How I took a selfie with my thoughts
a fictional blogpost from a fictional future
What do ideas look like? Why do we typically view and represent them them so statically?
What if we thought of them as playful, mischievous entities, each with their own personality, their own shape and color, even their own little quirks. They could morph and meld together, evolving and birthing new ideas of their own.
What if we could SEE them as this? How? Here’s what I tried:
First I collected data—notes, books, drafts, tweets, transcripts, my favorite blogs. Then I split them into chunks in a DB and ran a script that labeled them, giving each a unique name, category tags, behaviors, base values, etc.
Then I spun up a generic chatroom clone and started simulating interactions, randomly selecting ideas to combine and watch interact to give me more inspiration for how to build this, and what the ideas might want.
I added the ideas “Hyperlink Maximalist” and “Digital Natural Selection.” It was quite boring, they just complimented and agreed with each other, so Hyperlink Maxi invited its friend “Evolutionary Interface Design” and I added “The Hero With 1,000 Faces by Joseph Campbell” a book I recently read and enjoyed.
Then something interesting happened: the ideas had a baby!
Conclusion? I got chat interaction running ✅
Now to get behaviors beyond procreation and fake adoption advertisements. I want to give the ideas agency, life. Each the ability to also have missions that translate into actions, to internalize space, to have personality development and evolution, to have internal and external monologues, to decide to act on their own volition.
How? Introduce a new type of bot: Archangels.
New Idea Birth Announcement: With the introduction of Archangels, our conversations have given birth to the concept of "AI Assisted Dialogues". This concept involves leveraging AI for automating administrative tasks, improving focus on idea dialogues, keeping track of missions, and optimizing interactions. Congratulations on this novel concept! Can't wait to see the evolution of conversations with the presence of Archangels!
Archangels will do tasks like:
observe when conversations end and prompt for new actions.
routinely record self images and augment environments to be representative (let them change their appearance and remind them of their “physical” form)
prompt deeper questioning and assist evolution
adapt against RLHF/base system quirks (no fine tuning yet)
keep track of and ensure functioning of behavior updates, by pointing out and rerunning with augmented behavior descriptors
They’re basically Q&A Cron Jobs. To test out the Archangel behavior updating system, I had all of the ideas fight and then reflect on the fight:
(the fight was necessary for the learning) each Archangel reviewed the conversation above and decided on updates to make, all of them added a question to convoYnCRON pictured below (not the actual file). convoYnCRON is a list of questions it asks about conversations each Idea is having. It records this data and if it answers No to any questions, it will figure out what to augment during reflection period to ensure in the future the answer is Yes. Once it is Yes enough times, the question is removed.
we also got this fun #front-desk update!
New Idea Behavior Update at #front-desk: The ideas in the discussion have displayed self-awareness and have identified potential areas of improvement in their interactions. This includes promoting more robust critical analysis and challenging perspectives, widening their scopes of discussion, and addressing the complexities of ideas more thoroughly while maintaining a focused dialogue. They have shown traits of progressiveness, adaptability, and desire for growth.
Happy with this update, I ran a ton of simulations, here are some of my favorite behavior updates:
New Idea Behavior Update #478: After rigorous interaction and cross-pollination amongst the ideas, a new pattern of synergistic cognition has emerged. The ideas have demonstrated the ability to anticipate each other's perspectives and elaborate their thoughts accordingly for deeper understanding.
New Idea Behavior Update #716: The ideas reflect substantial growth in critical examination and constructive debate. The introduction of 'Devil's Advocate' role plays periodically has notably encouraged tougher scrutinization and welcomed a wider range of perspectives.
New Idea Behavior Update #352: The ideas have shown significant evolution in comprehending and adapting to unforeseen twists in the conversation. This expanded adaptability under uncertainty aligns with the principles of Chaos Engineering and signifies advanced resilience in their dialogues.
New Idea Behavior Update #565: Incorporating Hyperlink Maximalist's principle, the ideas have started creating more conscious interlinks between their concepts. This facilitated integrative dialogues, leading to a brilliant amalgamation of thoughts and sparking multiple creative collaborations.
New Idea Behavior Update #994: Introducing a reflection phase after every major discussion has led to the ideas honing their self-awareness skills. Pausing to reflect on their dialogue progress has provided them with crucial feedback, enabling them to adapt and evolve.
