🤖 AI learns to recognize schizophrenia from speech
On September 9, Knowable Magazine reported on two approaches: speech acoustics (88 features — volume, pauses, vowels, intonation) distinguished patients from healthy individuals with 86.2% accuracy (AUC 0.92) on a sample of 142+142, while semantic drift analysis in transcripts showed 87% versus 68% for doctors without AI.
🌍 In the US, the diagnosis is made on average 1.5 years after the first symptoms, and psychiatrists' assessments differ by 30–50%. Speech markers could standardize diagnosis, but a 2026 review warns: samples are small, and age, stress, medication, and non-native English mimic the markers.
👤 This is classic ML to assist the doctor, not replace them. Clinical trials of the semantic method are promised by 2030.
Source 1: https://knowablemagazine.org/content/article/mind/2026/how-ai-can-listen-for-signs-of-schizophrenia