We Make Stupid

Note / 001

Intelligence

The noetic
baseline.

The minimum evidence an intelligent system needs before it can form a sufficiently reliable model of what is happening and choose a useful next action.

The idea

Intelligence is always working from partial evidence. A person says something. The environment contains more. Memory adds history. Tools can reveal facts that are not immediately visible.

At some point, those signals become enough to support a coherent model of the situation.

That point is the noetic baseline. Below it, the system does not know enough. At it, the system knows enough to proceed. Beyond it, asking for more can become unnecessary friction.

Where evidence comes from

Signal

Instruction.

What the person explicitly asks for.

Signal

Environment.

What the system can observe about the current situation.

Signal

Memory.

What remains true or relevant from what happened before.

Signal

Tools.

What the system can verify by looking something up or taking a measurement.

The threshold moves

The baseline is not a fixed amount of context. It changes with the action.

A reversible, low-cost action can tolerate more uncertainty. An irreversible or high-consequence action should require a stronger model, better evidence, or explicit confirmation.

The more an action matters, the higher the noetic baseline should rise.

Three states

01

Below.

The model is too weak. Ask, observe, verify, or stop.

02

At.

There is enough evidence to choose a useful next action.

03

Above.

More questions add friction faster than they add understanding.

What it is not

It is not the shortest possible prompt.

It is not mind reading.

It is not a confidence score by itself.

And it is not permission to guess.

It is a threshold for justified understanding.

Design implication

Traditional software makes the person translate intent into the exact structure the system expects: fields, menus, filters, steps, and repeated explanations.

An intelligent system can absorb more of that translation work by observing, remembering, verifying, and reasoning before it asks another question.

The person should provide what only the person knows. The system should gather the rest when it can.

Application / Harv

In Harv, the noetic baseline is treated as a moving threshold rather than a target prompt length. The system should keep collecting signal until it has enough to act responsibly, then stop making the person do work the machine can do itself.

If the evidence is weak, ask. If the evidence is sufficient, act. If the evidence is strong, do not ask for ceremony.

Notes / Subscribe

Get the next note.

No schedule. No content calendar. We send one when there is something worth understanding.