KI_
TCHEN

EVENT - 2021

Artificial Intelligence has an enormous amount of human input behind it. Data sets have to be defined, compiled, made understandable. The algorithms have to be written. Computers have to be built. Time and a lot of energy flows into a system exclusively only designed for one specific task. This begs the question, at what point is it even worth to train an AI versus just teaching another human the task at hand.

We trained a multiple models of GPT-2 on cookie and cocktail recipes. Generating new recipes at different dataset sizes and stages of training and then serving them at an event allows the visitor to directly taste the level of human input and judge for themselves, if they would like an AI Bartender.

Disclaimer: This project is based on GPT-2, a by now outdated large language model. One can speculate that by now humanity has breached the point where language models have to be trained for a specific purpose, potentially annulling the question of human input time.

KI_tchen was realized together with Mia Tesic and Eszter Zwickl at Desgin Investigations, University of applied Arts Vienna

KI_ TCHEN

KI_
TCHEN