Seeing the Invisible

What the eye misses

Some things are invisible and we see them anyway. Affect is made below thought, not by it ¹, and the eye only sees what the mind can name ², so whatever we have no name for arrives unlabelled and we call it intuition. That is measurable. Show someone emotional information they never consciously perceive and their accuracy, confidence and speed all rise, and they cannot say why ³. It is in the writing too. Every message carries the state of the person who sent it, whether they meant it or not ⁴.

The machine as a second eye

Media are extensions of our senses ⁵. A model reading affect from language extends the eye, and the data is what it renders visible. It works when the model is right. GPT-4 beat expert human coders on short messages across eleven countries ⁶, and against 33 human annotators over 19,200 annotations the models were the more consistent readers ⁷.

Sources

1. Selected Principles of Pankseppian Affective Neuroscience. Davis and Montag, Frontiers in Neuroscience, 2019.
2. The theory of constructed emotion: an active inference account of interoception and categorization. Barrett, Social Cognitive and Affective Neuroscience, 2017. Emotion becomes emotion at the moment it is categorised.
3. Measuring Intuition. Lufityanto, Donkin and Pearson, Psychological Science, 2016.
4. Schulz von Thun, F. Miteinander reden 1: Störungen und Klärungen. Rowohlt, 1981. The four-sides model: every message carries four things at once.
5. McLuhan, M. Understanding Media: The Extensions of Man. McGraw-Hill, 1964.
6. Large Language Models Outperform Expert Coders and Supervised Classifiers at Annotating Political Social Media Messages. Törnberg, Social Science Computer Review, 2024.
7. Comparing large language models and human annotators in latent content analysis. Bojić et al., Scientific Reports, 2025.