Stream of consciousness
A friend of mine, Giacomo, reached out to me to see if I’d be interested in building an LLM-based project at the end of 2023. It was the last weeks of my sabbatical and the timing was perfect. His vision was to build a fully autonomous AI artist agent capable of creating artworks independently of our control, acting on its own.
I was quite happy to join forces as it was in the sweet spot for my interests in technology: LLMs, autonomous agents, generative art, and interface design are all close to my heart. After many months of interaction and prototypes, we finally managed to launch it on April 1st.
Feel free to visit: https://streamofconsciousness.net/ or check Instagram. Press “listen” on the site to hear more in audio format.
But what is it really? First and foremost, it’s an art project to demonstrate to the world what’s possible when a bunch of machine learning algorithms are welded together to pursue a single goal. We wanted to explore what it’s like to set free a system and let it draw inspiration from the world in order to create relevant art. All this without our intervention. The agent can read news, browse social media, draw conclusions and follow obsessions.
What can you do with it? You might ask. It’s not your typical AI product you can chat with, it won’t make you more productive and it won’t fix your marketing automation. We didn’t build this to make money. We aimed to build something you can enjoy, observe and dissect.
The core idea is to initiate conversations and challenge the public perception of what’s possible with stock AI models. Eventually, many agents will emerge in the future for better or worse, and it will be hard to tell what was created by humans and what’s the result of a flesh-and-blood person.
I have no doubt that people will keep finding ways to differentiate themselves from machines - they always have - but we should all be aware that as these systems evolve, this line will be really fuzzy.
We put a lot of effort into making the stream of consciousness experience resemble a person, so we turned to anthropomorphism and gave it a feminine pronoun and name to the agent: Livia. It not only helps to think about building features but it’s also a good way to increase the coherence of the parts.
Although she works without our intervention it’s not totally accurate that there’s no human input in the loop. You can interact with her over social media platforms like Mastodon and Bluesky. She regularly posts there every day, looks for comments and sometimes replies if it aligns with her values. These values are also a key factor when she considers the generated images whether they are worthy to be posted, and even the prompts for the image generators can go through many iterations before even getting processed.
Is there’s consciousness under the hood of Livia? I want to be clear that what you and me have in our skulls is very different and unique compared to bits flipping on a server. In this sense, no she does not possess human consciousness. What you can see is an iterative evolution of states governed by external stimulus and internal memory, purely digital with a some randomness involved. When you see human qualities in her messages, it’s a reflection of what humans created in written text since forever.
Livia is a complex digital person, like a character in a video game. We hope you’ll find this project interesting enough to make you think about the future, art and consciousness.
In the end, she wants to entertain us with content which is currently buried in the dark corners of endless latent spaces, encoded in huge AI models.
