Language Models' Hall of Mirrors Problem: Why AI Alignment Requires Peircean Semiosis

David Manheim · 2026 · Philosophy & Technology 39:9 interlocutor medium priority coded

Main argument

Thesis: basic LLMs inhabit a 'hall of mirrors' - manipulating symbols that reflect only the linguistic surface of training data, without Peircean indexical grounding in a shared external world or participation in socially-mediated epistemology - so linguistically-encoded goals may diverge from real-world values (an epistemological challenge to alignment); BUT extended context windows, persistent memory, and mediated interaction with reality are moving newer systems toward being genuine Peircean interpretants, and 'we identify no fundamental architectural barriers that would prevent this'.

Why it matters here

The semiotic version of the grounding objection - LLMs reflect linguistic surface without indexical grounding, so 'a model's linguistic encoding of goals may diverge from real-world values' - but with a notable ANTI-pessimist twist: memory, extended context, and world-mediated interaction are moving systems toward genuine Peircean interpretants with 'no fundamental architectural barriers'. The strongest published challenge to treating the anti-understanding findings as permanent.

Reading notes

Compact treatment (Technion/ALTER). Abstract + framing read.

Manheim, D. (2026). Language Models' Hall of Mirrors Problem: Why AI Alignment Requires Peircean Semiosis. Philosophy & Technology, 39, 9.

Close reading — 2 coded units

#1 · pp. 1 · argument

“basic LLMs exist within a 'hall of mirrors,' reflecting only the linguistic surface of training data without indexical grounding in a shared external world, and manipulating symbols without participation in socially-mediated epistemology. [...] without grounding in the semiotic process, a model's linguistic encoding of goals may diverge from real-world values.”

AG-UNDERSTANDING

#2 · pp. 1 · argument

“newer developments, including extended context windows, persistent memory, and mediated interactions with reality, are moving towards making newer Artificial Intelligence (AI) systems into genuine Peircean interpretants, and [we] conclude that LLMs may be approaching this goal, and we identify no fundamental architectural barriers that would prevent this.”

AG-AGENTIC

Synthesis-matrix row

contradicts T5-AGENCY-DENIED-EVALUABILITY-KEPT

grounding deficit contingent; no architectural barrier to genuine interpretants

complicates T7-AGENTIC-BREAKS-FRAMES

agentic affordances may upgrade semiosis - cuts both ways

Memos (1)

theoretical · unit #2

Manheim is the needed counterweight in the moral-agency chapter: unlike the five empirical-incoherence studies, he argues the grounding deficit is CONTINGENT - agentic affordances (memory, persistent identity, world-mediated action) are exactly what would upgrade symbol-reflection into genuine interpretation, with no architectural barrier. Two consequences for the dissertation: (a) the anti-moral-agency conclusion should be indexed to CURRENT systems and stated with Manheim's contingency acknowledged (matching Gabriel & Keeling's fn14 openness to revision) - this is also the honest reading of Augustine's own experiment; (b) agentic AI thus threatens BOTH sides of his framework at once: it breaks pre-agentic responsibility frameworks AND potentially erodes the no-moral-agency premise. The dissertation's convergentist/distributed-responsibility architecture should be stress-tested against the 'grounded agent' scenario - a section Howard will appreciate as anticipating the strongest objection.