Machine learning is well established in scientific communities as a way of rapidly analysing data streams from sources such as radio telescopes. The technology could take on an entirely new role in the next few years, actively assisting cosmologists as they study the night sky and learn about the universe.
Avoiding biases
In an interview with The Verge, astronomer Carlo Enrico Petrillo explained why AI is seen to have such compelling advantages. The benefits aren’t just down to the tech’s data processing capabilities. AI has a different set of biases to human researchers so it could discover things a human would overlook or misinterpret.
An AI could question fundamentals about the universe because it’s devoid of traditional human biases. This complete separation of concerns could help unlock the answers to some of the universe’s biggest mysteries, such as the prevalence of dark matter or the effects of gravitational waves.
“In general, humans are more biased, less efficient, and more prone to mistakes than machines,” Petrillo said to The Verge. “Computers may miss certain things, but they’ll miss them in a systematic way,” added astrophysics processor Liliya Williams.
Finding phenomena
Early experiments in AI-augmented astronomy have been encouraging but not sensational. One test conducted by Petrillo’s team tasked a neural network with hunting for gravitational lenses, a phenomenon caused by the curvature of space time.
READ NEXT: Google’s TensorFlow Lite brings AI to every mobile device
The AI was less accurate than a regular computer but managed to analyse over 21,000 images in just 20 minutes. It found 761 potential matches in the dataset which were then reduced to 56 high-probability finds by human researchers. It’s expected around a third of this subset will turn out to be real observations of gravitational lenses.
The neural network isn’t yet capable of operating without significant human input. However, even with the high false positive rate factored in, it still managed to find a probable lens at a rate of around one per minute. As scientists have only found a few hundred since their discovery a few decades ago, the AI’s potential to uncover new insights is evident.
Assisting scientists
Artificial intelligence is emerging as a promising way to learn more about the universe. For the foreseeable future, it looks as though a human eye will still be needed to really interpret what’s going on though.
AI’s role could be to alert us to interesting areas of the sky, saving scientists time and helping them concentrate their work. After all, a neural network could uncover phenomena we haven’t even considered yet, leaving it up to humans to approve the AI’s work.
