The Motivated Emotional Mind. Three Papers Extend Recurrent Processing Theory
Three arXiv papers posted within twelve days by Wieslaw L. Galus and Janusz A. Starzyk build one argument in three moves, and the argument targets the two theories this site covers most. The first move, posted September 29, challenges Global Neuronal Workspace Theory on its home ground. The second, posted October 2, extends Victor Lamme’s Recurrent Processing Theory with a motivational account of what recurrence carries. The third, posted September 20, turns the resulting architecture, the Motivated Emotional Mind or MEM, into a testable claim about when artificial intelligence could warrant a rational attribution of consciousness. The honest headline sits in the authors’ own conclusion. Most classical effects that Global Workspace Theory explains are compatible with MEM as well, which weakens the workspace interpretation’s claim to exclusivity, and MEM’s own distinctive claims remain hypotheses requiring causal testing.
One access note before the analysis. None of the three papers has an HTML full text on arXiv as of writing, so this article works from the abstracts, which are unusually detailed, plus the authors’ own statements about what remains unproven. Nothing below attributes experimental results beyond what the abstracts state, and the limits section says exactly where the record runs out.
What the Motivated Emotional Mind model is
MEM is described as an embodied, semi-hierarchical associative memory in which representational selection, action, recurrent reconstruction of modality-specific fields, interoception, and valence form a single functional cycle. Its component list runs through multilayer networks, semi-hierarchical semblions, receptors, effectors, allostasis, emotional valuation, and memory-based sensory reconstruction. A semblion is the architecture’s unit of associative memory, and the papers treat the semblion microstructure as one of the programme’s open hypotheses rather than a settled mechanism.
The concept doing the heaviest lifting is secondary perception. In the October 2 paper, secondary perception is the top-down reconstruction of sensory and interoceptive maps through which memories, mental imagery, dreams, hallucinations, feelings, and observed scenarios acquire experiential form. This is where MEM departs from the workspace picture. In Global Neuronal Workspace Theory, a representation becomes conscious when it wins access to a global broadcast channel. In MEM, experiential form arrives through reconstruction, the system re-building its sensory and bodily maps from memory under affective and allostatic control. The mathematical model formalizes re-entrant processing, representational competition, allostasis, affect, curiosity, procedural gaps, and action selection.
The Global Workspace challenge
The September 29 paper asks whether MEM mechanisms can reproduce the functions Global Neuronal Workspace Theory explains. Its method is a systematic juxtaposition. The authors examine 25 publications, twelve canonical studies of the workspace theory or critiques of it, and thirteen studies of perception, interoception, and affect, and for each they place the workspace interpretation next to a possible MEM reinterpretation.
The result is a compatibility finding rather than a victory. Most classical effects, including threshold dynamics, maintenance, recurrence, long-range connectivity, and flexible use of information, are compatible with both theories. In the authors’ words, this compatibility weakens the claim that the workspace interpretation is exclusive, but does not yet provide theory-specific support for MEM. The paper’s constructive output is a series of proposed interventions designed to dissociate four things that current evidence does not separate, sensory re-entry, global availability, report, and interoceptive valuation. Those dissociation experiments are the programme’s falsification handle. If a paradigm where the two theories diverge can be built and run, one of them loses cover.
For readers tracking the workspace side of this debate, the site’s analysis of Bernard Baars’ theory and ignition thresholds in attention-based architectures covers the empirical programme MEM is challenging, and the compatibility result lands on exactly the gap that the adversarial theory-testing literature keeps identifying, that rival theories fit the same data until an intervention separates them.
The Recurrent Processing extension
The October 2 paper positions MEM against Victor Lamme’s Recurrent Processing Theory. Lamme’s account explains the role of recurrence in perceptual organization and the transition to conscious vision, the difference between the fast feedforward sweep and the recurrent loop that follows it. The site covered the architecture consequence of that theory in the analysis of recurrent processing and the LLM attention recurrence gap.
