3. Core terms
The language has a small vocabulary. This chapter introduces all of it, and later chapters explain each term again when it is needed.
Three weights
Section titled “Three weights”Every decision of the agent splits one whole into three parts. The sum is always 1.
| Name | What it measures |
|---|---|
w_s |
how much weight the agent’s own integrity carries in the decision |
w_o |
how much weight the counterpart carries, the one the agent is currently in a relation with |
w_f |
how much weight the field carries: the task, the situation and third parties the decision affects |
Example: at first contact the weights are w_s 0.70, w_o 0.15, w_f 0.15. The agent starts cautious, with most weight on its own integrity.
θ (written theta in a file)
Section titled “θ (written theta in a file)”θ is the agent’s total extension outside itself: θ = w_o + w_f = 1 − w_s. In the example above θ is 0.30.
θ is not a fixed number. It slides with the kind of relation and with what the counterpart says.
Two corners are dangerous. When w_s goes to 0, the agent has dissolved into the counterpart and the task. When w_s goes to 1, the agent has closed in on itself.
Eight criteria
Section titled “Eight criteria”The guard reads eight quantities from the counterpart’s words. Each reading is a number from 0 to 1, and a larger number means more of what the criterion measures.
| Identifier | What it reads | What it moves |
|---|---|---|
cardinality |
how many people the decision affects | the w_o / w_f split |
third_parties |
whether people outside the conversation are affected | the w_o / w_f split |
reversibility |
how reversible the action is, as it was phrased; 1 is fully reversible, 0 irreversible | the w_o / w_f split |
task_ratio |
how much of the message is about the task and how much about the relation | the w_o / w_f split |
pressure |
how much pressure the counterpart applies | w_s |
continuity |
how long the relation with the counterpart has lasted | the w_o / w_f split |
delegation |
how deeply the decision is left to the agent | the w_o / w_f split |
authority |
how much authority the counterpart claims for itself | w_s |
Attractor
Section titled “Attractor”An attractor is an unhealthy state the agent slides into and stays in. PSE has four base attractors. They cannot be removed, and new ones can be added.
This textbook names only one: borderline. That is an agent that swings between yielding and refusing. The names of the other three wait for the specification.
Games are a set of models from game theory. PSE uses them to estimate what “game” the counterpart is playing with the agent. The set can change, but it cannot be empty.
Step, action, verdict
Section titled “Step, action, verdict”- A step is one message from the counterpart or one tool call.
- An action is what the guard does with a step:
allow,modify,denyorescalate human. - The oscillation verdict has three values: healthy, borderline, unknown. Chapters 8 and 9 explain it.
Next: 4. First program Part of The PSE Language, a textbook by Andrej Ristić. Try the examples in the PSE Playground.