Unreasonable Cube

A physical AI that maps challenges to multidimensional perspectives through tangible interaction. 

By replacing the static screen with a voice-driven, gear-based interface, the system transforms the act of problem solving into a tactile dialogue. Users speak their challenges directly to the object, which then projects an analytical framework across its six faces, allowing for the physical manipulation of abstract thought through rotation and speech.


YearMedia
Team
Special Thanks 2026
Machine Learning, Physical Computing, Speculative Design
Tim Qi, Haoyang Li & Shu Ou
Anthropic (Claude Feature)



MULTIDIMENSIONAL REASONING


When facing a complex problem, the standard digital interface often reinforces linear thinking. Most AI chat paradigms encourage a one directional thread message after message that can trap the user in a tunnel vision groove. Unreasonable Cube is a physical AI interface designed to break this pattern by making problem solving multidimensional. By mapping a challenge across the six faces of a custom built cube, users can physically hold, rotate, and rebalance different perspectives before committing to a direction of thought.

Drawing inspiration from the Rubik’s Cube a cultural shorthand for complexity and transformation the device translates a spoken challenge into six distinct lenses: relevance, emotional resonance, relationships, blind spots, systemic forces, and the unspoken. As the user rotates the cube’s faces, a Rubik’s adjacency weight model reconfigures how these perspectives interact. The embedded screens pulse with high contrast color, reflecting the shifting balance of the user's inquiry. This generative loop allows for a speculative glimpse of a solution, shaped not just by a single prompt, but by the physical configuration of the object.




TECHNICAL ARCHITECTURE


The system utilizes a tightly coupled hardware software loop to maintain real time physical to digital synchronicity:

  • Hardware Integration: The cube is a custom mechanical assembly housing six ESP32 microcontrollers and six 4" ST7796 TFT displays. To solve the mechanical decoupling problem, the team developed a gear driven system where outer gear rings drive internal rotary encoders, allowing users to rotate the faces without stressing the internal screens or wiring.

  • Computational Logic: A Python based weight engine coordinates the interaction. The AI layer leverages OpenAI Whisper for robust speech to text with tuned RMS thresholds for noisy environments, and Claude 3.5 Sonnet to generate both the initial six perspectives and the final speculative narrative.

  • Interaction Design: Rather than hardcoded mappings, the cube uses the full weight vector influenced by every rotation and press to shape follow up questions. This creates a feedback loop where movement informs music like variations in logic, and the machine, in turn, inspires further thought.




By leveraging large language models to interpret open ended challenges through a constrained physical form, the project explores how tangible interfaces can create a deeper sense of agency and investment in AI assisted creative processes.


Unreasonable Cube was featured by Anthropic as a pioneering example of tangible AI. By moving the interaction from a screen into a physical object, the project demonstrates how LLMs can be used to navigate complex, non-linear human challenges through tactile feedback.

@claude.ai