Emotions as Abstract Evaluation Criteria in Biological and Artificial Intelligences

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Neuroscience

Background:

• Both biological and artificial intelligences (AIs) face the challenge of selecting and prioritizing goals.
• Nature's solution involves a continuously adjusted categorical weighting mechanism, experienced as emotions.
• Phylogenetic observations show a correlation between cognitive capabilities, intelligence levels, and the number of emotional states.

Purpose of the Study:

• To explore emotions as a fundamental mechanism for valuing behavioral options in biological and artificial systems.
• To propose and discuss a functional framework that mimics emotions for artificial intelligence.
• To investigate how emotions can guide goal pursuit and time allocation in AI.

Main Methods:

• Reviewing nature's solution to the time allocation problem using emotional weighting.
• Proposing a framework based on time allocation via emotional stationarity (TAES).
• Implementing emotions as abstract criteria (e.g., satisfaction, challenge, boredom) to evaluate activities.

Main Results:

• Emotions serve as a generic mechanism for attributing values to behavioral options, especially when not specified at birth.
• The proposed TAES framework compares an agent's emotional timeline with its defined 'character' (preferred emotional states).
• Optimizing task selection frequency allows agents to align their experiences with their character, leading to stationary emotion statistics.

Conclusions:

• Emotions are crucial for understanding the mind and essential for intelligent agents to make decisions.
• The TAES framework provides a functional model for implementing emotion-like mechanisms in AI for goal-directed behavior.
• Aligning experienced emotions with an agent's character through optimized task selection is a key outcome.