Source : Perth Now news
A small team of CSIRO researchers from Perth have recreated one of the greatest scientific achievements this century by imaging a supermassive black hole, and they did it with a bespoke AI model they trained themselves on pictures of cats and dogs.
This improbable technique was crucial to establishing total independence from the team of hundreds of scientists who famously captured the first-ever image of a black hole in 2019 using the Event Horizon Telescope, a global network of synchronised radio observatories.
By applying their new technique to the same EHT data, the Perth team validated the landmark image of the black hole at the centre of galaxy M87, 55 million light years away, confirming the ring shape that had been under fire from rival theories in recent years.
As far as technical challenges go, it was equivalent to standing in your backyard and attempting to get a decent photo of a doughnut sitting on the surface of the moon.
The findings have been published on Thursday in the prestigious journal, Nature.
CSIRO astronomer and lead author Dr Samuel Lai called the original EHT image of black hole M87* “iconic”, but being iconic is not the same as being correct.
As Albert Einstein said, “No amount of experimentation can ever prove me right; a single experiment can prove me wrong”.
That was almost the case with M87*.
“There was one independent group that questioned the validity of the ring structure, and they suggested that an alternative interpretation is also equally valid,” Dr Lai said.
“So we developed a new data product, which has a fair few advantages compared to what is traditionally used in radio astronomy, and we also developed (GenDIReCT), a bespoke imaging algorithm.”
This new data product didn’t involve feeding EHT data into Claude of ChatGPT, but rather building an AI model from scratch, and training it on terrestrial images such as animals and automobiles rather than existing astronomical imagery.
CSIRO astronomer and project leader Dr Nithyanandan Thyagarajan said this decision removed any chance of bias in the model’s results.
“We intentionally avoided exposing it to any kind of previous assumptions or expectations about what the black hole shape or size should look like, and wanted to see what came out of it completely blindly,” Dr Thyagarajan said.
EHT collaboration director Laurent Loinard couldn’t be happier the 2019 image was confirmed.
“It is extremely gratifying to see that GenDIReCT, an elegant, entirely independent, method, recovers the ring structure of M87* across several years,” Professor Loinard said.
The algorithm developed by the team, which also included CSIRO team leader Dr Ivy Wong and CSIRO Principal Research Scientist Dr Foivos Diakogiannis, accounted for atmospheric errors and those produced by individual observatories, which had historically left room for data sets to be misinterpreted.
“Studies like this strengthen our confidence in what we have learned about black holes, and enrich the toolkit available to the whole community,” Professor Loinard said.
While machine-learning has been integral to science long before any of us heard of generative AI, the future potential of GenDIReCT won’t just be felt in the toolkits of astronomers studying black holes.
“This powerful imaging algorithm could be applied even beyond astronomy, wherever such combination of interferometric measurements are involved,” Dr Thyagarajan said.
“I can immediately think of a couple of areas, for example, seismography, seismic imaging, as well as in medical imaging with ultrasound or MRI.
Einstein can rest easy knowing GenDIReCT is at least one experiment that didn’t prove him wrong.
“The ring-like structure is one of Einstein’s general relativistic predictions for how a black hole would distort space-time around it,” Dr Thyagarajan said.
“I think it says a lot to the strength and the resilience of his theory, because one experiment after another over decades have yet to disprove what what his theory has predicted. It’s pretty amazing.”
