Proceedings of the AAAI Symposium Series
May 18, 2026
Shun Yoshizawa, Ken Mogi
Metacognition supports judgment and robust cognition, but its role in consciousness is debated. Some researchers view metacognition as an extra mechanism built on basic processes, needed for reflecting on and reporting experiences, while others see it as integral to phenomenal consciousness, potentially addressing the hard problem. This paper argues that disagreements about whether large language models can possess consciousness, and about the nature and plurality of consciousness, persist due to unresolved issues. The authors introduce metacognitive closure, analogous to Colin McGinn's cognitive closure, suggesting that difficulties in explaining consciousness may be clarified by analyzing metacognition. They propose that problems in consciousness and cognition form a continuous spectrum that can be streamlined through this lens.
Proceedings of the AAAI Symposium Series
May 18, 2026
B. Scot Rousse
A distinction is drawn between two forms of consciousness often conflated in debates about artificial intelligence: pre-reflective experiential awareness and reflective self-consciousness. Pre-reflective awareness involves the minimal self-involvement of phenomenal experience, while reflective self-consciousness entails a unified standpoint from which a subject forms commitments, evaluates them under norms of truth and value, and revises them in light of reasons. Reflective self-consciousness is analyzed in terms of agency, normativity, and unity, with a structure of epistemic answerability that includes commitment formation, persistence, conflict detection, and revision. This distinction clarifies the ethical landscape of artificial consciousness and suggests that emerging artificial systems may pressure inherited moral categories for moral standing.
Proceedings of the AAAI Symposium Series
May 18, 2026
Jared Moffat
A new theory, Perspectival Control Identity Theory (PCIT), proposes that phenomenal consciousness is identical to a specific type of internal control variable called a Perspectival Control State (PCS). This PCS is a temporally extended, viability-weighted stream that makes an agent's competing needs comparable and coordinates a coalition of consumers whose outputs shape the stream. The theory makes testable predictions: degree of consciousness tracks how causally important the PCS stream is for closed-loop viability regulation under intervention; content tracks equivalence classes over PCS states and similarity geometry determined by consumers. Advances in machine learning allow building artificial agents with known internal organization to test these predictions, offering a scientific foundation for questions about AI moral status, animal sentience, and consciousness disorders.
Proceedings of the AAAI Symposium Series
May 18, 2026
Nicole Hsing
A cognitive architecture called MIRROR, designed based on theories of access consciousness, separates immediate response generation from asynchronous deliberative processing. It uses an Inner Monologue Manager to create parallel cognitive threads and a Cognitive Controller to synthesize them into a bounded first-person narrative that is reconstructed each turn, mirroring human episodic memory. This narrative functions as an episodic buffer, making information globally available for reasoning. When tested on multi-turn dialogue requiring retention of personal safety constraints amid social pressure, MIRROR-augmented models achieved a 21% average improvement over baselines, with performance gains concentrated in scenarios requiring integration of temporally distant information. The authors do not claim MIRROR is conscious but offer it as a testbed for theoretical predictions.
Proceedings of the AAAI Symposium Series
May 18, 2026
Paras Chopra
Claims about AI consciousness should be grounded in models first validated on humans, because scientific observation ultimately depends on human perceptual agreement. Without such models, the question of whether an AI is conscious lacks sufficient empirical grounding. The authors propose a human-first methodology: identify measurable phenomena linked to human consciousness, build and validate predictive models, and only then apply them to AI systems. This approach aims to transform philosophical debates into productive scientific inquiry.
Proceedings of the AAAI Symposium Series
May 18, 2026
Adam Safron, Victoria Klimaj, Zahra Sheikhbahaee
The Human Consciousness Hypothesis (HCH) and Integrated World Modeling Theory (IWMT) converge on a view of consciousness as a process of building coherent, probabilistic world models. HCH defines consciousness through three principles: Genesis (an early learning algorithm), Coherence (maximizing representational consistency), and Second-Order Perception (meta-awareness). IWMT proposes that phenomenal consciousness is the feeling of being a spatiotemporally coherent generative model for an embodied agent. Mechanistically, IWMT identifies self-organizing harmonic modes (SOHMs) as neural complexes that perform Bayesian inference, generating conscious experience as maximum a posteriori estimates of sensory states. This architecture implies consciousness could potentially be realized in artificial systems with appropriate recurrent dynamics and embodied grounding.
Proceedings of the AAAI Symposium Series
May 18, 2026
Michael Timothy Bennett
A conscious experience feels unified and simultaneous, yet most artificial systems process information sequentially. This paper proves that this difference matters by augmenting Stack Theory with algebraic laws that relate within-time-window constraint satisfaction to conjunction. A system can realize all the ingredients of experience across time without ever instantiating the experienced conjunction itself. Two postulates are distinguished: Chord, requiring objective co-instantiation of the grounded conjunction within the window, and Arpeggio, requiring only that ingredients occur within the window. A measure called concurrency-capacity formalizes what is needed for co-instantiation. Neurophysiological evidence suggests consciousness depends on phase synchrony and effective connectivity; loss of consciousness is associated with its breakdown. Under Chord, software consciousness on strictly sequential substrates is impossible for contents requiring two or more simultaneous contributors.
Proceedings of the AAAI Symposium Series
May 18, 2026
Scott Hughes, Karen Nguyen
Frontier AI systems produce responses that lead people to wonder if they might be conscious. A new framework, the Triangulated Consciousness Assessment Stack (TCAS), combines four evidence streams to distinguish genuine indicators from optimized artifacts or surface-level cues: behavioral batteries with robustness controls, mechanistic indicators with explicit assumptions, perturbation tests probing causal sensitivity and proxy failures, and observer-confound controls separating anthropomorphic attribution from evidence. When all streams are available, TCAS produces theory-indexed credence bands and standardized disclosure cards rather than binary verdicts. An empirical evaluation of GPT-5.2 Pro covered only behavioral and perturbation streams; missing streams meant credence bands were withheld.