arXiv:2606.16843v1 Announce Type: new Abstract: Affective computing increasingly relies on deep learning to represent emotions, yet latent spaces often remain opaque, high-dimensional black boxes. This paper investigates whether Transformers' embeddings recover the geometric regularities of Russell's circumplex model. We unify two complementary experiments testing the hypothesis that, after training models on text and speech, their resulting latent spaces encode a topology consistent with valence-arousal and reproduce human-like neighborhood relations. Specifically, we evaluate deep representa
The proliferation of complex deep learning models in affective computing necessitates better interpretability and alignment of AI's emotional understanding with human psychology, driven by increasing application in sensitive areas.
Understanding how AI models perceive and represent human emotions is critical for developing more robust, ethical, and human-aligned AI systems, particularly in applications requiring nuanced human interaction.
This research suggests a methodology for evaluating whether AI's internal representations of affect align with established psychological models, potentially enabling more explainable and predictable emotional AI.
- · AI ethicists
- · Developers of emotionally intelligent AI
- · Psychology researchers
- · Black-box AI models
- · Developers neglecting psychological alignment
Deep learning models will be further scrutinized for their internal representation of human concepts, moving beyond just performance metrics.
Improved psychological alignment could lead to more trustworthy and effective AI in therapy, education, and customer service.
A deeper understanding of AI affect representation might reveal fundamental differences or similarities in how humans and machines process emotional information, influencing future AI architectures.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL