Here's an experimental text. For a while now, I have been creating pop music in collaboration with machine learning systems in a project called The Organizing Committee
. The lyrics are generated, mostly, from a version of the OpenAI GPT-2 language model. However, using RunwayML and Inferkit, I've conditioned the language model to draw from an archive of texts including cybernetics, critical and feminist theory, and Situationist writings.
I thought I might share some of the outputs that are too dense for lyrics.
The result is not a cut up or collage, though some lines may be "direct" from the sources because my training library is relatively small. Instead, it is a reassemblage of words and sentences that emerges from the Neural Network where pairings are weighted and associations between them loosely held. In other words, it's an attempt at recontextualizing the writing based on what we ask the system to do.
Or, more simply, the system is just sorta vibin' on Deleuze.
In this case, it was to autocomplete the starting phrase: "In cybernetic forests..." I find reading them in abstract ways to be quite productive, while reading them literally is maddening. What do you all think? I'm curious to hear about your individual reactions to "uncanny" writing by contemporary neural nets. Text below.
### In cybernetic forests, contagion attacks upon machines are mitigated or eliminated. Rather, enclosure within the periphery of the forest reduces network possibilities, artificially limits the variety available beyond the periphery, and offers the computer a kind of imprisonment. We have seen that the semiotic regime of the forest does not give the computer freedom of action without conditions. The world supplies conditions for a certain amount of freedom. But it could not have been created without imposing the affects of decoded language upon the computer, and preventing it from understanding the world in which it is placed.
The computer implicitly demonstrates the limits of the world in which it is operating. This is clear not only for the generic language, but also for a way of thinking about concepts such as computation.
Where the trees are data, the forest is data processing. Computers as creative writing devices are structured and manipulated by a labour of enunciation. Drawing on the work of Jacques Riviere, who started his own development of the symbolic-combinatorial approach towards reading and writing in France in the mid-1950s, called 'assemblage sans communautaire', and Debord, who described ontological, semantic and linguistic forces of enunciation, we see computation as shaping language in “indistinguishable forms,” and asking “which ones are to be absorbed and which are not?”
A tree is partially structured into “graphs,” so that each leaf points in certain directions, in relation to its nearest neighbours. Next, the tree forms a graph within itself (and with neighbouring generations); so, once again, the traits in question are in correspondence with one another. (It could also be said that the same process is going on to form two other kinds of graphs — graphs of resources, and graphs of spatiotemporal relations. These are in turn part of the tree. A tree -- a living, breathing creature -- is also an infographic.) “Competition and diversity” mean “confusion” in the “forest,” but not “in the trees.”