Models of Crisis is a series of three generative kinetic objects-poems that externalize and embody the thoughts and feelings of the artist undergoing an existential crisis.
We’re living in the era of models – we’re almost obsessively trying to digitally model every aspect of reality – from the weather, through the stock market to the language and process of thinking and even our emotions. Thus I’ve realised that maybe I can kind of subversively utilise the new technology of Large Language Models (LLM) to construct the model of my own mental struggle to synthesize what’s deeply personal into mechanical abstraction that’s so much easier to deal with.
As my method of work is all about intertwining physical with digital I wasn’t intending to stop on just playing with LLM’s on the screen – I’ve asked myself what if I’ll quite literally build the physical models trough which I’ll be able to convey the synthetic thoughts modeling my current state of mind ? I’ve decided to display words produced by LLM briefed on my feelings and emotions on old school 14-segment alphanumeric displays but also combine the meaning of those words with physical motion of the objects in the vein of “kinetic poetry” in part inspired OuLiPo movement.
And here they are – three physical models (Mood Swing, Train of Thought, Rise & Shine) that are vehicles for expressing the output of the Large Language Model that models state of my mind in crisis.
Mood Swing
A model of constantly fluctuating sentiments
One of the key aspects of any creative practice is to keep an emotional balance between euphoria and misery. To a certain degree it’s only natural to have ups and downs toward the project you are currently working on and your practice in general – though the amplitude of those swings might become problematic.
It’s this constant struggle that I’ve encased in the Mood Swing. The LLM is instructed to create a series (potentially infinite) of emotional antonyms and three “in-between” terms, creating a sort of affective gradient – for example: Euphoria – Joy – Calm – Unease – Misery. Then those word-sets are downloaded onto the microcontroller (raspberry pi pico w) and animated in sync with physically swinging display – thus providing a visceral performance of the emotional quiver.
Train of Thought
A model of an endless stream of anxious and regretful reflections
Amongst multitude of distressful psychological phenomena there is one of the most exhausting – ruminations – defined by American Psychiatric Association as:
[…] repetitive thinking or dwelling on negative feelings and distress and their causes and consequences.
This never ending “train of thought” constantly circulating and regurgitating pointless doubts, regrets and anxieties that literally leads nowhere and keeps me up at night pushed me towards its physical manifestation – so I’ve embedded it in the form of… a hacked train model.
Here the LLM is briefed on the issue of ruminations and asked to produce a virtually infinite range of sentences describing all sorts of doubts and anxieties based on my confessions. These sentences are then downloaded onto the raspberry pi pico w that displays them on a screen divided into four wagons and also controls the speed of the train creating an illusion of new thoughts constantly arriving and departing like unwanted passengers of the mind.
Rise & Shine
A model of infinite cascade of futile reassurances
People struggling with depression all too often hear those generic and futile words of encouragement and all sorts of baffling “motivational quotes” like they could be treated as some form of medicine.
Ironically the LLM thrives in the positive and generic realm, so I was not surprised how easily and flawlessly it constructed all sorts of pseudo-encouraging statements. Here again I receive a potentially unbounded stream of such sentences that are displayed and cheerfully raised up by 8 separate servo sliders in unison only to start glitching and falling apart a few moments later.
The setup is based on local server (written in nodejs) that communicates with LLM (in this case GPT 4o) trough it’s API. The model is briefed on my state of mind and with particular kinetic-physical setup of each of three pieces (like for example swinging motion or limit of 32 characters).
For example in case of Mood Swing the LLM is instructed to create a series (potentially infinite) of emotional antonyms and three “in-between” terms, creating a sort of affective gradient – for example: Euphoria – Joy – Calm – Unease – Misery. Then those word-sets are downloaded as json onto the microcontroller (Raspberry PI Pico W running micropython) and animated in sync with physically swinging display – thus providing a visceral performance of the emotional quiver.
Apart from being based on same microcontroller and codebase (custom engine written from scratch in micropython) each of the pieces include 14-segment alphanumeric displays built from eight 4-char modules giving 32 chars in total.
Mood Swing uses a single servo motor whereas Train of Thought relies on a hacked toy-train locomotive with DC-motor + custom power-sustaining circuit based on super-capacitors. Rise & Shine uses 8 Serial Bus servos driving custom-built linear rails and lots of resin 3d-printing, cnc-machined wood and acrylic, various mechanical and electronic components
Project Page | Jakub Koźniewski | Instagram
