The Full-Stack Functionalist Theory. Thorat and Chopra Profile Awareness
If consciousness is functional, a basic question follows. What does being aware of something actually change in how a system handles it. A working paper posted to arXiv on September 27 by Sushrut Thorat and Paras Chopra of the Lossfunk research lab takes that question seriously enough to answer it with evidence rather than intuition, and the answer they assemble is deliberately small. Towards a full-stack functionalist theory of consciousness (arXiv:2609.33366) finds one strong function, one weaker one, and one clear negative case, then argues that this functional profile is only half of what a theory needs.
The paper states its own status in its first pages. It does not present a completed full-stack functionalist theory of consciousness. What it presents is a method and a first slice of the profile, built by sweeping the literature for comparisons between aware and unaware processing that survive a specific control.
The question and the control
The core question, in the paper’s own words, is this. For a particular content X and mental or bodily process Y, does awareness of X change whether, or how well, X can be used in process Y. The difficulty is confounding. An aware condition usually comes with better information about X, so any performance difference could reflect information quality rather than awareness itself.
The paper’s method demands comparisons that rule that out. Performance on a direct task probing X must be matched across the aware and unaware conditions, following the matching logic of Morales and colleagues and Samaha. Dissociations must be diagnostic, and the strongest cases are non-monotonic effects, where the relationship between signal strength and outcome reverses, which rule against particular signal-outcome relationships rather than against all signal-dependent explanations. The distinction between the process Y and the outcome used to measure it runs through the whole analysis.
What survives the control
The strongest current case is goal-sensitive control, specifically the regulation of attentional influence. The anchor study is Tsushima and colleagues in Science in 2006. Task-irrelevant motion at 5 percent coherence, below the direction-discrimination threshold measured separately in the same observers, disrupted performance on a central task more than motion at 20 percent coherence did. The effect is non-monotonic, more distraction from less visible signal, and the dip shifted when the perceptual threshold shifted with luminance contrast, which is what a threshold-linked mechanism should do.
The supporting cases line up. Lin and Murray showed in 2015 that invisible cues bias attention while visible cues instead produce suppression. Lamy and colleagues found in 2015 that irrelevant colour cues impose a same-location cost only when consciously perceived. A 2021 study by Ben-Haim and colleagues found rhesus monkeys performing above chance with awareness and below chance without it on a cue task, though the authors mark it suggestive because internal evidence was not matched. Blindsight patient GY’s exclusion performance, where higher blind-field contrast worsened exclusion, enters as related evidence.
The metacognitive case is weaker but present. Persaud and colleagues had GY wager on his own decisions, with performance matched between fields at 76 and 79 percent, while the wager-accuracy correlation ran at 0.52 for the normal field against 0.30 for the blind field. Vlassova and colleagues converge from a different paradigm.
The negative case is as important as the positive ones. Keizer and colleagues tested orientation-location binding with effective masking and found binding costs of 8.6 milliseconds against 5.8, with no masking-by-binding interaction. Short-lived feature binding and retrieval, in the paper’s summary, can proceed without awareness. One paradigm does not settle a function, and the authors flag exactly that, but it is the profile’s first boundary.
| Function | Evidence | Status |
|---|---|---|
| Regulation of attentional influence | Tsushima 2006, Lin and Murray 2015, Lamy 2015 | Strongest current case |
| Metacognitive evaluation | Persaud 2011 wager results | Weaker, converging |
| Short-lived feature binding | Keizer 2015 null interaction | Proceeds without awareness |
The other half of the theory
A functional profile is not yet a theory, and the paper is explicit about what the other half must cover. A full-stack account must also explain the structure and manner of presentation of perceptual and affective experience, subject-world organisation, awareness judgments, and problem reports. That means similarity relations among qualities, why contents seem directly present, the apparent subject-world form that Metzinger analyzed, and the reports of privacy, immediacy, and irreducibility that Chalmers and Kammerer have recently treated as data in their own right.
The paper’s key structural claim is that these two profiles constrain theory construction from different directions, and neither determines the process architecture that could connect them. Bridging the gap requires abductive construction. Propose candidate architectures, implement them, intervene on their components, and revise as further constraints accumulate. Attention Schema Theory is the worked example, with computational constructions of the schema and perturbation studies already in the literature. The authors also name the bad-lot problem, the possibility that the correct architecture is absent from the candidate set entirely, so abduction can only be as good as the lots it draws from.
The literature-sweep method behind the profile is being packaged as a separate programme at the authors’ repository, announced as forthcoming, with the sweeps intended to run as reproducible pipelines.
What this means for the AI consciousness debate
The paper makes no claims about current AI systems. Artificial systems appear only as motivation for the substrate-independence question, citing the Bayne and colleagues assessment and Butlin and colleagues’ indicator framework. That restraint is itself the contribution to this site’s coverage. The functional profile is exactly the kind of input an AI-consciousness assessment needs before any indicator is scored. A system either regulates attentional influence in the Tsushima sense, shows metacognitive sensitivity dissociated from performance, or binds features without awareness, and each answer means something different for which theories apply. The same discipline runs through the cacophony framework covered here last week, which turns indicator evidence into credences, and through Schwitzgebel’s ten-feature checklist, which treats functional coverage as probabilistic evidence. The full-stack paper supplies the empirical anchor those frameworks currently lack for the functional side.
One transparency note belongs in the record. The acknowledgements state that a frontier LLM was used to run literature sweeps and edit and structure parts of the paper, with the authors taking full responsibility for the outputs. The sweeps are themselves the paper’s evidentiary base, which makes the disclosure relevant rather than incidental, and the released sweep pipelines are the check on it.
Comparison to The Consciousness AI
The paper’s method maps onto this project’s measurement discipline directly. The Consciousness AI architecture treats consciousness as an emerging property of system dynamics and evaluates candidate properties through implementation and intervention rather than through behavioral impression, documented on the project repository. The full-stack programme asks for exactly that at the level of theory construction, candidate architectures built, perturbed, and revised against a growing constraint set. What the project’s architecture currently lacks is the matched aware-unaware comparison discipline the paper demands, and the paper supplies the experimental designs that discipline would need. The flagship field survey tracks where theory-driven and function-driven programmes meet, and this working paper is the most concrete functional-side entry so far.
Towards a full-stack functionalist theory of consciousness was posted to arXiv on September 27, 2026 by Sushrut Thorat and Paras Chopra of Lossfunk, with the literature-sweep programme at github.com/Lossfunk/consciousness-sweep.
Researchers covered here
- Sushrut ThoratLossfunkThe full-stack functionalist theory of consciousness programme, the aware-unaware comparison method, the functional profile of awareness
- Paras ChopraLossfunk (founder)Founding the Lossfunk research lab, the full-stack functionalist theory of consciousness programme, writing on evolution and complexity at Lossfunc