AI Trained Only on Landscapes Confirms M87 Black Hole Ring, Settling 2022 Debate
Reported by Tech Times1 source ↓
M87 black hole image confirmation: CSIRO's GenDIReCT AI, trained only on landscape photographs and never on astronomical data, independently recovered the same 42-microarcsecond ring from Event Horizon Telescope data, matching the 2019 EHT measurement within 0.5% and countering…
Key points
- GenDIReCT measured the ring at 42.0 ± 0.2 microarcseconds
- The EHT’s 2019 measurement was 42 ± 3 microarcseconds
- The method uses closure measurements to avoid station-based phase calibration
- The study was published in Nature on November 11, 2026
Sources
Details
The CSIRO researchers applied GenDIReCT to the public Event Horizon Telescope dataset for M87*, the black hole at the center of galaxy Messier 87. The system reconstructed a ring measuring 42.0 ± 0.2 microarcseconds in angular diameter, close to the EHT’s 2019 estimate of 42 ± 3 microarcseconds.
GenDIReCT uses closure phases and amplitudes, measurements that the researchers say cancel errors tied to individual telescope stations. Its diffusion model was trained on natural photographs, including landscapes, animals and objects, rather than radio astronomy data. The Nature paper addresses a 2022 reanalysis that argued the EHT ring could result from sparse data coverage and sampling bias; that challenge had also prompted responses from other researchers.
Written from coverage as of
Images
- Tech Times