Experiments with AI

My professional background lies in Data Science and AI. While creativity is present in this field, it is often expressed within abstract, technical, and efficiency-driven frameworks. Entering the Master Design programme created a space to shift from problem-solving towards exploratory making, and from optimisation towards experience. This shift strongly informed the methods and design decisions that followed.

Throughout the research, making functioned as a primary mode of inquiry. Early experiments focused on the development of a visual language that could operate independently of written text. Through iterative sketching and testing, I arrived at a deliberately naïve yet functional drawing style. Central to this approach was the art of omission: determining what visual information is strictly necessary to communicate a concept, and what can be left out without compromising meaning. This principle became a recurring design heuristic across the project.

At the same time, practical constraints—limited time, limited access to participants, and the need for repeatability—necessitated methodological choices that balanced depth with feasibility. Here, my technical background did not disappear but was reframed as a supporting resource within a design-led research process.

Visualisation, AI, and the Limits of Substitution

As part of the exploration into non-textual communication, I investigated whether generative AI could support the translation of written language into visual representations. This phase positioned AI as an experimental tool within the research, rather than as a proposed solution.

The findings from these experiments were instructive and led to an explicit design decision to move away from AI-based visual translation:

• Current generative systems are not capable of converting complex or procedural text into visual representations in a way that is both unambiguous and reliable at scale.
• Language is inherently contextual and situated. Any attempt to replace text with images involves interpretation. Conducting such interpretation responsibly requires intensive co-creation with low-literate participants and results in highly contextual, bespoke outcomes.

From a methodological perspective, this reinforced a critical insight: exclusion caused by text-based systems cannot be resolved through one-to-one substitution of modalities. The failure of AI to “solve” this problem became valuable data in itself, redirecting the research towards experiential and systemic approaches rather than representational fixes.

Experiments with AI Audio as Medium and Research Instrument

Subsequent experiments shifted towards audio as an alternative to written language. Initial tests using recorded speech demonstrated the potential of audio to convey information while simultaneously shaping tempo, tension, and dependency. However, they also revealed practical challenges: inconsistency in delivery, variable audio quality, and the large volume of material required to support repeated gameplay.

The introduction of a specialised text-to-speech AI tool addressed these challenges and constituted a deliberate methodological choice. AI was used here not to generate content autonomously, but to stabilise and scale a carefully authored script. This enabled consistency in tone, pacing, and clarity, which proved essential for the functioning of the game as a research instrument.

An unexpected outcome of this process was that audio made procedural complexity tangible in a new way. Instructions that appeared acceptable in written form became conspicuously long and convoluted when heard. This necessitated a process of rewriting and compression, not to simplify the underlying systems, but to make their complexity playable within a game context.

Crucially, even in compressed form, the audio instructions are still experienced by players as lengthy and demanding. This effect is not treated as a flaw, but as an intentional design outcome. The friction generated through audio contributes directly to the experiential understanding of bureaucratic complexity and literacy-based exclusion.

AI and the Narrative of the Research

While writing remains an important component of this research, the project explicitly seeks to explore alternatives to text-dominant knowledge dissemination. In line with this aim, I am currently experimenting with the use of generative AI to translate the narrative of the research into an audio-based podcast format.

This experiment is ongoing and remains provisional. The prompts used to guide the AI require further iteration and refinement. However, the initial results suggest that AI can function as a mediating tool for reframing and retelling research narratives across media. Rather than replacing authorship, AI operates here as a reflective surface—revealing which aspects of the research translate, which resist translation, and which require reinterpretation.

Within the broader methodological framework of this project, AI is thus positioned neither as a neutral tool nor as a technological solution, but as an active participant in the design process—one that exposes limits, introduces frictions, and informs subsequent design decisions.