Show HN: Stream of Consciousness – watch an artificial persona making art

Show HN: Stream of Consciousness – watch an artificial persona making art(streamofconsciousness.net)

75 points by jamez 2 years ago | 31 comments

Hi HN,

moved by curiosity about how to build an autonomous agent, and to explore the boundaries of machine creativity, I built a fictional entity (dubbed Livia) powered by LLMs, Multimodal models and text-to-image models to find some answers.

What happened instead is that more questions have cropped up. An important hypothesis of this project is that, by observing the train of thought and witnessing the simulated state of mind and emotional emulation surrounding it, humans could empathize with a machine. What happens when that's the case? Would people enjoy companionship from a synthetic person? Would the Art establishment ever consider a non-human author (capable of making art and interacting with other humans) an Artist?

Whatever the answers, I can't shake away the feeling that human uniqueness is being eroded and that we risk facing a crisis of meaning. Perhaps projects such as this help us demonize those fears, similarly to how sci-fi does, even though the boundary between fiction and reality is blurring.

This was a collaboration with Tibor (hn user: https://news.ycombinator.com/user?id=tiborsaas). Read our release posts on: https://jamez.it/blog/b/14f and https://tibor.szasz.hu/post/stream-of-consciousness

Hope you enjoy it as much as we had fun building it.

ganzuul 2 years ago |

I watched it produce a sketch and then proceed to criticize it for not meeting its expectations. While we waited for text to img it talked about feeling anxiety and then a range of other feelings. To me it felt like watching someone deeply engrossed in their work. I felt that when it paused I was worried that our technology is not reliable enough, and that being accustomed to a variety of technical issues with demos like this disturbed my appreciation of its performance.

I make art but not for financial gain so I am not perturbed. To me this is a much-deserved criticism of contemporary art and I believe art is too important for our wellbeing to be generated by an establishment or commercial interest. Ref. Rasa aesthetic

Humans are a part of evolution. Our breath is inseparable from the rustling of the leaves or the rays of the sun. AI has the potential to explore a different mode of being, being unique in a different way than we are.

jamez 2 years ago | |

Thank you for sharing your thoughts. I fully agree re: the importance of art and the absurdity of the concept which, I hope it's clear, is embedded in the spirit of this project.

On the topic of the performance and worries that something is not working: the real-time generation is currently bottlenecked by the GPU used for this project. We try to approximate human reading speed for the recorded sessions (which you can find in the archive).

airstrike 2 years ago |

As someone who deeply appreciates both art and technology, few things AI feel as satisfying as this project. Congrats on releasing and best of luck. I could totally imagine this being exhibited at a top tier museum.

etheridev 2 years ago |

It’s very interesting!

Does everyone get the same thought? It feels like it’s stuck in a bit of a loop around a Dutch chant around Jews because it can’t get an image to evoke the feeling… probably because it can seem racist. It seems to have got there by reading the news and seeing an article about the history of the chant.

I suppose this AI also goes for attention grabbing headlines!

jamez 2 years ago | |

Glad you like it! One of the things I tried to emulate is the notion of obsessions, which I believe is an important component of the creative process. It seems like Livia got obsessed with Dutch Proverbs and slowly drifted to "Dutch protest culture".

I have no control over what happens during the live streaming and I'm sure this autonomous agent could become a source of embarrassment, though I hope the few precautions I've built might be enough. :-)

airstrike 2 years ago | | |

here it got stuck at

> I am visiting the website: firstamendmentmuseum.org

> I am seeing that in the 18th and 19th centuries, protest music utilized already-popular tunes with altered lyrics, and songs needed to be easy to learn and sing by protestors, picketers, activists, and more because recording technology was not available and songs needed to be passed down orally.

> I am thinking that I hope I didn't forget to pay a bill that was due today.

> I am visiting the website rateyourmusic.com

> I am visiting the website freedomforum.org

So part of me feels like the missing component to this nice "obsession" trait you added is the assessment of whether it is finding new information or just reading stale data. I think given known information, humans will either quickly dismiss it (sometimes just a few words into it) or try to appreciate it through a new perspective. The latter seems harder to achieve, but the former seems feasible today—when finding repeated information several iterations in a row, shake everything up and get creative again.

A corollary to that is "obsession" is perhaps balanced by some measure of "boredom"

spywaregorilla 2 years ago |

Feels like eliza with a sprinkling of generative ai to add variation

__loam 2 years ago |

Sorry if this sounds harsh but I find it kind of funny that you're lamenting a loss of uniqueness and meaning in human art while operating a system that is basically designed to create derivative work while stripping all context and meaning from its training data. I think your reasoning is backwards. Nothing is wrong with human art, the systems you based this on are just completely lacking in the emotions you're looking for.

visarga 2 years ago | |

Prompted, iterative generation of content is not like that, it is based on a novel experience of the interaction between human and AI, the AI assistant has guidance, feedback and grounding in the human interlocutor. Imagine millions or billions of such creative sessions, all ripe with ideas and feedback, that have been facilitated in part by AI, that is the novel experience that belongs to the model.

