How AI and an astronomer’s laptop can bring new galaxies within reach

{"text":[[{"start":7.8,"text":"At the European Space Agency, David O’Ryan and Pablo Gómez have discovered hundreds of new astronomical objects in the vast archive of images transmitted to Earth by the Hubble Space Telescope over the past 36 years."}],[{"start":21.8,"text":"Using a neural network called AnomalyMatch, trained to detect unusual patterns, they found galaxies galore, with peculiar shapes that astronomers had not previously seen through manual inspection of Hubble data."}],[{"start":35.4,"text":"“Of course people look at the exciting new generation of telescopes coming online but this work shows how AI can enhance the scientific return from archival data sets,” Gómez says. “They are treasure troves in which we can still find very exciting things.”"}],[{"start":51.9,"text":"At Oxford university, physicist Héloïse Stevance has led the development of an AI-based Virtual Research Assistant. This has cut by 85 per cent the amount of human “eyeballing” required to distinguish the rare genuine signals caused by supernovae — ultrapowerful explosions of dying stars — among hundreds of alerts received every day from Atlas, a network of five terrestrial telescopes scanning the sky for high-energy events in distant galaxies."}],[{"start":78.8,"text":"“Manual verification would take several hours each day,” says Stevance. “Thanks to our new tool, we can free up scientists’ time for . . . creative problem solving and questioning the nature of our universe. It’s the astrophysical equivalent of having a bot doing your laundry so you can focus on your art!”"}],[{"start":null,"text":"

"}],[{"start":96.9,"text":"At Birmingham university Guy Davies, who studies the behaviour and structure of stars, uses AI “emulators” to help interpret telescope observations. “In much of the work I do our biggest challenge is not so much getting the data but building models of stellar evolution fast enough to be able to figure out what we are seeing in the data,” he says."}],[{"start":116.9,"text":"“We have been successful in taking a neural network and teaching it to simulate complex models of stellar evolution,” Davies adds. “Training can take weeks or months but once it is trained we can evaluate a model in less than a millisecond.”"}],[{"start":131.45000000000002,"text":"His Birmingham colleague Anjali Piette depends on emulators for research into the atmospheres of planets beyond our solar system. “If we see a signal in the spectrum of an exoplanet, which we can attribute to a particular molecule, we want to know why it is there,” she says. “We want to use all the nuanced information in the spectrum to link it to different specific processes that we can model. Getting the ‘why’ involves connecting the models and data.”"}],[{"start":null,"text":""}],[{"start":157.55,"text":"Budget-busting AI bills; transforming pharma; AI and trust in air traffic control; chatbots and mental health; factory robots on the march; rewriting gaming rules; agentic travel agents"}],[{"start":172.25,"text":"These examples — making new discoveries in scientific archives; detecting significant signals in incoming data; and building models to make sense of observations — illustrate some of the ways astronomers are enlisting AI as they look forward to a hugely increased flow of data from a new generation of telescopes in space and on the ground."}],[{"start":193.05,"text":"The greatest torrent of observations will come from the new Vera Rubin Observatory in Chile, which includes an 8.4-metre telescope and a 3bn-pixel digital camera, the largest ever. It will take an image every 30 seconds for 10 years, creating an ultra-wide, high-definition, time-lapse record of the southern sky — “the largest astronomical movie of all time”, as the observatory puts it."}],[{"start":216.70000000000002,"text":"The US National Science Foundation has set up the NSF-Simons AI Institute for the Sky (SKAI Institute) in Chicago to develop new AI tools to detect scientific treasure in Rubin’s output. Astronomers look forward to discovering myriad objects such as asteroids and comets, pulsating stars, supernova explosions and others so novel that they do not yet have a name."}],[{"start":null,"text":""}],[{"start":241.95000000000002,"text":"Supercomputers and intensive processing are not required to exploit AI for astronomy. “The surprising thing is how little data it took,” says Stevance of her Virtual Research Assistant. “With just 15,000 examples and the computing power of my laptop, I could train smart algorithms to do the heavy lifting and automate what used to take a human being hours to do each day . . . With expert guidance, AI can transform astronomical discovery without requiring enormous data sets or computational power.”"}],[{"start":272.90000000000003,"text":"Gómez made a similar comment about the European Space Agency project to explore the Hubble archive with AI. “We have a very efficient algorithm,” he says. “It ran on a single GPU [graphics processing unit] in a few days. We are not on the scale of using 100,000 GPUs burning gigawatts of energy. If we pick the right methods and inputs, like an efficient AI tool with expert annotations, we can still be sustainable.”"}],[{"start":298.75000000000006,"text":"On the whole, astronomers are using more specialised AI tools than the generative AI and large language models from companies such as Anthropic and OpenAI that have caught the public imagination."}],[{"start":310.25000000000006,"text":"“Do you need a general purpose system to answer a specific question in astronomy?” asks Gómez. “In many cases not.”"}],[{"start":318.20000000000005,"text":"While astronomers continue to develop and use specialised tools for their research, generative AI will improve their efficiency, says Amaury Triaud, another expert on exoplanets at Birmingham university. “For example it will help us to design user interfaces for our instruments, work online with colleagues around the world and also carry out intricate work on the optical systems of telescopes — aligning mirrors, focusing and tracking your star.”"}],[{"start":353.40000000000003,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1783089314_8336.mp3"}