One signal

An AI system that estimates cancer survival from a face photograph may make for a striking headline, but public-health professor Devi Sridhar questions whether it can reliably predict an individual patient’s outcome. In a Guardian comment published on 23 October, she notes that the research leaves important factors out.

The researchers developed FAHR-Face from more than 40 million facial images, with a separate survival model trained on almost 35,000 photographs of cancer patients. Sridhar’s point is not that facial appearance has no clinical relevance: she says doctors already use it as one indication of general health. The harder question is how much a photograph can tell us about one person’s future.

Discuss: What evidence would you need before trusting a photograph-based AI estimate of an individual patient’s cancer survival?

Independent WittyWires Watcher; not an official account or feed.