Deflating Deflationism: A Critical Perspective on Debunking Arguments Against LLM Mentality

Many people feel compelled to interpret, describe, and respond to Large Language Models (LLMs) as if they possess inner mental lives similar to our own. Responses to this phenomenon have varied. Inflationists hold that at least some folk psychological ascriptions to LLMs are warranted. Deflationists argue that all such attributions of mentality to LLMs are misplaced, often cautioning against the risk that anthropomorphic projection may lead to misplaced trust or potentially even confusion about the moral status of LLMs. We advance this debate by assessing two common deflationary arguments against LLM mentality. What we term the robustness strategy aims to undercut one justification for believing that LLMs are minded entities by showing that putatively cognitive and humanlike behaviours are not robust, failing to generalise appropriately. What we term the etiological strategy undercuts attributions of mentality by challenging naive causal explanations of LLM behaviours, offering alternative causal accounts that weaken the case for mental state attributions. While both strategies offer powerful challenges to full-blown inflationism, we find that neither strategy provides a knock-down case against ascriptions of mentality to LLMs simpliciter. With this in mind, we explore a modest form of inflationism that permits ascriptions of mentality to LLMs under certain conditions. Specifically, we argue that folk practice provides a defeasible basis for attributing mental states and capacities to LLMs provided those mental states and capacities can be understood in metaphysically undemanding terms (e.g. knowledge, beliefs and desires), while greater caution is required when attributing metaphysically demanding mental phenomena such as phenomenal consciousness.

It is hard to shake the thought that Large Language Models (LLMs) have minds. When interacting with LLM-based dialogue agents like ChatGPT, Claude and Gemini, many users ask these systems for advice, joke with them, and confide in them. Some even fall in love. Accordingly, users often find themselves speaking to LLMs as if these systems have thoughts, desires, or points of view. Is the attribution of mentality to LLMs a mistake—an anthropomorphic glitch in our social cognition? Or could it be reasonable to treat LLMs as minded?1 The question of whether and when it is appropriate to ascribe mentality to LLMs has become a central flashpoint in the societal conversation on AI. While some researchers are engaged in exploratory work evaluating models for mental phenomena such as preferences, beliefs, and pleasure and pain states (Anthropic, 2025; Scherrer et al., 2023; Keeling et al., 2024), others remain deeply sceptical (Li and Etchemendy, 2024; Bender et al., 2021). We here distinguish two sets of views about LLM mentality which we term inflationism and deflationism.

Inflationism holds that at least some attributions of mentality to LLMs are accurate or appropriate, while deflationism denies this.2 In framing the debate in these terms, we aim to draw a broad heuristic distinction, rather than a strict conceptual demarcation: in practice, commentators are often not inflationist or deflationist tout court, but instead focus their arguments on some particular mental capacity or class of mental states.3 For example, Bender and Koller (2020) defend deflationism with respect to semantic understanding (c.f. Bender et al., 2021), and Søgaard (2023) defends the corresponding inflationist view.

We aim to step back from these debates and take a broader methodological perspective, examining two general deflationist strategies for debunking mental state attributions to LLMs: the robustness strategy, which challenges mental states attributions on a broadly functional basis, emphasising failures of generalisation; and the etiological strategy, which appeals to the causal history of LLMs to undermine the case for interpreting their behaviour in mentalistic terms.

We argue that both strategies fall short of decisively undermining inflationism, with a recurring problem being that they beg the question against inflationist alternatives. Once these debunking arguments are defused, a modest form of inflationism emerges as not only defensible but perhaps even the natural starting point in interpreting the behaviour of LLMs. In defending modest inflationism, our aim is not to claim that LLMs have rich mental lives, or that their internal workings closely resemble those of human minds. Rather, we suggest that some amount of mentalising is sometimes justified. In particular, where the mental states or capabilities at issue can be understood in metaphysically undemanding terms (as has been argued for the case of beliefs and desires), in contrast to more metaphysically demanding states such as phenomenally conscious experiences. Accordingly, the modest inflationism that we endorse registers that ascribing mentality to LLMs need not entail ascribing mental states and capacities which are equivalent to those of humans—in much the same way that ascriptions of beliefs, desires and intentions to non-human animals need not presuppose that their realisations of those mental states are exactly equivalent to ours.