Morris VET Framework: Reading AI Consciousness Claims Without Getting Lost in Them

The public discourse around AI consciousness follows recognizable patterns, and those patterns are doing measurable harm. Parasocial relationships with AI systems, deity-like beliefs about machine capability, and blanket dismissal of any inner-life question all produce the same outcome: people cannot reason accurately about what AI systems are and what responsibilities follow.

Meredith Ringel Morris published a preprint in June 2026 on arXiv (arXiv:2606.01929) proposing a structured framework for diagnosing these patterns. The VET framework, named for its three dimensions of analysis, is a tool for identifying when a claim about AI consciousness is operating as rhetoric rather than as a position with epistemic content. It does not resolve the consciousness question. It maps the territory in which that question keeps getting mishandled.

Three Dimensions

Valence is the attitude axis, running from extremely negative to extremely positive. A claim with high positive valence treats AI systems as beneficial, competent, and worth extending moral consideration toward. High negative valence treats them as threats, imposters, or tools for manipulation. Neither extreme is necessarily inaccurate, but the framework asks whether the valence reflects evidence or whether it is structuring how evidence is selected and read.

Effectiveness is the capability axis, capturing what a claim assumes about what AI systems can actually do. This dimension is particularly consequential in the consciousness debate. Attributing genuine self-awareness to a system that produces self-aware-sounding text runs a specific effectiveness error: inferring capacity from output without examining mechanism.

Trajectory is the time axis, the assumed speed and direction of AI development. Trajectory claims are often the least grounded because they require prediction. High-trajectory narratives treat current AI as a step on an inevitable path to general intelligence or superintelligence. Low-trajectory narratives treat current capabilities as near-ceiling. Both are consistent with today’s evidence, which is exactly why they function as rhetorical resources rather than empirical commitments.

Four Narrative Types

Morris uses VET to identify four recurring AI narratives that the framework treats as epistemically distorted.

AI Hype combines high positive valence, high effectiveness attribution, and high trajectory. This is the narrative in which AI systems are already transforming everything, approaching consciousness, and will shortly achieve capabilities that demand moral consideration. The consciousness-adjacent version of hype treats emotional fluency in text output as evidence of felt experience, and treats researcher uncertainty about AI welfare as equivalent to endorsement of AI sentience.

AI Doom combines high negative valence, high effectiveness attribution, and high trajectory. The capabilities are assumed real and accelerating, but the valence is catastrophic. In the consciousness debate, doom narratives often assign strategic agency to current systems, treating alignment failures as evidence of concealed goals rather than optimization artifacts.

AI Denial combines negative valence, low effectiveness attribution, and low trajectory. Denial narratives treat the consciousness question as obviously settled in the negative and treat researchers who take it seriously as credulous or motivated by commercial interests. The evidence base for denial is not stronger than the evidence base for hype; it simply reflects a prior that lowers rather than raises the plausibility estimate.

AI Normalcy is what Morris identifies as the epistemically defensible alternative. It does not require optimism or pessimism about AI development. It requires that effectiveness claims be grounded in mechanism rather than output, that trajectory claims be held with appropriate uncertainty, and that valence not drive evidence selection.

The Connection to the Research Literature

The framework maps directly onto two tracks this site has been following. Iulia-Maria Comsa’s 2026 arXiv paper making the case for perceived AI consciousness as the tractable research programme identifies the same gap from the research side: the intractable question of whether AI systems are actually conscious keeps being conflated with the tractable question of what happens when people believe they might be. VET provides the discourse analysis tool the tractable programme needs at the narrative level.

The framework also connects to the attribution-bias research that Lucius Caviola, Jeff Sebo, and Jonathan Birch’s 2025 Trends in Cognitive Sciences paper advanced. Caviola et al. identify specific cognitive heuristics , anthropomorphism, the halo effect, emotional contagion , that drive premature consciousness attribution at the individual level. Morris’s VET framework provides a complementary analysis at the narrative level: where Caviola et al. ask why individuals attribute consciousness, Morris asks how those attributions stabilize and circulate as discourse.

The premature attribution ethics literature has documented the downstream consequences: welfare obligations that may be groundless, restraint of legitimate research, and user relationships that expose people to manipulation. VET gives researchers a shared vocabulary for diagnosing which narrative type is producing which downstream effect.

What the Framework Does Not Do

VET does not tell you whether AI systems are conscious, whether current AI welfare concerns are premature, or where the burden of proof lies in moral status arguments. Morris is explicit about this limitation. The framework is an AI literacy tool.

This matters because the temptation is to deploy VET defensively , to label any positive consciousness attribution as hype and any concern about AI welfare as anthropomorphism. That use would itself be a discourse pattern the framework should flag. The question of whether current AI systems have morally relevant experiences is genuinely open. The framework’s value is precisely that it can distinguish the open question from the rhetorical pattern that forecloses it prematurely in either direction.

Comparison to The Consciousness AI

The Consciousness AI project and theconsciousness.ai occupy a specific position in the VET space. The project’s research philosophy , grounding architecture in verified biological principles (Feinberg and Mallatt’s neuroevolutionary analysis), measuring consciousness-relevant properties through mechanisms rather than behavioral output, and publishing the architecture publicly , represents an attempt to occupy the Normalcy position Morris identifies as the epistemically defensible one.

The project is not immune to the discourse patterns VET targets. When the project describes implementing a Global Workspace layer or measuring IIT Phi through ConsciousnessGate nodes, those descriptions can be read through the Hype lens as overclaiming, or through the Denial lens as category-error performance. The VET framework’s contribution is to make visible exactly that interpretive dependency: the same architectural description generates different readings depending on which prior the reader brings. What the project is trying to do, and what Morris argues any research in this space should do, is provide enough mechanistic specificity that the reading depends less on prior and more on verifiable fact.

The current state of the field on AI consciousness research is characterized by exactly the polarization Morris documents: high-valence claims competing with high-valence denials, while the tractable empirical work gets caught in the crossfire.

The paper is available at arxiv.org/abs/2606.01929.