I recently uploaded a paper to PhilArchive that proposes a formal, information-theoretic model of consciousness—not as subjective experience, but as the alignment between beliefs and objective descriptions.
In this framework:
Consciousness is the proportion of an object’s inherent description that an observer believes correctly.
Schizo-Consciousness refers to misbeliefs—statements the observer believes but which contradict the object’s true description.
Unconsciousness refers to unknowns—parts of the object for which the observer holds no belief.
Formally:
\text{Consciousness} = \frac{\text{Complexity of true beliefs (T)}}{\text{Complexity of full description (D)}}
Descriptions are represented using O(x)-Q(y) statements (objects and their qualities), and observers are modeled as possessing internal belief-updating codes influenced by stimuli.
Key features:
The model allows vector-based representation of belief alignment—two observers might have the same consciousness score but over different parts of the object.
Complexity can be measured via bit-length, code length, or entropy (Shannon or Kolmogorov).
The model supports comparative consciousness and simulates evolving belief states.
I’d love critical feedback. Is this a useful lens to formalize epistemic accuracy? Can this be applied to AI alignment or belief modeling? Are there already better formalisms I’ve missed?
🧠A Formal Model of Consciousness as Belief Alignment: Conscious, Schizo-Conscious, and Unconscious States
I recently uploaded a paper to PhilArchive that proposes a formal, information-theoretic model of consciousness—not as subjective experience, but as the alignment between beliefs and objective descriptions.
In this framework:
Consciousness is the proportion of an object’s inherent description that an observer believes correctly.
Schizo-Consciousness refers to misbeliefs—statements the observer believes but which contradict the object’s true description.
Unconsciousness refers to unknowns—parts of the object for which the observer holds no belief.
Formally:
\text{Consciousness} = \frac{\text{Complexity of true beliefs (T)}}{\text{Complexity of full description (D)}}
Descriptions are represented using O(x)-Q(y) statements (objects and their qualities), and observers are modeled as possessing internal belief-updating codes influenced by stimuli.
Key features:
The model allows vector-based representation of belief alignment—two observers might have the same consciousness score but over different parts of the object.
Complexity can be measured via bit-length, code length, or entropy (Shannon or Kolmogorov).
The model supports comparative consciousness and simulates evolving belief states.
The full paper is here (PDF, ~8 pages):
[https://​​philpapers.org/​​rec/​​HUGAIT-6]
https://​​drive.google.com/​​file/​​d/​​1IMexDlOqZuwNDtE4SAbB8AbdCsnx1qjg/​​view?usp=drivesdk
I’d love critical feedback. Is this a useful lens to formalize epistemic accuracy? Can this be applied to AI alignment or belief modeling? Are there already better formalisms I’ve missed?
—Anonymous