AI at the bedside and at the telescope

A deep-learning model trained on the historical observations of the Dark Energy Survey now schedules observing time on the Blanco telescope: it decides where to point, and adapts in real time as conditions change.

Two observing runs were completed that way this spring and summer. What is at stake is not discovery but telescope time — the most contested resource in astronomy.

↑ Back to the summary

An MIT study published in Nature Medicine measures what AI assistance does to skin-disease diagnosis. Assistance improves accuracy, for non-experts and clinicians alike. The explanations are the problem.

Non-experts defer to the model, especially when it justifies itself in natural language: their confidence rises, including on wrong answers. Clinicians score best with the prediction alone, with no explanation attached.

↑ Back to the summary

Work reported this week describes new routes, opened by AI, to diagnosing and treating Crohn's disease earlier — a condition whose diagnosis often arrives years after the first symptoms.

This is a signal, not a summary: the methods and the cohorts are in the original publication, and the link below is the point of this brief.

↑ Back to the summary