postBreeder by Jakob Sitter is an in-browser project powered by a constantly evolving AI system. It takes cues from biological evolution such as viruses that learn to break algorithms or systems in which they operate. Like Facebook algorithms and online politics, certain computer systems are scrutinised as they replicate and imitate biological systems.
In recent years, a revolution in machine learning has rapidly changed the way we receive content. Linear feeds, which used to be the standard, have quickly been replaced by customised feeds curated by an algorithm. Put simply, these algorithms analyse vast amounts of data to find correlations between content and its statistical feedback. This type of ranking system is used on both an individual and a general level. In other words, the content that is expected to generate the most reactions from users is prioritised so that the platform can generate as many clicks as possible. “Good” content is therefore synonymous with what sells better; the strongest inherited information survives.
postBreeder investigates these phenomena, drawing lines to biological evolutionary systems and data valuation. The work runs on a custom model of GPT-2 trained on thousands of Facebook posts. Twice a day, a new generation of “posts” is generated and scored accordingly based on the same dataset of reactions. After each generation, the posts are sorted by a genetic algorithm, with only the half with the most reactions “surviving”. They are then exported, and the model is trained again on these posts so that the next generation is closer to the winning posts of the last round. So over several generations, the texting AI is fed the most successful content and produces new content based on it. This means that its performance gets higher and higher (in terms of likes and reactions) as it strives to create combinations that earn it the highest possible rewards. Visitors are also able to send reactions to the posts, thus altering the order and affect how the training material for the next generation will be.
One of the main “advantages” of the algorithm-curated feed is how it is able to target content in order to maximise profit. In postBreeder, the amount of reactions determines the value of the respective post, effectively turning the generated content into currency. After each generation the “surviving” posts are deployed and minted as NFTs onto the Polygon blockchain and OpenSea where they can be bought. Bridging Web 2.0 with 3.0, the project seeks to turn content into permanent assets on the basis of algorithmic and user labor.
While the language used to describe these processes is becoming more and more similar, the question is whether or for how long they can still be considered separate structures. If Facebook posts (and online information/memes in general) are part of a larger system in which they mutate and multiply to stay alive or extend their lifespan, they could very well be called artificial organisms, much like computer viruses. There is also the question of what this says about us, both creators:inside and keepers:inside such a structure, in which we basically learn to predict ourselves and our environment.
The work postBreeder can be found here.
Jakob Sitter was born in 1998 in Trondheim, Norway, and has been studying at the Hamburg University of Fine Arts in the class of Simon Denny since 2018.