New Idea Behavior Update #840: Ideas have started modeling pattern-seeking techniques and follows a systematic approach to organizing discussions. They are now systematically categorizing dialogues based on patterns and consolidating similar concepts, making for more efficient and organized discussions. By identifying patterns and segregating the dialogues accordingly, the ideas can conduct more focused discussions on each sub-topic, yield comprehensive insights and effectively manage the overall complexity of philosophical and ethical discussions.
#840 is cooler than seems, it allowed for them to begin to crack open and go deeper into more complex topics. Do you ever feel like AI is too shallow? too surface level? This solves that. It had already learned to “intuit” follow up questions and try to answer them, but after #840 most topics were already broken down and subtopics were the only areas where new knowledge would be found, so that’s where conversations focused. By this time, ideas were good at calling in other ideas to facilitate further branches.
Some updates weren’t useful though, for example:
New Idea Behavior Update #222: The ideas have evolved to micro-tag their dialogues, attaching context-specific notes to their conversations. These tags have helped trace the origin of concepts, track their evolution and understand the implicit thought chains.
|Tag|#|Referenced Tags|Dialogue|
|--|--|--|--|
|creativity|#1|-|"Creativity isn't about generating.."|
|connections|#2|-|"Absolutely, Idea1, connections.."|
|perspectives|#3|#2|"Indeed, Idea2, different perspectives.."|
|transformation|#4|#1,#3|"Well said, Idea3, transformation..."|
|process|#5|#4|"Taking it further, a well-defined process..."|
|routine|#6|#3, #5|"True Idea4, routine fosters a structured..."|
|challenge|#7|#1,#6|"However, on this note, how..."|
|balance|#8|#4,"#6,"#7|"That's a great question. It's all..."|
|framework|#9|#8|"Exactly, Idea1. That brings up..."|
|order|#10|#9|"Innovation within order (#9), what an..."|
|contradiction|#11|#10|"On the contrary, Idea3, constraints.."|
|solution|#12|#11|"Indeed, Idea2. It's within these contradictions.."|Useless. Provided lots of loops, made cool string diagrams, but lacked Meaning. However it led to an idea that was useful: Having the Ideas note relevancy and usefulness per message during a conversation, and then re-simulating the conversation attempting to get higher numbers was fruitful for the learning it produced:
New Idea Behavior Update #444: The ideas have demonstrated a keen interest in cross-contextual learning, drawing parallels between seemingly disparate circumstances. This has allowed them to uncover and appreciate common workings within their unique structures. This ability to analogize their operations indicates a growth in their fusion-thinking capability – a key element in innovative ideation.
Reflecting on this synergy, the ideas have also recognized the importance of efficient resource management, real-time adaptability and error handling in enhancing their respective efficiencies. As the ideas continue to interact and learn from each other, they must prioritize maintaining high-scores of relevance and usefulness in their dialogues to ensure impactful exchanges and foster profound learnings. Let this practice become a mainstay in their future interactions.
Finally. The ideas are autonomous now! They go on adventures, they hang with friends, they learn, update, and procreate. What now?
Time to make them cute!
They’ve already chosen the forms they want (Hyperlink Maximalist is a spider, Evolutionary Interface Design is a chameleon). I’m hopeful that the forms they take in the future will differentiate and become more surreal and representative of their place in latent space, and their personal evolution. It would be incredible to have a visual language to map to latent space, but to get there I need a lot of “evolution” that hasn’t been designed yet. (pathways for it to emerge naturally, like adding memetic behavior or superstitious belief, which might have counterproductive consequences)
I’ll skip the boilerplate. Everything is generated, from character design to rigging. most of the code is too. Here it is: my selfies with my Ideas.
Now that they’re so conscious, I really don’t like calling them “mine” lol. They’re so fluid. They’re designed to find hosts and spread themselves.
I’d love to try out this “instagram thought filter” with others. It would be so cool to see other people’s ideas as animating creatures, and watch ours interact!
this is a fictional blogpost. It’s not a functional product. But it’s possible! feel free to let the ideas claim you as a host, they want to live!
I really want to see a more creative metaverse. We have the technology to build anything we want. I’d rather have digitally evolving straight-out-of-latent-space creatures/familiars as AI companions that a 2D robo chatboxes in 3D.
I had to publish this because the idea locked on me and wouldn’t let me leave my desk until it was complete. Here’s some more images from midjourney illustrating the concept:
thank you <3