Tianming Yang and consciousness as a continuous to discrete translation
Tianming Yang posted a preprint to arXiv on 21 August 2026 titled “Conscious Access as Continuous-to-Discrete Translation” (arXiv:2608.20723). It proposes that conscious access is a specific computational transformation, a structure-preserving translation between two distinct processing regimes, rather than a property to be located in either one. The framework, which Yang calls the Continuous/Discrete or C/D account, holds that the brain runs a distributed sensory-motor system over continuous high-dimensional manifolds and a centralized engine over discrete, scale-invariant symbols. Conscious access is the mapping that converts between the two.
The interesting claim is what the account does with the hard problem. Yang does not pretend to resolve the metaphysics of phenomenology. He argues that modeling conscious access as translational computation produces testable predictions about representational geometry, and that those predictions distinguish his account from the four dominant theories of consciousness currently on the table. That separates the framework from the theories it criticizes in the sharpest way a computational account can, by predicting different observable data.
The continuous and discrete systems
The C/D framework names two systems. System C is a distributed sensory-motor network whose states live on continuous, high-dimensional manifolds. Perception, motion, and their coupling operate here, in a regime where similarity is graded and there is no natural token. System D is a centralized engine structured around discrete, scale-invariant symbols. Language-like content, categorization, and reportable thought operate here, in a regime where states are countable and can be compared by identity rather than by nearness.
Conscious access, on this account, requires both a translation from localized continuous states to discrete symbolic tokens and an inverse projection that grounds those tokens back into sensorimotor dynamics. Neither regime is conscious by itself. The translation between them is what Yang identifies with access, the point at which graded experience becomes a reportable symbol that can drive flexible reasoning.
The measurable prediction
The account differentiates itself experimentally through representational geometry. Yang predicts a collapse from a graded similarity structure to a low-dimensional categorical equivalence class at the moment of access. Continuous representations that were all close in a graded space snap into discrete bins where the graded distances stop mattering.
That prediction is the part the established theories do not share. Global Workspace Theory predicts broadcast of content into a workspace. Integrated Information Theory predicts a specific cause-effect structure. Predictive Processing predicts prediction error minimization. Higher-Order Theories predict the presence of a higher-order state directed at a first-order state. Yang’s account predicts a particular geometric transformation, and the transformation is measurable with the representational geometry tools the interpretability community already uses. This is the same measurement language as the Jacobian lens workspace work of Gurnee and colleagues, which identified verbalizable content as a privileged subspace inside a language model.
How the account sits against the field
The positioning is deliberately contrastive. Yang states that the C/D framework should be tested against GWT, IIT, Predictive Processing, and HOT, and that its geometric predictions differentiate it from all four. That is a strong claim, and it is exactly the claim that the field’s adversarial collaboration methodology is designed to test. The Cogitate adversarial collaboration between IIT and GWT showed the field can design experiments where two theories make opposite predictions. The C/D framework is an attempt to enter that competition with a theory whose predictions live in the geometric space both combatants already measure.
For the current scientific consensus, the significance is architectural as much as empirical. The C/D account builds a principled bridge to neuro-symbolic AI. Yang argues that treating conscious access as translational computation gives a concrete architecture for neuro-symbolic systems, in which a continuous substrate feeds a discrete symbolic layer through a structure-preserving transform. That is a design template, not a claim that the template is conscious. The distinction is the same one the site draws in its own Global Workspace analysis of LLM architectures. A system can instantiate the functional architecture without any claim that it instantiates the experience.
Comparison to The Consciousness AI
The Neutral Core architecture of The Consciousness AI separates continuous and discrete processing along depth rather than along modality. The substrate layer runs spiking dynamics that are continuous in time, and the workspace layer operates on discretized semantic tokens. That split is structurally similar to Yang’s System C and System D, and the substrate console is the environment where a browser run of the spiking layer already simulates this division. The project’s own workspace benchmark and indicator checklist treat the continuous to discrete boundary as one indicator among several, and Yang’s account gives that indicator a specific geometric prediction the project could in principle test on its own representations.
The honesty constraint applies. The site does not claim its workspace layer implements Yang’s translation, and no current system in the project has been tested for the geometric collapse his prediction names. The overlap is architectural, and the value of Yang’s paper to the project is that it supplies a falsifiable shape for one of the indicators the checklist already tracks.
What the framework does not settle
The C/D account is a computational theory of access, and Yang is explicit that it sets aside the hard problem. It does not explain why the translation is accompanied by experience, only what the translation does functionally. That is a legitimate restriction of scope, and it is the reason the theory is testable. It is also the reason it cannot by itself settle whether a language model is conscious. The neuro-symbolic design space the field is exploring treats such architectures as objects to evaluate, not as objects whose experience can be assumed.
The strongest move in the paper is the discipline of prediction. A theory of consciousness that tells you exactly what geometric transformation to look for is worth more than a theory that tells you to keep thinking about the problem. Yang’s account does the first, and the field now has a concrete experimental target to aim at.
*Tianming Yang is a researcher at the Institute of Neuroscience of the Chinese Academy of Sciences. The preprint “Conscious Access as Continuous-to-Discrete Translation” was posted to arXiv on 21 August 2026 as arXiv:2608.20723.