Machine Ethology: Interaction as Umwelt in the Behaviour of Conversational Machines

This paper proposes an ethological framework for the empirical study of the behaviour of conversational machines, paradigmatically large language models. The canonical programme of machine behaviour (Rahwan et al., 2019) adapts Tinbergen's four questions to machines under an ecological premise: the machine is an actor immersed in a social ecosystem that pre-exists and persists independently of observation. This paper argues that the premise fails for the conversational machine, whose entire behavioural output is linguistic and elicited in exchange with a human interlocutor. It advances a single foundational thesis — the conversational machine's Umwelt (von Uexküll) is human interaction itself — from which two methodological consequences follow: the dissolution of the observer's paradox and the coincidence of observation and experiment. Under a Vichian epistemological warrant, the four Tinbergian questions are re-articulated for the interactional Umwelt and a fifth is added, proper to artefacts, concerning absence (the unsaid and the untrained). Each question is operationalised through prompting alone. The proposal is situated within recent convergent work (AI Agent Behavioral Science; Machine Culture) and specifically demarcated from so-called Machine Psychology. The document is bilingual (English and Italian).