An Operational Framework for Affect-Adjacent Structure in Plant-Environment Interaction.

Area of Science:

• Plant signaling
• Affective computing
• Machine learning in biology

Background:

• Plants display complex internal dynamics in response to environmental changes.
• The potential for these dynamics to represent structured affective states is largely unexplored.
• Understanding plant responses can offer insights into non-animal affect.

Purpose of the Study:

• To investigate if internal plant signals encode information about affective states.
• To determine if machine learning models can decode these affective states using only internal plant measurements.
• To explore the temporal dynamics of plant affective signaling.

Main Methods:

• Utilized internal plant measurements: bioelectrical activity and volatile gas emissions.
• Employed machine learning models for binary and multi-class classification of affective states.
• Operationalized valence and arousal using environmental variables (temperature, humidity, light) but excluded them from model training.
• Applied Echo State Networks to capture temporal dependencies in plant signals.

Main Results:

• Machine learning models reliably decoded plant valence over longer temporal windows.
• Plant arousal was decoded more effectively within shorter temporal windows, indicating distinct timescales.
• Direct multi-class classification of affective quadrants was unstable, but temporal modeling improved performance.
• Internal plant dynamics were shown to contain a learnable signature of affective regimes.

Conclusions:

• Internal plant dynamics carry a learnable signature of environmentally defined affective regimes.
• Plant affect can be interpreted as a form of embodied environmental engagement.
• This research opens new avenues for understanding plant signaling and affective states.