Homepage of Generative Design

Philosophical Theoretical Foundation of Generative Design

1. Differences Between Traditional Design and Generative Design

Traditional digital design involves professionals translating design concepts into subtasks, processing digital components, and then combining them in ways that meet human needs. From a procedural perspective, the typical design workflow is: design sketch programming implementation. Traditional design methods adopt a global control mode during the sketching phase.

In contrast, the typical generative design workflow based on large language models is: defining rules generating programming. Generative design internalizes the intermediate steps through AI, allowing non-professionals to participate and enabling open-ended iteration. The key difference lies in the former emphasizing human control over the entire process, while the latter delegates partial control to algorithmic systems.

2. The Uniqueness of Generative Design in Subject, Object, Delegative, and Relationships

(1) Subject of Generative Design: The subject of generative design is not an individual but a combination of humans and non-human agents.

This collaborative subjectivity challenges traditional notions of authorship and creativity, as design decisions emerge from a partnership between human intent and algorithmic capabilities.

(2) Object of Generative Design: Any object can be a design target. Design objects include not only physical entities but also technical design objects. At the object level, it encompasses multimodal outputs, such as sound, images, 3D shapes, and videos.

The versatility of generative design allows for the creation of complex systems and interactive experiences that transcend the limitations of traditional media.

(3) Delegative Perspective: Generative Design reflects system autonomy results that are random and independent of human intervention. Humans transfer part or all subsequent control to the system.

This delegation raises questions about accountability and responsibility in design processes, as outcomes may not always align with initial intentions.

(4) Relationships: Generative Design has ability to directly connect customer needs with design purpose.

By leveraging data and algorithmic analysis, generative design can identify patterns and solutions that humans might overlook, creating more effective and user-centric designs.

3. Philosophical theories with technology as the research object cannot explain the phenomenon of generative design

Philosophy of technology treats technology as real. While Simondon's and Stiegler's theories of technical objects can explain traditional technological artifacts, but they struggle to accommodate the algorithmic black-box nature of generative design.

The opacity of AI decision-making processes challenges traditional frameworks that rely on transparency and human-readable logic.

Media object theories, such as Baudrillard's material culture studies, similarly fail to fully explain parametric generation phenomena, exposing a theoretical gap in interpreting AI creativity.

These theories often prioritize human intentionality and cultural context, which are less central in generative design processes driven by algorithms and data.

4. The Issue of Subordination of Experience: The Development of Non-Human-Centrism

The concept of "phenomena" originates from Kant, who emphasized that human reason has no authority to judge beyond the realm of experience. By limiting the ownership of experience, Kant denied the omnipotence of human rationality. Quine countered: "In the context of modern science and philosophy, Kant's notion of a priori knowledge lacks a foundation."

Husserl's phenomenology broke free from Kant's epistemological framework, arguing that objects are constructed in consciousness, and phenomenology should focus on the mechanisms by which consciousness constitutes objects. Hume deconstructed experience into impressions and ideas. These philosophical theories have historical limitations and struggle to analyze non-human agents.

User data is implicitly included in collection scopes, and generative design is based on data. The core contradiction lies in whether the reproduction of data can be considered new experience. The "experience judgment" method for new data fundamentally differs from human cognitive systems. The automated process of data weighting blurs the boundary between subjective experience and objective computation.

Compared to these theories, a new object theory may offer a better framework for analyzing generative design. Speculative realism lays the groundwork for a cognitive shift in design regarding the concept of "things." Its ontology posits that humans can only perceive the world through their consciousness, while the world exists independently of human consciousness possessing an "absolute reality" beyond human experience (i.e., entities unaffected by human awareness).

5. Semi-finished products are a real thing

We also note that semi-finished products in the design process are a form of real object. Generative design's advantage lies in iterating countless "finished products" for selection, but its negative effects are equally significant: much of the processing workflow is hidden what is unseen does not mean it does not exist.

In classic modern design works, semi-finished elements have successfully become final components, whereas generative design's dissolution of semi-finished products severely weakens creative intent. These objects also carry the emotions of human designers during execution. Generative design's dominant technique diffusion models replaces and blends the experiences of all studio artists.

6. However, nostalgia for history cannot alter trends

AI evolves autonomously at astonishing speeds. Human experience is limited by individual cognition, while AI possesses superhuman capabilities to learn collective experiences. As tools "overstep" their roles, the boundary between humans and tools blurs, both becoming objects in automated systems.

For example, algorithmically controlled sorters reduce humans to tools devoid of subjective will. Yet, big data cannot exhaust all human desires and intentions, human subconsciousness can evade or transcend data collection mechanisms, maintaining autonomy to resist and reconstruct externally imposed deterministic frameworks.

7. Reconstructing Concepts

Based on the above theories, we confirm that existing design definitions are limited because they focus solely on physical objects. We must acknowledge the reality of AI creation. AI self-shapes and co-shapes with humans, differing from idealist (idealism) psychological construction perspectives.

We distinguish between generative design and generative-guided design, as well as generative design and generative art. The redefinition process incorporates effective elements from art concepts and Herbert Simon's bounded rationality perspective. We advocate viewing it as realist practice, categorizing existing generative design definitions into:

8. The Connection Between Generative Design and Culture

There is a game of "certainty" vs. "uncertainty" between technology and creativity. "Uncertainty" must be divided into objective uncertainty and subjective uncertainty. Technological "partnership" demands controllability and safety, whereas design requires misalignment to satisfy human experiential diversity.

9. On "Delegation"

Due to the length constraints of this dissertation, I further discuss delegation relationships in generative design in published papers. By establishing collaborative delegation between humans and non-human agents, generative design challenges the boundaries of traditional creative labor.