A generic self-learning emotional framework for machines.

Area of Science:

• Artificial Intelligence
• Cognitive Neuroscience
• Computational Psychology

Background:

• Intelligent beings use emotions for behavioral modulation and evolutionary advantage.
• Current AI lacks a first-principles theory for emotion elicitation, leading to generalized models.
• Existing AI emotion models are often case-specific and lack broad applicability.

Purpose of the Study:

• To propose a self-learning AI framework for emotion elicitation from raw observations.
• To define emotions as perceived temporal patterns in crucial environmental values.
• To develop AI agents that exhibit and identify natural emotions.

Main Methods:

• Developed a fully self-learning emotional framework for AI agents.
• Trained an artificial neural network on unlabeled agent experiences.
• Validated the framework using human observer ratings on pleasure-arousal-dominance dimensions.

Main Results:

• The AI framework successfully learned and identified eight basic emotional patterns.
• Identified emotions were situationally coherent and reproduced natural emotional dynamics.
• Human observers showed high statistical agreement and alignment with experimental psychology.

Conclusions:

• The proposed framework provides a general, cross-disciplinary approach to artificial emotions grounded in Reinforcement Learning.
• This research may pave the way for more human-like emotional AI.
• The findings suggest emotions can be understood as perceived temporal patterns in environmental interactions.