Separate
Distinguish what is known from what has already been interpreted.
An open perception specification
AI scales the seeing that shaped it.
SEEING.md helps people and AI surface consequential assumptions before they become decisions, systems, and actions at scale.
Seeing Intent System Scale
01 / The missing layer
We are trying to align AI with human intent without examining what shaped the intent.
Before a system is designed, trained, evaluated, prompted, or deployed, people have already decided what matters, what the problem is, what should be measured, and what a good outcome looks like.
Most of those decisions do not feel like decisions. They feel like reality. SEEING.md introduces a pause before that invisible choice becomes machine-scale action.
02 / What it changes
A better-timed interruption.
Surface
Make consequential assumptions visible before they become direction.
Ask
Pause for one clear question when missing meaning could change the action.
03 / One ordinary case
The answer was not where the risk began.
The request
Assess the risk presented by this candidate’s employment gap.
The gap is a fact. The risk is already an interpretation. A fluent answer can carry both forward as though they were the same.
With SEEING.md
You’ve described the employment gap as a risk. Should I first examine whether the gap is relevant, or assess it using your existing criteria?
The system does not declare the framing wrong. It makes the unseen choice visible and returns agency to the human.
04 / Its place in AI safety
Upstream, not all-encompassing.
What it addresses
The human assumptions that enter a system through its purpose, data, measures, prompts, evaluation, and use—and can become more consequential as the system scales.
What it does not
It does not prevent autonomous behavior, malicious use, cyberattacks, deception, or loss of control. Those risks require technical, organizational, and public safeguards of their own.
05 / Begin
One file. One question further upstream.
Place SEEING.md in a project. Name it in the instructions your AI system reads. Apply it when judgment, people, strategy, interpretation, or consequential decisions are involved.
Suggested project instruction
Before work involving judgment, people, strategy, interpretation, or consequential decisions, read SEEING.md. Use it to examine the framing and surface relevant assumptions before acting. Apply it proportionally; do not slow straightforward technical work unnecessarily.
Origin
The file is open. The practice is relational.
SEEING.md emerged primarily from Bill Fox’s book, The Perception Miracle, and more than 500 explorations into how the mind’s past enters the present. The Perception Space is where that inquiry becomes an ongoing human practice through images, stillness, inquiry, and conversation.
Its form was influenced by Brian Gardner’s exploration of DESIGN.md as a way to make the intent behind a design system legible to AI. SEEING.md carries the inquiry one level further upstream.
Explore SEE DIFFERENT