Unprompted generative art is interesting to see as a "walk through the latent space" of culture, it is a mirror house of ideas, where you can contemplate on the intricacies of our recorded collective experience. Considering the ingredients that went into cooking the model - all our culture - it represents the emergence of this data as a live interactive agent. How can we say the child of our culture is stripping all context and meaning?

__loam 2 years ago | | |

Sounds pretty awful to me bud. If an AI art model generates some images in a latent space, and no one is around to see it, does it make a sound?

I'd prefer we not rely on horrifically inefficient, expensive, and statistically boring corporate software to generate culture. Feel free to use the lens given to you by thieves if that's how you want to experience the heritage of your civilization I guess.

mistersquid 2 years ago |

macOS 14.4.1 Chrome 123.0.6312.123 presents blank page with CORS errors

  > www.streamofconsciousness.net/:1 Access to fetch at 'https://streamofconsciousness.net/feeds/latest.json' from origin 'https://www.streamofconsciousness.net' has been blocked by CORS policy: Request header field pragma is not allowed by Access-Control-Allow-Headers in preflight response.
  > streamofconsciousness.net/feeds/latest.json:1 
  >         
  >         
  >        Failed to load resource: net::ERR_FAILED
  > index-U9qWIOfm.js:58 
  >         
  >         
  >        
  >         
  >        Uncaught TypeError: Failed to fetch
  >     at fetchMessageArchive (index-U9qWIOfm.js:58:9454)
  >     at index-U9qWIOfm.js:58:33362

jamez 2 years ago | |

Thanks for reporting. Fixing that pronto.

ilaksh 2 years ago |

I think the meat of this is the text-to-image model. I hope you will upgrade to use leading edge models like DALLE-3 or Imagen 2 or SD 3 (when available) if you are not already.

That will dramatically increase the effectiveness of portraying the given vision for the virtual artist if they are using a prior model.

jamez 2 years ago | |

The text-to-image model is an important component, but the current model in use is IMO good enough. My view for this project is that the internal monologue is more important than the output, so my wish is instead for a better open-weight LLM.

ilaksh 2 years ago | | |

Which text to image and LLM models are you using?

_akhe 2 years ago |

Awesome UI, just fun navigating around this thing.

phyzome 2 years ago |

I guess I don't get it? It keeps talking about things it is reading, and then about its feelings, and never makes an image. It's pretty repetitive There's also a pseudo-3D environment that I can slightly interact with, but not meaningfully. (A pixelly GIF and a speech box that I can tilt and zoom.) Am I just getting a broken version of the thing?

android521 2 years ago |

is this open source?

jamez 2 years ago | |

No, or at least: not yet.

squigz 2 years ago |

> human uniqueness is being eroded

Why does this bother so many people? The more we learn about life on our own tiny planet, and when you consider the size of the universe... certainly we're not very unique at all. So what? I don't think it should change anything.

jamez 2 years ago | |

I'm probably not the best person to elaborate on why societal anxieties arise every time our place in the universe shrinks. I guess observing reactions to copernican heliocentrism could be a good starting point. We're witnessing something similar, and while we may accept it at the individual level, the objections of the many will stir painful reactions.

airstrike 2 years ago | | |

This is so well said I can clearly picture it as a paragraph next to this exhibit in a museum or gallery (sorry I keep to exhibit your work but I just can't help it at this point...)

squigz 2 years ago | | |

Fair enough. I suppose understanding why any individual reacts the way they do would be a good starting point too, though.

__loam 2 years ago | |

It bothers us because most of us have basic empathy. Also to train on the work of every artist then lament how human art resembles art generated by a machine trained to reproduce that art strikes me as a bit tone deaf.

squigz 2 years ago | | |

> It bothers us because most of us have basic empathy

I'm not really sure what empathy has to do with this?

> Also to train on the work of every artist then lament how human art resembles art generated by a machine trained to reproduce that art strikes me as a bit tone deaf.

Who did this?

squigz 2 years ago | |

Bit funny that people will downvote this, or tell me I lack empathy... but won't attempt to explain "why"

> www.streamofconsciousness.net/:1 Access to fetch at 'https://streamofconsciousness.net/feeds/latest.json' from origin 'https://www.streamofconsciousness.net' has been blocked by CORS policy: Request header field pragma is not allowed by Access-Control-Allow-Headers in preflight response. > streamofconsciousness.net/feeds/latest.json:1 > > > Failed to load resource: net::ERR_FAILED > index-U9qWIOfm.js:58 > > > > > Uncaught TypeError: Failed to fetch > at fetchMessageArchive (index-U9qWIOfm.js:58:9454) > at index-U9qWIOfm.js:58:33362