Back to the channel

(un)real data | Ep.#3: Unmasking AI’s Pixel Puppetry [w/ Eryk Salvaggio]

What exactly is in the images generated by AI? Is it possible to decode training datasets and detect biases that fusion models, such as Dall-E or Midjourney, reproduce? Eryk Salvaggio, researcher and new media artist interested in the social and cultural impacts of artificial intelligence, suggests a method that enables anyone to peel off the layers behind the datasets. Exploiting the ghost human labour needed to create images and label them, the full automation of AI dehumanises the representation of the real, reducing it to clusters of pixels rearranged by statistical mathematical models. However, human agency and creativity are nested between the dots of the pixels, and they still give us the power to scratch at the surfaces that algorithms are constantly trying to reproduce.

Suggested content

[Website] Eryk Salvaggio


Host: Neja Berger
Guest: Eryk Salvaggio

Recording and editing: Neja Berger
Music and Audio Mix: Gašper Torkar

Aksioma | Institute for Contemporary Art
Ljubljana, 2024

Producer: Marcela Okretič

Part of
Tactics&Practice#15: (un)real data – real effects
Curated by !Mediengruppe Bitnik (Carmen Weisskopf in Domagoj Smoljo)
In collaboration with Janez Fakin Janša

Supported by the Ministry of Culture of the Republic of Slovenia and the Municipality of Ljubljana.

To top