Vpliv umetne inteligence na vrednotenje vizualnih vsebin v grafičnem oblikovanju

Ekart, Barbara (author), Ahtik, Jure (mentor)

Namen diplomskega dela je bil raziskati, kako ljudje dojemajo oglase, ki so ustvarjeni z umetno inteligenco, in hkrati ugotoviti, ali se njihova vrednost zmanjša v primerjavi z oglasi, ki jih samostojno oblikujejo ljudje. V raziskavi smo preverjali, ali umetna inteligenca s svojimi generiranimi vizualnimi vsebinami preslepi potrošnike na način, da ti ne zaznajo, da gre za delo umetne inteligence. Metodologija raziskave je temeljila na dveh anketah, ki sta vsebovali enake pare oglasov nakita z enakimi spremenljivkami. V vsakem paru je bil en oglas generiran z uporabo ChatGPT-ja v kombinaciji z DALL·E, medtem ko je bil drugi ustvarjen z uporabo konvencionalnih orodij. Pri prvi anketi podatek o avtorstvu ni bil znan, pri drugi pa je bil naveden. Sodelujoči so imeli v obeh primerih informacijo, da je v vsakem paru en oglas generiran z umetno inteligenco, in so izbirali oglas, ki jim je bil v paru bolj všeč. Rezultati raziskave so pokazali, da so bili oglasi, generirani s ChatGPT-jem v kombinaciji z DALL·E, redkeje izbrani kot oglasi, ustvarjeni z uporabo konvencionalnih orodij. Ugotovljeno je bilo, da se je pri znanem podatku o avtorstvu delež izbranih oglasov, generiranih z umetno inteligenco, zmanjšal, ženske so bile izraziteje kritične in so posledično večkrat izbrale oglase, ustvarjene s konvencionalnimi orodji, moški pa so bili bolj nevtralni in ta podatek nanje ni bistveno vplival. Na podlagi ugotovitev je bilo zaključeno, da se lahko oglasi, generirani z umetno inteligenco, primerjajo z oglasi, ustvarjenimi z uporabo konvencionalnih orodij, po kakovosti in vizualni vsebini, vendar so kljub temu vrednoteni nižje kot vizualne vsebine, ustvarjene s strani oblikovalca, kar potrjuje, da ima podatek o izvoru pomemben vpliv na doživljanje in vrednotenje vizualnih vsebin.

• Language: Slovenian
• Keywords: grafično oblikovanje, percepcija oglasov, umetna inteligenca, vizualne vsebine
• Work type: Bachelor thesis/paper
• Organization: NTF - Faculty of Natural Sciences and Engineering
• Year: 2025
• PID: 20.500.12556/RUL-172048
• Publication date in RUL: 05.09.2025

Secondary language

• Language: English
• Title: Impact of artificial intelligence on evaluation of visual content in graphic design

Abstract

The aim of this work was to investigate how people perceive adverts created with artificial intelligence (AI) and to determine whether their value is reduced compared to adverts created by humans themselves. In the study, we tested whether AI deceives consumers through the visual content it creates, so that they do not recognise it as AI-produced. The research methodology was based on two surveys using identical pairs of jewellery ads with the same variables. In each pair, one advert was created using ChatGPT in combination with DALL-E, and the other using conventional tools. In the first survey, the authorship information was withheld, while in the second, it was disclosed. In both cases, participants were informed that one ad in each pair had been AI-generated, and they chose the one they preferred. The results showed that adverts created using ChatGPT in combination with DALL-E were chosen less frequently than those created with conventional tools. It was found that the proportion of AI-generated adverts chosen decreased further when authorship information was disclosed. Women were more critical and therefore more likely to select ads created with traditional tools, while men were more neutral and were not significantly influenced by authorship information. The findings suggest that AI-generated ads are comparable in quality and visual content to those created with conventional tools, but are still rated lower than visuals created by designers. This highlights the significant influence of provenance information on the perception and evaluation of visual content.

• Keywords: artificial intelligence, graphic design, perception of advertising, visual content