Gilestro, Giorgio F. (2026) The Mimicry Trap How We Define Intelligence to Exclude Inconvenient Minds. [Preprint]

	Text (manuscript) mimicry_trap_v11.pdf - Submitted Version Available under License Creative Commons Attribution No Derivatives. Download (169kB)
	Text main.pdf Download (218kB)

Abstract

When entities presumed incapable of intelligence produce behaviour that would ordinarily count as evidence of it, a recurrent response is to concede the performance while denying its evidential force: the behaviour is redescribed as imitation, simulation, contamination, or surface patterning. I call this structure the mimicry trap. The article argues that the trap is a cross-domain epistemic pattern that recurs wherever intelligence is at stake in an unfamiliar substrate: in humans of disfavoured groups (the eighteenth-century reception of Phillis Wheatley), in non-human animals (anthropodenial and anthropofabulation in comparative cognition), and in machines (contemporary debates over Large Language Models). Existing labels—anthropodenial, anthropofabulation, and the AI Effect—capture local instances; the present framework identifies their shared diagnostics: failure to specify falsifiers, asymmetric standards, shifting criteria, mechanism-based dismissal, and appeals to an unobservable missing essence. Applied to LLMs, the framework distinguishes legitimate scepticism about grounding, agency, embodiment, and benchmark validity from blanket denial that cannot say what evidence would change its verdict. I argue that recent behavioural and mechanistic evidence—including recoverable world models, algorithmic circuits, calibrated self-assessment, and high performance on some theory-of-mind and mathematical-reasoning tasks—raises the evidential cost of crude "mere mimicry" accounts. A Bayesian reconstruction of Turing's imitation game treats behavioural performance and mechanistic evidence as jointly relevant to intelligence attribution. Finally, drawing on Ockham's razor, I argue that "real understanding," "true semantic content," and "genuine intelligence" are legitimate theoretical posits only if they make independent empirical or explanatory differences; otherwise they function as non-identifiable entities preserving a prior verdict.

---

Export/Citation:

EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL

Social Networking:

Share |

---

Item Type:	Preprint
Creators:	Creators Email ORCID Gilestro, Giorgio F. giorgio@gilest.ro 0000-0001-7512-8541
Keywords:	artificial intelligence, philosophy of mind, functionalism, intelligence attribution, large language models, Ockham's razor, anthropodenial
Subjects:	Specific Sciences > Neuroscience > Cognitive Neuroscience Specific Sciences > Cognitive Science Specific Sciences > Artificial Intelligence General Issues > Science and Society General Issues > Social Epistemology of Science
Depositing User:	Dr. Giorgio F Gilestro
Date Deposited:	11 May 2026 12:41
Last Modified:	11 May 2026 12:41
Item ID:	29558
Subjects:	Specific Sciences > Neuroscience > Cognitive Neuroscience Specific Sciences > Cognitive Science Specific Sciences > Artificial Intelligence General Issues > Science and Society General Issues > Social Epistemology of Science
Date:	9 May 2026
URI:	https://philsci-archive.pitt.edu/id/eprint/29558

Available Versions of this Item

• The Mimicry Trap How We Define Intelligence to Exclude Inconvenient Minds. (deposited 04 Feb 2026 13:47)
  • The Mimicry Trap How We Define Intelligence to Exclude Inconvenient Minds. (deposited 11 May 2026 12:41) [Currently Displayed]

Monthly Views for the past 3 years

Monthly Downloads for the past 3 years

Plum Analytics

Actions (login required)

View Item