Between transitions
Artificial Intelligence is in the constant process of becoming, with no determined place or purpose. If AI is far from human intelligence can it be treated as another type of intelligence, finding its ways in monitoring data and giving away the information? When training on images and datasets, who is doing the seeing, is it even a seeing, what is happening behind that? Could we teach machines to see better? And by playing with these narratives and technologies of machine vision could we ourselves learn to see better?
When we talk about AI we have a certain vision towards it, our lack of terminology to describe its aesthetic and not quite understanding how it functions put limitations on us. By creating practices we might redirect our attention towards the strange, the construction of possibilities and new language.
To delve into these questions I trained an RNN model on a certain dataset to discover how the model will interpret given information and how it could be envisioned on a human level. RNN is a type of neural network that runs on a particular set of text data with a next-word prediction. What happens is that since the data is usually not in a large amount and never can be learned perfectly, it becomes creative in some way, giving very random word sequences.
The model is trained on Jorge Luis Borges’ book “The Book of Imaginary Beings”. It is a catalogue of cultural and folklore gathering of different fantastical living forms which is never complete and should be read as “shifting patterns of kaleidoscope”. Each of them is a fruit of human imagination that reflects human nature and surpasses its agency.
RNN data training with ASCII art
When the model was trained I gave it one prompt:
Describe a creature of the 21st century
It gave me following results with second one being more reliance on trained data:
With this inofrmation i asked ChatGPT to create me an imaginary being in ASCII art, and I got the following results: