20.500.12556/RUL-189027 Vzgojiti umetno inteligenco? Educating artificial intelligence? Delo obravnava vprašanje, ali je mogoče vzgojiti umetno inteligenco, osredinja pa se predvsem na velike jezikovne modele kot trenutno najrazvitejšo in najbolj množično razširjeno obliko generativne umetne inteligence. Temelji na enaktivistični teoriji udejanjenega postajanja in avtopoetski paradigmi vzgoje, ki jo predstavlja ob konceptu uravnavanja umetne inteligence. Delo temelji na konceptualni raziskavi s pomočjo narativnega pregleda literature, v katerega so bili vključeni viri s področij pedagogike, sociologije, filozofije, kognitivne znanosti in umetne inteligence. V delu predstavimo tri glavne ugotovitve. Zdajšnjih velikih jezikovnih modelov ni mogoče vzgajati, če vzgojo razumemo skozi dimenzije udejanjenega postajanja, jezikovanja, moralnega razsojanja in subjektifikacije. Procesa uravnavanja in vzgoje nikakor nista enakovredna. Prvi predstavlja optimizacijo k vnaprej določenemu cilju, drugi pa odprt dialoški proces postajanja subjekta. Vseeno pa lahko etični cilji vzgoje informirajo načrtovanje prihodnjih modelov umetne inteligence v smeri razvoja multimodalnih sistemov, ki bodo lahko vsaj deloma udejanjeni. Pojem vzgoje je za opis trenutnih procesov uravnavanja umetne inteligence neustrezen in ostaja vezan na specifične pogoje človeškega postajanja. This work examines the question of whether it is possible to educate artificial intelligence, focusing primarily on large language models as the currently most advanced and widely used form of generative artificial intelligence. It is based on the enactivist theory of embodied becoming and the autopoietic paradigm of education, which is presented alongside the concept of aligning artificial intelligence. The work is based on conceptual research using a narrative review of the literature, which included sources from the fields of pedagogy, sociology, philosophy, cognitive science and artificial intelligence. In this paper, we present three main findings. Current large language models cannot be educated if we understand education through the dimensions of embodied becoming, language formation, moral judgement and subjectification. The processes of alignment and education are by no means equivalent. The former represents optimisation towards a predetermined goal, whilst the latter is an open-ended, dialogical process of becoming a subject. Nevertheless, the ethical goals of education can inform the design of future artificial intelligence models towards the development of multimodal systems that will be at least partially embodied. The concept of education is inadequate for describing current processes of regulating artificial intelligence and remains tied to the specific conditions of human becoming. vzgoja veliki jezikovni modeli umetna inteligenca udejanjeno postajanje uravnavanje jezikovanje education large language models artificial intelligence enactive becoming alignment languaging true false false Slovenski jezik Angleški jezik Magistrsko delo/naloga 2026-10-01 07:20:07 2026-10-01 07:20:08 2026-10-12 08:34:26 0000-00-00 00:00:00 2026 0 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 1970-01-01 635304 293365763 75071.pdf 75071.pdf 1 C045EEE45DD8928B1BE428C679F5D231 17e1d364525863a8cefcf45195c994a9f7903b668493b7a0ceb9e457439f8500 a8d8a174-bd57-11f1-8bc5-0050569b8976 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=249470 Filozofska fakulteta 0 0 0