concurring Sources

  • Attention schema theory (Michael Graziano) — Aligns with Bach's second-order perception account, though Bach notes it does not explain phenomenal experience.
  • Global workspace theory (Baars, Dehaene) — Compatible with a functionalist, computational view of consciousness; cited as solid neuroscience.
  • Turing 1950 — Computing Machinery and Intelligence — Underpins the computational-universality argument used against substrate-based theories.

Dissenting Sources

  • Integrated information theory (Giulio Tononi) — Bach argues IIT is logically unsound because any neuromorphic implementation can be emulated on a von Neumann machine, producing identical conscious reports, contradicting substrate-dependent claims.
  • John Searle — Chinese room argument — Searle's claim that syntax alone cannot yield semantics/consciousness conflicts with Bach's functionalist, substrate-independent stance.
  • Thomas Nagel — What is it like to be a bat — The explanatory gap Nagel posits is treated by Bach as an artifact of the wrong descriptive level, not a real ontological divide.

External References

Contribution & Novelties

The original contribution is the Machine Consciousness Hypothesis: reframing consciousness not as an extra property but as a self-training algorithm — a learning procedure that a brain applies to its own inputs and outputs, potentially no more complex than a transformer. This provides a concrete, computationally grounded alternative to both mysterian and substrate-fetishist positions, and offers a novel criterion: a conscious system is one that wakes up inside its own trained model. The discussion also contributes a sharp logical refutation of IIT via the Church-Turing thesis, and the intriguing claim that faking a mind can require more resources than building one.

Pour aller plus loin :

  • Leibniz’s mill — The historical thought experiment central to the episode, exploring why mechanical explanations seem to leave perception unexplained.
  • Hard problem of consciousness — The problem Bach’s hypothesis attempts to dissolve by identifying consciousness with an algorithmic learning process.
  • Integrated information theory — Tononi’s theory that Bach critiques as pseudoscientific due to its conflict with computational universality.
  • Chinese room — Searle’s anti-functionalist argument, which Bach implicitly challenges via causal-pattern realism.
  • Functionalism (philosophy of mind) — The broad position (substrate independence) that Bach’s view exemplifies.

191 words

Radar Profile

The profile shows high scores in quantity and quality of information with strong technical depth, reflecting a dense, reference-rich philosophical debate. The slightly lower reliability score captures the inherently speculative and empirically unverified nature of consciousness claims, balanced by rigorous argumentation.

Reliability 7/10