Generative Algorithms

Pengembangan arsitektur etnik Nusantara yang mengkini dalam proses kreatif desain menjadi cukup sulit untuk dilakukan. Perubahan bentuk namun tetap harus mempertahankan karakteristik acuannya adalah kendala yang sering dihadapi seorang perancang. Permodelan parametrik memungkinkan eksplorasi varian bentuk arsitektur etnik Nusantara dengan menggunakan logika dalam komputasi. Sebuah acuan bangunan akan dimodelkan dengan menyusun algoritma setiap konfigurasi elemen bangunan beserta parameternya. Nilai parameter elemen bangunan dapat diubah untuk membuat bentuk-bentuk generatif. Eksplorasi dalam model parametrik dilakukan dalam program Grasshopper. Setiap perubahan pada parameter memberikan kesempatan perancang memilih beragam output bentuk. Termasuk di dalamnya  varian yang mudah dikenali atau menjauh dari acuannya.
Kata kunci: algoritma , arsitektur, bentuk, Nusantara, parametrik.

In this thesis, I seek to explore alternative ways of thinking about sound material and different strategies to treat this material in time, namely by the usage of generative algorithms and a network-inspired data flow. For this purpose, I have designed and implemented a framework that is meant to facilitate this exploration in the context of computer music composition; it is therefore a dissertation concerned with aspects of relationships, inter- dependencies and interactions of sound objects. A detailed account of these processes will be covered along with musical concerns that motivated the creation of the framework. The chosen environment for the implementation of the system is SuperCollider. The thesis will cover as well examples of own compositions made with the framework and a concise analysis of potentials and consequences of these strategies.

The paper discusses the implementation of a multi-agent system as an integral component of a hybrid, digital-physical architectural environment. It contributes to the existing practice-based architectural research in two ways: 1) by describing an innovative integration of a multi-agent system for surface patterning; and 2) by discussing this integration in terms of emergent materiality. This case-study demonstrates suggestive creative approaches and observes in the field the operation of a concept that promises to be useful for future analysis, research and design."

In this work we will analyze some recent proposals about how the result of the application of a generative algorithm is recognized as a unit for the purposes of further computations, and how each correlates, either explicitly or implicitly, with a model of the mind. We will propose, in line with previous works, that a mixed approach to mental computations (including both Markovian and non-linear dependencies) can prove useful at both theoretical and empirical levels when stated explicitly. We will argue that phrase markers include both kinds of dependencies between terminals and non-terminals: labeling requirements in a theory (which, in turn, interact with other conditions on phrase structure, binarity being the most notable) give away the conception a certain theory has about the (uniform or not) functioning of the mind.

The new class of neural network architectures introduced in 2014, called generative adversarial networks (GANs), can be used to mimic any distribution in any domain: images, music, speech, prose etc. We propose a pipeline SDCGAN to produce paintings from extremely small datasets using a combination of VAE, DCGAN and image translation frameworks.

The last decades witnessed an exponential proliferation of AI music composition programs. The hard-coded algorithmic composition systems of the outset are progressively giving way to a more advanced use of neural networks and Deep Learning software, with a consequent
increase of the sophistication and quality of the music produced. The Mozart and Lady Gaga of the future are a set of silicon chips: Jukedeck, Flow Machines, Aiva and other programs are gaining more and more followers in many music platforms, raising a growing enthusiasm and
consent among their fans.
While the latest developments in the field of AI music are matter of excitement among both listeners and researchers, they also raise less practical and more philosophically oriented questions. In this paper I discuss the impact of AI on one of the key topics in the philosophy of
art: the nature of musical works. The question I will address is the following: “Can a computer create a musical work?”. In attempting to provide an answer, further questions about the creativity and intentionality exhibited by AI will emerge.
In the first section of the paper I identify a necessary requirement for musical works: being created by an act of intentional creativity. I then argue that the evaluation of something as creative or intentional is influenced by our subjective judgement. As a consequence, it seems impossible to objectively assess whether a computer can create a musical work. What we measure when we provide an answer to this question, in fact, are not the computer’s accomplishments but instead our subjective evaluation of them (Boden and Edmonds 2009).
In the second part of the paper I suggest an alternative definition of minimal creativity (CREATIVITYm) which focuses on the autonomy needed by a system to produce a creative output. I will claim that software like Jukedeck or Flowmachines, where the human presence is still essential for performing the task of composing music, do not create musical works. They at best can be considered an extension of the programmer’s or user’s mind (Clarks and Chalmers 1998). This point will lead to the related discussion about the boundaries of the self and the relative autonomy of humans and machines in respect to each other.
In the last part of the paper I will claim that the use of Generative Adversarial Networks (GANs) in software for music composition may grant AI a sufficient level of autonomy for deeming it able to create musical works (Yang 2017). In addressing the inherent difference of GANs from other kinds of software previously used in algorithmic composition, I will borrow some insights from the discussion on integrated information theory (Tononi 2008). Indeed, I will discuss the quantity of integrated information, or Φ (‘phi’), that a system for music generation which makes use of GANs has and what does this tell us about its levels of
creativity.