ТЕСТ ТЬЮРИНГА И РАСПРЕДЕЛЕННОЕ ЗНАНИЕ

Эпистемология и философия науки. 2026. Т. 63. № 2. С. 64–69.

Малышкин Е. В.

The article begins with several critical remarks about Konstantin Ocheretyany's paper. The conclusion of these remarks is that the dangers presented in the article under discussion, in the spirit of epistemic pessimism, are not frightening, and the problems raised are much older than the emergence of language models. The article is conceived in terms of expansion, although the conclusion seems to contrast competition for knowledge with "standing silently before the mystery". The author sees a way out in the attitude of existential romanticism. The latter, however, presupposes a set of techniques that can only produce limited effectiveness. This article proposes an alternative: since language models today pass the Turing test quite successfully, we cannot accept large language models as a tool, and we must describe interaction with them as distributed knowledge. For objects of distributed knowledge, it is essential whether this object is accessible to everyone or whether it cannot be accessible to everyone or even many (such as secrets). Since interaction with large language models, in our view, belongs to the second class, we will need to pursue a corresponding strategy. Finally, we propose an improvement to the Turing test, the Pico test, which would require the test subject to describe a selected object (for example, a literary work) in as much detail as possible and subsequently change their perspective in describing and interpreting it. The tester would need to pay attention to the natural difficulty humans have in abandoning their initial position, whereas a machine would be more easily able to change perspective, description, and style.

Keywords: Тест Тьюринга, Turing test, DISTRIBUTED knowledge, распределенное знание