# Not Minds, but Signs: Reframing LLMs Through Semiotics  ## Metadata - Author: [[arxiv.org]] - Full Title: Not Minds, but Signs: Reframing LLMs Through Semiotics - Category: #articles - Summary: insert summary - My notes: - Summary: LLMs are not minds but machines that recombine and circulate signs. They don't understand like humans; their outputs gain meaning through human interpretation and cultural context. This view shifts focus from whether LLMs think to how they shape meaning, discourse, and practice. - URL: https://arxiv.org/html/2505.17080v1 ## LLM Chats ## NotebookLM ## LLM Audio ## Highlights > Building on the critiques of the cognitive paradigm outlined above, we propose an alternative point of view: instead of interpreting LLMs as digital minds that simulate cognition, we suggest understanding them as dynamic semiotic machines. This reframing marks a decisive shift away from metaphorical analogies with the human mind, rooted in the historical approaches of symbolic AI and neuro-inspired architectures, and turns our attention toward what these systems concretely do: organize, recombine, and circulate linguistic forms across diverse cultural and communicative contexts. ([View Highlight](https://read.readwise.io/read/01k2y5e0frrf883149hatcx0rq)) > Rather than speculating about hypothetical internal states such as intention, consciousness, or understanding, the semiotic approach focuses on observable sign processes. LLMs do not “understand” language in a human sense; they manipulate symbols probabilistically, producing outputs that gain significance only through situated interpretation. As such, these models function not as thinking entities but as operators within a broader ecology of meaning, where prompts, patterns, and cultural references interact to generate texts that invite human engagement and interpretive labor. ([View Highlight](https://read.readwise.io/read/01k2y5ezkjrcq99099x91af9y8)) > Rather than assuming that LLMs understand language or simulate human thought, we propose that their primary function is to recombine, recontextualize, and circulate linguistic forms based on probabilistic associations. By shifting from a cognitivist to a semiotic framework, we avoid anthropomorphism and gain a more precise understanding of how LLMs participate in cultural processes—not by thinking, but by generating texts that invite interpretation. ([View Highlight](https://read.readwise.io/read/01k2y4xcwdwnjcddzfgyt1bg68)) > Ultimately, this approach reframes LLMs as technological participants in an ongoing ecology of signs. ([View Highlight](https://read.readwise.io/read/01k2y4z650nafr6k9jmt9qd02w)) > LLMs should be seen as semiotic machines. That means we should think of them not as thinkers but as semiotic means that manipulate and organize signs, like words, phrases, and meanings, within cultural and linguistic systems. ([View Highlight](https://read.readwise.io/read/01k2y51mmxvm2p6zxywdhtx7dz)) > Chalmers [[8](https://arxiv.org/html/2505.17080v1#bib.bib8)] argues that current LLMs lack key features associated with consciousness, such as recurrent processing and unified agency, though he acknowledges the possibility that future models might achieve such states. Goldstein and Levinstein [[20](https://arxiv.org/html/2505.17080v1#bib.bib20)] explore whether LLMs possess minds by analyzing their internal representations and action dispositions, concluding that while LLMs exhibit some characteristics of intentional agents, definitive evidence of genuine mental states is lacking. Despite this popular tendency to regard agents as human-like, Kang et al. [[24](https://arxiv.org/html/2505.17080v1#bib.bib24)] investigate human perceptions of AI consciousness, finding that certain features in AI-generated text, such as metacognitive self-reflection and emotional expression, significantly influence these perceptions, despite the absence of actual consciousness in the models. ([View Highlight](https://read.readwise.io/read/01k2y52rprm6b8erap19ghz0v1)) > Then, we’ll develop a semiotic framework that treats LLMs as machines that work within systems of signs, not as intelligent agents, but as semiotic means that reshape meaning. ([View Highlight](https://read.readwise.io/read/01k2y57c0rm01zkxe1r9pyb25y)) > Every generated sentence is shaped by probabilistic proximities among linguistic tokens, yet it is not wholly determined by them: the prompt acts as a local semiotic perturbation, setting parameters within which the model’s generative pathways unfold. ([View Highlight](https://read.readwise.io/read/01k2y5j5n7arvze7ww1c1pyjya)) > Their utility and significance lie not in internal semantic processing as we argued in the Section [2](https://arxiv.org/html/2505.17080v1#S2), but in their capacity to dynamically recombine signs in culturally legible, and often surprising, ways. ([View Highlight](https://read.readwise.io/read/01k2y5r1cpe09m807f9r9hfmzg)) > So if we look at prompts, they function as semiotic acts: they act as structuring interventions that frame (in the Eco’s perspective) the interpretive conditions of the LLM’s output. ([View Highlight](https://read.readwise.io/read/01k2y5r73phavtt67hk4mvm0sr))