Takafumi Horiuchi — Design Engineer

Design Philosophy

AI can generate a wide range of content within a given set of conditions. Yet what can be produced—and how broad those possibilities can be—is determined by the design space provided to the AI. I believe that designing not only individual outputs, but the “space of possible content” that AI explores, will become central to design in the age of AI.

This space is shaped by factors such as objectives, values, priorities, constraints, prohibitions, and evaluation criteria. By defining what must be preserved, how much may change, and what kinds of variation are acceptable, designers give AI a bounded space in which to explore multiple possibilities and turn them into concrete content.

I call this approach “constraint-driven design”. Here, constraints are not intended to suppress creativity. They are a way to exclude undesirable possibilities while creating room for AI to explore safely. Rather than specifying a single answer, the designer shapes the structure of the space from which desirable answers can emerge.

The starting point is an understanding of people. Insights into cognition, perception, the body, and behaviour are translated into conditions and rules that AI can work with. Drawing on my experience across computer science, design engineering, business consulting, and cognitive and behavioural science, my work centres on using constraints to connect human understanding with AI’s generative capabilities.

As AI becomes more capable of making decisions and taking action autonomously, the designer’s role expands from refining a single answer produced by AI to shaping the structure that determines what kinds of answers can emerge. Design must address not only what AI creates, but also the space it explores and the conditions under which it creates.

Recent Works

• Personal Fit UI - Personalising Interfaces Through Constraint-Driven Design (August 2026) Generating interfaces tailored to cognitive, sensory, and physical characteristics within the boundaries of scientific evidence and design constraints. A car-navigation prototype was used to examine both the potential and the limitations of the approach.
• My Own Pokémon Generation (July 2026) A new Pokémon species is generated from the player's words during play, then brought into the game as an encounter they can catch. The prototype focuses not only on what AI creates, but on the rules that preserve balance and coherence.

Academic Works

• Master's Thesis on Human–Robot Interaction Design (July 2021) The operator's field of view becomes a condition that determines where a robot may move. This research proposed and implemented a safety mechanism that incorporates human attention into collision prevention without requiring objects to be registered in advance.
• Bachelor's Thesis on Constraint Programming (June 2019) The simulator identifies conditions that can never become true again and removes them from the search. By reducing the possibilities it must consider without changing the result, the method roughly halved execution time in the evaluation model.