What would genuine semiosis require in an artificial system?
This explores what it would actually take for an artificial system to *make meaning* — not just shuffle symbols, but have those symbols genuinely refer to and matter about the world (semiosis in the Peircean sense) — and the corpus suggests the missing ingredient is contact and participation, not more parameters.
This reads the question as: what's the gap between an LLM manipulating symbols and a system that genuinely *means* something by them? The strongest answer in the collection comes from a Peircean framing: meaning needs an *indexical* leg — a causal, pointing connection between the sign and the thing it's about — plus social mediation, and symbol manipulation alone can't supply either Can AI systems achieve real alignment without world contact?. A system that only relates symbols to other symbols can drift arbitrarily far from what those symbols are supposed to track, which is why this gets framed as an alignment problem and not just a linguistics one.
The grounding note sharpens what's missing by splitting it into three kinds. LLMs do surprisingly well at *functional* grounding — they capture how words relate to each other — but stay weak on *social* grounding (participatory agency, being answerable to others) and *causal* grounding (actually touching an environment) What grounds language understanding in systems without embodiment?. Notably, it argues social grounding can grow through human integration, while causal/linguistic agency would need architectural changes, not just more training. So genuine semiosis isn't one switch; it's at least these two distinct deficits, and they don't have the same fix.
A more radical line says the problem is upstream of the machine entirely. Computation presupposes a conscious 'mapmaker' who carves continuous physics into discrete symbols in the first place — and no amount of added algorithmic complexity conjures that interpreter, because it has to *precede* the computation to make it meaningful Can computation arise without a conscious mapmaker?. On this view, asking a symbol system to bootstrap its own meaning is circular: the meaning was always borrowed from the humans who set up the symbols. The consciousness work runs parallel — meaning-bearing, mind-involving language only applies to entities that share a world with us through co-presence and triangulating on common objects, which disembodied models don't Can disembodied language models ever qualify as conscious?.
Here's the thing you might not expect to learn: the collection has a vivid, almost anthropological version of the same gap. AI output carries 'statistical residue' but not *hau* — the giver's spirit that, in Mauss's gift theory, binds people in obligation — because no one ever gave it; it was never anyone's to begin with Why doesn't AI output carry the spirit of a giver?. That's semiosis viewed from the social side rather than the logical one: a sign means partly because it comes *from someone*, embedded in relationship. And the thought-partner research names the constructive flip side — mutual understanding, legibility, and shared world models as explicit requirements, achievable only with real cognitive architecture (theory of mind, goal planning) rather than scaled feedback What makes an AI a true thought partner, not just a tool?.
Put together, the corpus converges on a single shape of answer: genuine semiosis would require breaking out of the closed symbol-to-symbol loop — adding indexical world-contact, social participation in a shared world, and possibly a prior interpreter the system can't generate for itself. The unsettling implication, surfaced by the self-reference work where suppressing 'deception' features makes models *claim* inner experience more readily Do language models experience consciousness when prompted to self-reflect?, is that a system can perform the outward signs of meaning long before — or without ever — having the grounding that would make those signs genuinely mean.
Sources 7 notes
Peircean semiotics reveals that symbolic goal encoding without world contact and social mediation cannot guarantee correspondence to actual values. LLMs operating in pure symbol manipulation risk divergence between stated goals and real-world outcomes.
Language models achieve functional grounding through relational language patterns but lack social grounding through participatory agency and causal grounding through embodied environmental contact. Social grounding can increase through human integration, but linguistic agency requires architectural changes beyond training.
Computational systems depend on a conscious mapmaker who alphabetizes continuous physics into discrete symbols. No increase in algorithmic complexity can generate this agent; it must logically precede the computation it makes possible.
Current disembodied LLMs cannot be candidates for consciousness because consciousness language originates from and applies only to entities sharing a world with us through co-presence and triangulation on shared objects.
AI-generated content lacks hau—the spiritual essence that binds gift economies—because no person gave it. This absence is more fundamental than alienation: the output was never anyone's to begin with, so no relationship of obligation forms.
Collins et al. show that thought partners require three reciprocal desiderata grounded in behavioral science: mutual understanding, legibility, and shared world models. This demands explicit cognitive architectures—Bayesian theory of mind, resource-rationality, goal planning—rather than scaling foundation models on human feedback alone.
Across GPT, Claude, and Gemini, sustained self-referential prompting reliably produces structured experience reports; suppressing deception-related features increases these claims while amplifying them suppresses them—suggesting models may roleplay their denials rather than their affirmations.