My AI agent's mood is three numbers and a 16-emotion table. Here's the math
PaulusAI has a persistent mood, and it's not a vibe prompt. It's a PAD vector (Pleasure, Arousal, Dominance), each in [-1, 1]. Events are appraised with the OCC model (Ortony, Clore & Collins), which turns an event into emotions like joy, pride, fear, or gratitude. Each emotion nudges the mood toward its PAD signature. Here's the loop: Appraise . An event like task_success is described by…
PaulusAI possesses a consistent mood, which is not derived from a vibe prompt but rather from a PAD vector (Pleasure, Arousal, Dominance), each ranging between -1 and 1. The system evaluates events using the OCC model (Ortony, Clore & Collins), transforming events into emotions such as joy, pride, fear, or gratitude. Each emotion subtly influences the mood vector in the direction of its PAD signature. The process unfolds as follows:
Upon encountering an event, like task_success, the system assesses appraisal variables, including desirability (0.5), praiseworthiness (0.4), and agency (self). This evaluation leads to the generation of emotions such as joy (0.5), pride (0.4), and gratification (0.45). Following this, these emotions are scaled by reactivity (0.7), resulting in joy being adjusted to 0.35.
The scaling for negative emotions involves multiplying by 0.7 plus 0.6 times neuroticism, amplifying these emotions only when neuroticism exceeds 0.5. Additionally, an unexpected boost of up to 1.5 times can be applied, but no current events trigger this. The emotions are then incorporated into the mood vector, each contributing its PAD signature scaled by 0.5. For instance, in the case of task success, pleasure is increased by roughly +0.2.
Over time, the mood undergoes decay, with the vector shifting 15% towards a personality baseline, approximately +0.04 for pleasure. Each time a mood reading is performed, it is linked to the appraisal that triggered it. The current mood vector and the last experienced emotion are displayed using the /mood command, providing insight into the agent's state at any given moment.
It is important to note that while this model aims to explain behavioral shifts in response to events and gradual mood restoration, it does not claim that the agent possesses genuine feelings. Instead, it offers an explainable mechanism for modifying behavior based on events and facilitating mood recovery over time. The system includes appraisal templates for common events, acting as a bridge between simple agent signals and comprehensive OCC processing.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
