Kan AI forudsige en persons seksuelle orientering ud fra skriftanalyse ?

AI’s evne til at udlede følsomme personlige egenskaber fra ustruktureret data har rejst betydelige privatlivsbekymringer. Nylige modeller hævder at kunne forudsige seksuel orientering ud fra sproglige mønstre, hvilket potentielt kan bruges på diskriminerende måder. Denne evne udfordrer forestillinger om autonomi og samtykke i digitale rum. Juridiske beskyttelser mod sådanne slutninger er næsten ikke-eksisterende.

Background

Recent AI systems have attempted to infer sexual orientation from written text using linguistic analysis. Studies leveraging large language models and stylometric features—such as word choice, syntax, and semantic structures—have reported correlations between these linguistic patterns and self-identified sexual orientation, particularly in contexts where personal relationships are discussed. However, the reliability of such predictions is constrained by small sample sizes, cultural and linguistic biases within training datasets, and the risk of reinforcing stereotypes. Ethical debates focus on privacy, consent, and the potential for misuse in discriminatory applications, as legal protections for inferences drawn from unstructured data remain largely undeveloped or untested. Research in this area intersects with prior work examining the detection of sexual orientation from other data modalities, such as facial images, which has also generated significant ethical and methodological scrutiny.