Document details

Artificial intelligence (AI) is transforming EEGinterpretation faster than training is adapting to definewhat competent interpretation now requires. Errors arisebefore the first abnormal waveform appears, in whattrainees encounter, how they examine evidence, and howreadily a signal becomes a diagnostic label. That path isalready shortening: a deep neural network has achievedexpert- level detection of interictal epileptiform dischargesin selected settings.1 As automated recognition growsmore capable, human verification becomes more, notless, consequential. Protecting patients requires readers toreconstruct, defend, and revise the reasoning from signalto conclusion. What makes verification credible? Presenceis insufficient unless readers can explain why an output isplausible and recognize when it is not.