Black Box: How AI will Amplify the Best and Worst of Humanity
summary
A hospital deploys a machine learning model from Epic that scores cardiac patients with probabilities derived from past records — examples like 87% or 79% are shown to clinicians — and it measurably reduces cardiac codes. Clinicians, however, collapse these decimals into a few buckets because human brains handle only five probabilities (99, 100, 1, 0, 50/50). The team must calibrate percentages constantly because too low is ignored and too high triggers emergency responses.
AI-written summary · about
related talks
39mHuman Values in Clinical Algorithms, Old and New
47mBreaking Black-Box AI
37mAI in medicine: where do we stand?
35mResponsible implementation of Machine Learning models for life-and-death decision making
09mHow AI is changing the healthcare system
18mMedical AI Beyond Prediction: From Normative Learning to Agentic Healthcare AI
23mEngineering AI-Powered Predictive Healthcare Systems
35mML: Beyond confusion matrix, model performance analysis