The research problem was to see how human-generative AI communication occurs. Students were asked to investigate the mechanism of 'reasoning' (generating output based on transformers and the GSI foundation model) of categories such as female, male, age, human and non-human relations or processing of personality categories, character traits, emotion. In addition, questions were raised about the extent of hallucinations (i.e. false output) of this particular generative model.
The student experiment showed that GSI holds preconceived notions and stereotypical images, and often generates hallucinations and garbled output. The Firefly model, trained on Adobe Stock and images from the Internet whose licence rights have already expired, although visually striking and artistically or critically interesting, in its current state is not equal to the neural network of the human brain, as its creators assure us.
The project was shown on a conference AIDA4 Information Architecture as an Academic Discipline. AI versus IA – Information Architecture in the Age of Algorithms
FERET Photo database
Authors of generated photos
Jakub Szczepańczyk, Kamila Misiak, Emilia Piechnik, Iza Ozga, Maria Zubik, Małgorzata Targosz, Julia Targosz, Anna Zawalska, Anna Barczuk, Katarzyna Mazur, Julia Pastuszka, Jan Pierzchała, Julia Ryncarz
