The images above represent the output of a network with two Simplex noise operators in a feedback loop inside TouchDesigner. Simplex noise is an evolved form of Perlin noise, an algorithm, or a procedural texture developed by Ken Perlin in 1983 to give an impression of randomness: “It needed to have a variation that looked random, and yet it needed to be controllable, so it could be used to design various looks.”  It is a pseudo-random noise function that is often used in computer graphics to make visual elements appear more natural, for instance it is often used in video games to generate landscapes and terrain. The fact that it is an algorithm rather than an image-based texture, allows it to be scaled and altered in many ways. Perlin explains noise as a seasoning to make things irregular and more interesting, he also notes that the fact that noise does not repeat makes it useful in a similar way that a paint brush is useful when painting:
You use a particular paint brush because the bristles have a particular statistical quality - because of the size and spacing and stiffness of the bristles. You don't know, or want to know, about the arrangement of each particular bristle. In effect, oil painters used a controlled random process (centuries before John Cage made such a big deal about it) - Ken Perlin 
American composer John Cage began to use chance operations in his compositions during the 1950s, leaving decisions in the score, and later decisions in performance up to chance, for example by employing coin tosses, or consulting the I-Ching (the Chinese book of changes). One of his ambitions with indeterminacy was to place himself in the role of listener and discoverer rather than creator, and to move towards viewing compositions as processes rather than objects. He notes that his own piece Music of Changes (1951), although composed indeterminately, did not afford the performer the ability to perform from her own center, but from the center of the work as written, it was still an object.
As hinted by Perlin, randomness has been used as a compositional tool for centuries, both in the visual- and the sounding arts. In overview Aleatoric music can be described as a music whose composition and/or performance is to a greater or lesser extent undetermined by the composer, and while this can be made to be true for almost all music, the term is usually restricted to music in which the composer has made a deliberate withdrawal of control.  European composers Stockhausen and Boulez also had their go at employing aleatoric techniques during the 50s and 60s, often not as a way of writing the actual music, but as a way to give performers some flexibility to choose between alternative routes, e.g. different pre-written parts, or alternative tempos during performance. A generalization made of early aleatoric music in Europe is that it was a matter of choice rather than chance. However, composer Iannis Xenakis made use of chance by turning to stochastic processes, the laws of probability, to compose most of his music. Probability is the branch of mathematics putting numbers to how likely an event is to occur, adding percentages to the heads and tails of a coin toss for instance. Stochastic music can be seen as a fully composed music, where chance only enters the compositional process, not the performance. It often relies on computer-calculated random distribution, which can be applied to pitches, or any other compositional parameter you would like to distribute over a range. 
Building modular synth patches and TouchDesigner networks is a real-time activity, meaning that calculations of any kind often are left to the computer, or modules inside the network to carry out as it unfolds. Donald Buchla describes that the design choices behind the Buchla modules 265 and 266, or The Source of Uncertainty as they are often reffered to, was to aid the composer:
The predictabilities could be highly defined, or you could have a sequence of totally random numbers. We had voltage control of the randomness and of the rate of change. In this way you could make patterns that were of more interest than patterns that are totally random. — Donald Buchla 
Even though I have not used the Buchla 266, I often use Eurorack modules that were partly inspired by it, such as the Turing Machine by Music Thing Modular. What I do feel is that my compositions are aided by controllable randomness, or pseudo-randomness, it is at the core of all the sounding and all the visual work presented in this exposition. It is sometimes the clay, the core voltage, the raw material. It is sometimes a helping hand, sometimes the decisionmaker. In the context of generative networks, it is a function among other functions. It can be indeterminant, noisy, stochastic, and highly controlled all at the same time. The amount of control can itself be as probabilistic as the rest of the network.
I do relate to Cage's ambition of becoming a listener and seeing a composition as a process rather than an object, and I do think some level of indeterminacy helps. I believe it echoes in Brian Eno’s thoughts about a generative approach, to let go of some control to the system is a rethinking of one's own position as a creator: “You stop thinking of yourself as me, the controller, you the audience, and you start thinking of all of us as the audience, all of us as people enjoying the garden together, gardener included.” Noise can be a seasoning, and it can be it, the seed that grows into a work of art. It can be both, at the same time.
 John Cage, "Composition as a Process: Indeterminacy," in Audio Culture: Readings in Modern Music, ed. Christoph Cox et al. (New York: Bloomsbury, 2017), 259.