Perceptron: look inside an AI decision

The decision network

Teal = what you set. Purple = what the model does with it. The weights (amber) are what training would tune.

Your controls

budget score 7

budget weight 3

team score 6

team weight 2

market score 8

market weight 4

approval threshold 100

The whole trick, visible

Inputs are multiplied by weights and summed; the total is compared to a threshold. Nothing else happened in there — and that is the point. The data did not change between runs; the decision did. The decision was never hiding in the data. It lives in the weights (what training tunes) and the threshold (what the business chooses).

The same mechanic at every scale. A spam filter, a credit call, a chatbot choosing its next word — each is stacks of this multiply-add-threshold step. Scale changes how many run and what sets the weights. It does not change what a decision is made of.

System	Decision nodes	Roughly
Your perceptron	1	one manager's judgement call
A small neural network	~100–1,000	a department voting together
Image recognition	millions	a corporation of specialists
Frontier LLMs	hundreds of billions of parameters	weights, not wisdom — tuned, not taught