Chang-Eop, Kim
(2026)
Understanding Is Not a Scalar: What the Chinese Room Could Not Imagine.
[Preprint]
Abstract
Searle's Chinese Room argument derives its persuasive force from a specific implementation image: a person following a static rulebook of symbol-manipulation rules. The ``stochastic parrots'' critique revived the same intuition for contemporary systems. We argue that this line of argument constitutes an implementation-dependent thought experiment---one whose intuitive force diminishes when its implementation premises change. Two premises underlying the original image---static symbolic manipulation and internal opacity---are violated in significant respects by modern learning systems. Models trained on different modalities and domains converge toward shared representational geometries, and within individual models, non-spatial inputs give rise to geometric representations that mirror known structures of the domain. Large language models exhibit concept-selective self-reports causally grounded in internal activation states. These findings do not prove understanding exists, but establish that the system image presupposed by the Chinese Room no longer describes how modern systems operate. We propose a multi-dimensional framework that decomposes understanding into three conceptually separable dimensions: structural (capturing world structure in representational geometry), self-modeling (causally grounded access to internal states), and phenomenal (subjective experience). Under this framework, the Chinese Room's conclusion is valid for systems matching its implementation profile---high phenomenal experience but near-zero structural capture and self-modeling---yet this profile no longer applies to modern learning systems, and the ``stochastic parrot'' characterization is revealed to conflate absence on one dimension with absence on all. The question becomes which dimensions of understanding contemporary learning systems instantiate, and which remain genuinely open.
Monthly Views for the past 3 years
Monthly Downloads for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[img]](https://philsci-archive.pitt.edu/style/images/fileicons/text.png)
![[feed]](http://philsci-archive.pitt.edu/style/images/feed-icon-32x32.png){kind=icon}