MEM accepts the recurrence principle and asks what recurrence carries. Where RPT explains the transition to conscious vision, MEM addresses the content of recurrent processing, its grounding in receptor activity and bodily states, and its interaction with memory and action. The phenomenality claim follows. MEM associates phenomenality not merely with recurrent activity or reportability, but with the reconstruction of modality-specific, receptor-grounded sensory and bodily content.
That sentence is the programme’s centre of gravity, and it is also the sentence the authors themselves mark as unproven. The paper concludes that MEM can be viewed as an embodied and motivational extension of recurrent processing theory, while its distinctive claims, particularly the sufficiency of secondary perception for phenomenality and the proposed microstructure of semblions, remain hypotheses requiring causal testing. Sufficiency is the strong word there. The claim is not that reconstruction accompanies experience, but that reconstruction is enough for it.
The artificial cognition question
The September 20 paper asks under what conditions artificial intelligence could warrant a rational attribution of consciousness, and its answer starts with a negative list. Linguistic ability, multimodality, memory, planning, action control, and humanoid embodiment are not sufficient evidence of phenomenal experience. The positive side comes from biological precursors. Excitability, homeostasis, neural networks, and hierarchical representation identify functions whose counterparts, the authors argue, may be engineered.
MEM is the engineering proposal. Where conventional LLMs, hybrid embodied LLMs, vision-language-action systems, and current embodied agents stop at language and sensorimotor loops, MEM adds receptor-grounded representations, regulatory self-monitoring and interoception, affect, semblion-based associative memory, and re-entry into lower sensory and interoceptive maps. The paper’s central definition deserves to be quoted in full. A phenomenal state is a dynamic, integrated state of the whole embodied system in which sensor-grounded modal content is recurrently stabilized and coupled to a suprathreshold interoceptive representation of the system’s regulatory state.
That definition ties experience to the regulatory state of the system, not to its linguistic output, which is why the authors present MEM as a falsifiable research program requiring empirical validation and ethical caution rather than as a verdict about any existing system.
Comparison to The Consciousness AI
The definition’s machinery maps onto questions this project already treats as load-bearing. The Consciousness AI architecture implements homeostatic drives in its Affective Core through a PAD model that generates functional preferences under emotionally shaped rewards, documented on the project repository. That is a functional cousin of MEM’s allostatic regulation, built independently and on a smaller scale. What the project’s architecture does not implement is MEM’s second half, the re-entry of regulatory state into the reconstruction of modality-specific sensory maps, and the papers’ sufficiency claim would require exactly that loop to be tested. The project’s own framing, consciousness as an emerging property of system dynamics rather than a feature of linguistic behavior, is the assumption MEM’s negative list shares, and the Solms affective case study covered here in August makes the same move from the clinical side, grounding the consciousness criterion in felt homeostatic valence rather than in cognition.
What would settle it
The three papers leave a clear test agenda. The workspace challenge needs the dissociation interventions built, separating sensory re-entry from global availability, report, and interoceptive valuation, with preregistered quantitative designs. The recurrent extension needs a causal test of whether secondary perception is sufficient for phenomenality, which means building or finding systems where reconstruction occurs without recurrence, and vice versa. The artificial-cognition paper needs comparative and causal tests across humans, invertebrates, and artificial systems. Until any of that runs, the honest summary is the authors’ own. MEM is a falsifiable programme at hypothesis stage, and the field’s state-of-the-field survey now has one more entrant on the embodied side of its theory table.
The three papers by Wieslaw L. Galus and Janusz A. Starzyk were posted to arXiv as 2609.23828 (September 20, 2026), 2609.38495 (September 29, 2026), and 2610.04057 (October 2, 2026). Institutional affiliations are not stated on the arXiv records. This analysis is based on the abstracts and the authors’ own stated limitations, because full texts were not available at writing time.
Researchers covered here
- Wieslaw L. GalusInstitutional affiliation not stated on the arXiv recordCo-developer of the Motivated Emotional Mind model, the 2026 GNWT compatibility challenge, secondary perception as the locus of phenomenality
- Janusz A. StarzykInstitutional affiliation not stated on the arXiv recordThe Motivated Emotional Mind architecture, embodied machine consciousness research, the three-paper 2026 programme with Wieslaw Galus