First AI Model of the Universe Knows Science it was Never Taught
A new 3D model of the Universe developed by an international team of researchers is fast, accurate, and makes predictions for which it was never programmed. Creators of the system have no idea how the Deep Density Displacement Model (D3M) is able to simulate changes in dark matter it was never taught to calculate.
Artificial intelligence (AI) is modeling astronomical events from supernova explosions to the formation of exoplanets around alien stars. Now, researchers are harnessing the power of AI to model the Universe as a whole, with surprising results. As expected, the model quickly proved itself to be faster and more accurate than previous systems. However, without human operators training D3M to do so, the AI proved itself capable of describing models of the Universe with varying degrees of dark matter, baffling researchers.
“It’s like teaching image recognition software with lots of pictures of cats and dogs, but then it’s able to recognize elephants. Nobody knows how it does this, and it’s a great mystery to be solved,” stated Shirley Ho, group leader at the Flatiron Institute’s Center for Computational Astrophysics and adjunct professor at Carnegie Mellon University.
Our compliments to the M-5 unit
Of the four fundamental forces of nature (gravity, electromagnetic, strong and weak forces), gravity plays the largest role, by far, in interactions between distant objects. Therefore, researchers developing the D3M system directed its focus to the effects of gravity on the Universe.
Such models can require researchers to run thousands of simulations, and the most accurate models examine how billions of bodies behave over the age of the Universe.
Investigators provided D3M with 8,000 of the most accurate simulations ever created. Neural networks ran calculations on the data, and researchers compared those results to their expected results.
Following this training, D3M ran simulations on the behavior of virtual galaxies within a box measuring 600 million light years per side. While traditional faster models produce data in two to three minutes to complete, and more detailed scenarios take hundreds of hours to produce a result, D3M completed its task in less than 1/30th of a second.
Faster models typically produce an error rate around 9.3 percent when compared to the best known results. In comparison, this new system produced a model with an error rate of a mere 2.8 percent.
Researchers expected the speed and accuracy of their system to exceed that of previous models. What they did not expect was the AI to be able to carry out scenarios in which it had no training.
Modeling the behavior of vast numbers of galaxies requires extraordinary computing power, and detailed simulations can take days to complete. Faster simulations, able to return results in a matter of minutes, lack some of the detail and accuracy possible with older, slower, models.
The D3M model developed scenarios far more accurate than other models of the gravitational effect of gravity on galaxies, such as the second-order perturbation theory (2LPT).
“We can run these simulations in a few milliseconds, while other ‘fast’ simulations take a couple of minutes. Not only that, but we’re much more accurate,” Ho stated.
Computer simulations are utilized to model a wide range of astronomical scenarios, such as how galaxies would react if the amount of dark energy varies over time.
Dark Matter poses a Heavy Question
When we look out at the Universe, all the stars and galaxies we see does not come close to accounting for the effects of gravity. This finding was first discovered by Fritz Zwicky in the early 1930’s, and expanded by Vera Rubin four decades later.
Today, we know that dark matter makes up roughly 27 percent of everything in the Universe, while all ordinary matter represents less than five percent of the total. The remainder — 68 percent — is dark energy, fueling the expansion of the Universe.
The nature of dark matter, however, remains elusive.
“Dark matter may not be made up of the matter we are familiar with at all. The matter that makes up dark matter could different. It may be filled with particles predicted by theory but that scientists have yet to observe,” NASA explains.
As research continues, neural networks will become ever-more valuable to astronomers and astrophysicists seeking to uncover the nature of the Universe. This surprising new finding could assist researchers looking to advance both machine learning and artificial intelligence.
I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.”
― Alan Turing
“We can be an interesting playground for a machine learner to use to see why this model extrapolates so well, why it extrapolates to elephants instead of just recognizing cats and dogs,” Ho remarked.
Dark matter, along with dark energy, remains one of the greatest mysteries of the Universe. Artificial intelligence may prove essential in understanding this mystery — one of the most fundamental questions about the Universe that surrounds us all.
Did you like this article? Subscribe to The Cosmic Companion Newsletter!