Ethics of AI

See what AI can
design in space.

We give AI models spatial design tasks and report what they can and can’t do, so creators know what to expect before they build with one.

Explore the evals Our approach
A transparent spatial cube containing the Ethics of AI emblem

What we check when a model designs a space.

We open everything a model builds and report back on three questions.

Can it compose a space?

Scale, layout, light and materials, and whether the scene still holds together as you move around it.

Can it make the space respond?

Which objects a visitor can use, whether each action can be undone, and whether the scene resets cleanly.

Will it work where you ship it?

What runs in an ordinary browser, what needs Safari on Apple Vision Pro, and what has only been checked in simulation.

A model built this furnished garden in code. Switch the lanterns, open a journal or serve tea, then reset the scene and judge how well it did.

Open Living Pavilion
Inside Living Pavilion: hanging lanterns, a book console and a tea table beneath a circular timber canopy, surrounded by trees

OpenAI · GPT-6 Astra · Reasoning: high

A rendered still. Open the artifact to explore its browser interactions.

From results to guidance.

Spatial Web Evals puts each model’s work side by side, labelled with provider, model and reasoning level, so you can compare them before choosing one for your own project.

Explore Spatial Web Evals

  1. Brief

    Give each model the same kind of spatial design task, with clear goals and limits.

  2. Build

    The model produces a working experience you can open, explore and test.

  3. Report

    We record what worked, what broke, and what still needs checking on a device.

Choose a model on evidence.

Every result links to the code, media and notes behind it. Use them to set expectations, pick a model, or write a better brief.

Build an artifact