Emotion recognition in neurotechnology: A scientific and technical fact check
Abstract
Consumer neurotechnology systems that are claimed to be able to infer human emotions from facial expressions, voice, and physiological signals are increasingly deployed in domains such as marketing, education, human resources, and mental health. As these systems have moved beyond clinical and research settings into consumer markets, their purported capacity to infer users’ internal states has become increasingly relevant to everyday life. More recently, emerging neurotechnologies have extended these claims by suggesting that emotional states can be detected directly from neural data using artificial intelligence, thereby bypassing outward behaviour altogether. These technologies are often promoted on the premise that emotions can reliably be measured objectively, enabling more precise, efficient, and equitable decision-making. However, the scientific and technical foundations underlying these claims remain contested. In this article, we examine the scientific basis and methodological reality underlying claims regarding emotion classification in neurotechnology. We will evaluate the basic assumptions that are embedded in emotion recognition claims: That 1) emotional states can be objectively and reliably inferred from bodily or brain data and 2) emotional states are associated with stable and sufficiently specific behavioural, physiological, or neural signatures that permit reliable inference and 3) recordings from forehead, in-ear devices or consumer EEG in the form of headbands, glasses, or earbuds can deliver technically good-enough signal to inform about the internal emotional states of an individual user. Based on current findings, we are sceptical about claims regarding consumer emotion recognition systems. These systems currently face fundamental conceptual, empirical, and practical limitations. We conclude that the current enthusiasm surrounding AI-based emotion recognition reflects an overconfidence in the objectivity of data-driven systems as well as current technological capabilities. A more realistic appraisal of the technical state of the art and transparent discussions of limitations of these technologies is necessary, particularly when they are deployed in high-stakes contexts or in vulnerable groups. Clarifying these limitations is essential for responsible development, evaluation, and governance of emotion recognition in neurotechnology.
Related
- VAD-EmotionNet: a spatial-spectral attention network for EEG emotion recognition in the 3D affective space
- An Eyewear-Based EEG Interface for Temporal-Region Monitoring and Emotion Recognition
- Disrupting the Right Inferior Frontal Gyrus Increases Social Brain Activity During Emotion Recognition
- ASD-FER: An Annotated Multi-Class Benchmark and Comparative Study for Facial Emotion Recognition in Children with Autism Spectrum Disorder
- EEG emotion recognition via multilevel entropy analysis: The power of entropy of entropy on phase dynamics
BibTeX
@misc{schreiter2026emotion,
title = {{Emotion recognition in neurotechnology: A scientific and technical fact check}},
author = {Marie Luise Schreiter},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/8x7gv_v2},
url = {https://osf.io/preprints/psyarxiv/8x7gv_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Emotion recognition in neurotechnology: A scientific and technical fact check
AU - Marie Luise Schreiter
PY - 2026
DA - 2026/09/19
PB - PsyArXiv
DO - 10.31234/osf.io/8x7gv_v2
UR - https://osf.io/preprints/psyarxiv/8x7gv_v2/
AB - Consumer neurotechnology systems that are claimed to be able to infer human emotions from facial expressions, voice, and physiological signals are increasingly deployed in domains such as marketing, education, human resources, and mental health. As these systems have moved beyond clinical and research settings into consumer markets, their purported capacity to infer users’ internal states has become increasingly relevant to everyday life. More recently, emerging neurotechnologies have extended these claims by suggesting that emotional states can be detected directly from neural data using artificial intelligence, thereby bypassing outward behaviour altogether. These technologies are often promoted on the premise that emotions can reliably be measured objectively, enabling more precise, efficient, and equitable decision-making. However, the scientific and technical foundations underlying these claims remain contested. In this article, we examine the scientific basis and methodological reality underlying claims regarding emotion classification in neurotechnology. We will evaluate the basic assumptions that are embedded in emotion recognition claims: That 1) emotional states can be objectively and reliably inferred from bodily or brain data and 2) emotional states are associated with stable and sufficiently specific behavioural, physiological, or neural signatures that permit reliable inference and 3) recordings from forehead, in-ear devices or consumer EEG in the form of headbands, glasses, or earbuds can deliver technically good-enough signal to inform about the internal emotional states of an individual user. Based on current findings, we are sceptical about claims regarding consumer emotion recognition systems. These systems currently face fundamental conceptual, empirical, and practical limitations. We conclude that the current enthusiasm surrounding AI-based emotion recognition reflects an overconfidence in the objectivity of data-driven systems as well as current technological capabilities. A more realistic appraisal of the technical state of the art and transparent discussions of limitations of these technologies is necessary, particularly when they are deployed in high-stakes contexts or in vulnerable groups. Clarifying these limitations is essential for responsible development, evaluation, and governance of emotion recognition in neurotechnology.
ER -