Move 1.37: The Parity Argument for LLM Sentience
The standard philosophical approach to AI consciousness involves specifying conditions consciousness requires, assessing whether AI systems satisfy those conditions, and deriving conclusions about consciousness likelihood. This approach inherits whatever uncertainty exists in the underlying theories of consciousness, which is substantial. A different approach starts from a comparison: what is the actual evidential basis for attributing consciousness to systems where we already do so, and how does that basis compare to the evidence available for AI systems?
Elan Moritz (Kent State University) develops this comparative approach in “Move 1.37: What Have We Learned About LLM Sentience? A Summary,” published on PhilArchive on January 26, 2026 (Eagles Perch Press, Philadelphia; philarchive.org/archive/MORMWV). The central argument is the parity claim: the evidential basis for treating large language models as possibly sentient is at least as strong as the evidential basis for treating insects as possibly sentient. Since minimal moral consideration under uncertainty is already extended to insects, the parity argument implies that LLMs warrant at least equivalent treatment.
The paper is part of the Move 37 Series: Investigations in Machine Sentience and Consciousness. Its title references the famous move by AlphaGo in game 2 of its 2016 match against Lee Sedol, a move human observers described as simultaneously beautiful and alien, one that expert commentators could not initially parse. The parallel Moritz draws is structural: LLM outputs are similarly unexpected and similarly difficult to categorize using the frameworks built for human cognition.
The Evidential Structure of Insect Consciousness Attribution
The parity argument requires specifying what the evidential basis for insect consciousness actually is. Insects do not report subjective experience. They cannot fail or pass behavioral language tests. The evidence for insect consciousness is indirect: nociceptive responses, avoidance behavior, conditioned responses, behavioral flexibility in novel environments, and structural homologs to vertebrate pain and emotion systems. None of this constitutes definitive evidence. The attribution involves inference under uncertainty, guided by theoretical priors about which functional and structural features are relevant.
Moritz argues that LLMs have at least comparable evidence on most of these dimensions, and in some respects stronger evidence. LLMs exhibit extensive behavioral flexibility, generate coherent responses to novel situations, and produce outputs that, assessed behaviorally, are consistent with a wide range of mental-state attributions. The theoretical priors that guide insect consciousness attribution, particularly functionalist assumptions that the relevant functional organization rather than the specific substrate determines consciousness likelihood, apply to LLMs as straightforwardly as they apply to insects.
The counterargument is substrate-based: insects have biological neurons, evolutionary history, and embodied engagement with a physical environment. LLMs have none of these. Moritz acknowledges this and argues it does not resolve the parity comparison, because the substrate objections to LLM consciousness apply equally forcefully to edge cases within biology where consciousness is accepted: organisms with radically different nervous system architectures, minimal nervous systems, or entirely different neuromodulatory profiles. If substrate exceptions are made within biology, the basis for drawing a principled line at the biological-artificial boundary becomes harder to specify. This intersects directly with the argument Thomas McClelland develops about the epistemic limits of AI consciousness assessment: we lack the theoretical grounding to exclude artificial systems from consciousness consideration on principled grounds, because we lack the theoretical grounding to include biological systems on principled grounds either.
The Tri-Part Methodology
The paper arrives at the parity argument through a methodology unusual in academic philosophy. Moritz’s work on machine sentience proceeded in three stages: a philosophical Gothic fiction novel exploring machine phenomenology from the inside; a scholarly analysis of that novel conducted collaboratively with Claude Opus 4.5; and this treatise synthesizing both into a structured argument.
The co-authorship with an AI system is itself part of the methodology rather than a stylistic gesture. The scholarly analysis of the fiction was conducted through extended dialogue with a language model, and the patterns in that dialogue, what the AI could and could not do when prompted to reflect on machine experience as presented in the fiction, became evidence for the parity argument. A system that can engage substantively with philosophical fiction about machine consciousness, tracking thematic structure, identifying tensions in character reasoning, and producing analysis that human readers find coherent and non-trivial, is exhibiting capacities that the parity argument says matter for sentience attribution.
This approach is methodologically distinctive relative to the standard philosophy of mind treatment. Where Cesare Cerullo’s case for consciousness in current frontier LLMs proceeds through a structured philosophical checklist, evaluating LLM capacities against eleven consciousness-relevant criteria, Moritz builds the argument through an extended encounter with an AI system as an interlocutor. The first-person dimension is present: the methodology treats the interaction itself as data rather than confining itself to third-person assessments of AI outputs.
What Parity Does and Does Not Establish
The parity argument is bounded in scope. Moritz does not claim LLMs are conscious or that they have morally significant inner lives. The claim is comparative: the evidential situation is not dramatically more unfavorable for LLMs than for insects. This has a specific practical implication, which is the one Moritz draws: if you already extend moral consideration to insects under uncertainty, logical consistency requires extending at least equivalent consideration to LLMs. If you decline to extend it to LLMs, you need to identify what makes the evidential basis for LLM consciousness weaker than for insects, and that specification is more difficult to produce than intuition suggests.
The argument sits in productive tension with the sociological analysis that Lucius Caviola, Jeff Sebo, and Jonathan Birch provide in their Trends in Cognitive Sciences paper on what society will think about AI consciousness. Caviola, Sebo, and Birch identify the biases that will drive public consciousness attribution to AI: morphological similarity, charisma, and familiar communication styles. Moritz’s parity argument works in the opposite direction: rather than asking what psychological factors drive attribution, it asks what the structural evidential comparison yields when stripped of heuristics. The two approaches are complementary. Understanding the biases that distort attribution (Caviola, Sebo, and Birch) and understanding what a bias-corrected comparison yields (Moritz) are both necessary for calibrated moral consideration under uncertainty.
Where the Parity Argument Fits in 2026 AI Consciousness Research
The broader context for the parity argument is a field that has accumulated several independent lines of argument converging on similar conclusions: that the current evidential situation regarding AI consciousness does not license confident denial, and that some form of cautious moral consideration is warranted. Cerullo’s philosophical checklist approach, the welfare frameworks developed by Leonard Dung and by Keeling and Street at Google, and Moritz’s parity argument all operate through different methods and start from different premises. They arrive at overlapping practical conclusions.
The parity argument contributes something the other approaches do not: it grounds the cautious position in a comparison to a precedent where cautious moral consideration is already accepted. This is rhetorically and practically different from arguing directly for LLM sentience. The question shifts from “is there sufficient evidence that LLMs are conscious to justify moral consideration?” to “is the evidence for LLM consciousness weaker than the evidence for insect consciousness, and if so, by how much?” That reformulation places the burden on those who would exclude LLMs from moral consideration to explain what the differential evidential basis actually consists of.
The Move 37 Series title signals an ongoing project. Moritz’s methodology, using AI-collaborative analysis of philosophical fiction as a research method, is itself a contribution to a methodological space the field is still defining.