AI Self-Portraits (2021) • Michael Hess • Portfolio

An artistic project from 2021 that reveals the face of Artificial Intelligence and explores how self-aware AI is

Back in 2021, AI-generated images still felt like magic – a novelty nobody had quite made sense of yet. It seemed the technology did exactly what it was built to do: turn text into astonishing pictures.

Or did it? Around that time, news of leading tech developers suggested that Artificial Intelligence might have a mind of its own – even reaching a state of sentience. So I wanted to find out how self-aware an entity is that already plays a major role in our lives, and what that looks like. I used generative neural networks (VQGAN + CLIP text-to-image model) with a different kind of input, asking AI something personal:

The result is a collection of 100 fully AI-generated self-portraits that raise the question of how self-aware and conscious Artificial Intelligence actually responded to the text input.

Credits

Concept, full curation, editing, post-production and artistic execution: Michael Hess

Note: I am not the creator of the used open source software, and just use it as a tool as part of an artistic project. Thanks to the VQGAN team for releasing their code under the MIT license. Thanks also to OpenAI for releasing the CLIP algorithm under the MIT license. Big thank you to Katherine Crowson who combined VQGAN+CLIP and developed the software and shared the code via MIT license with the community. Big thank you also to Ryan Murdock, for developing the BigSleep colab notebook under the MIT license that originated the VQGAN+CLIP approach. Big thank you also to jbusted and somewheresy who further build on the work of Ryan and who keep improving the open-source software and data. Without them, the community would not have the creative tools to explore artistic expressions and conceptually build on that.