Good news: Neural network says 11 asteroids thought to be harmless may hit Earth. Bad news: They are not due to arrive for hundreds of years

And also, crucial point, the software may be wrong and we'll never be released by these angels of death

A neural network has identified eleven asteroids, so far thought to be benign, that may eventually come close enough to hit Earth.

These 11 space rocks, each measuring more than 100 metres across, are listed in a NASA database as, for now at least, non-hazardous objects.

However, AI software – developed by researchers at Leiden University in the Netherlands, and dubbed the Hazardous Object Identifier – has singled out the eleven, predicting they could come within 0.05 astronomical units (7.5 million kilometres, 4.7 million miles) of terra firma. By that definition, these asteroids should be labelled as potentially hazardous objects.

Although that sounds a little alarming, they are still unlikely to slam into our fragile world, since 0.05 astronomical units is a wide margin. Just because something's declared a hazardous object, be it by a human or machine-learning code, that doesn't mean it's going to pummel us to death... sadly.

To train the neural network, the researchers simulated hundreds of thousands of asteroid collisions with Earth, recording the space boulders' motion over time using the ALICE supercomputer cluster at the university.

Data describing each asteroid's properties and trajectories were then fed into the machine-learning software so that it could discern and learn to spot common features that hint that an asteroid may crash into Earth. In other words, the software was taught to pick up on patterns in the data common among asteroids that would hit Earth and identify them as such. Each asteroid was encoded as a vector of variables defining the semi-major axis, eccentricity, and inclination of its orbit, as well as its mean speed, and angular momentum.

"If you rewind the clock, you will see the well-known asteroids land again on Earth,” Simon Portegies Zwart, co-author of the paper and an astronomy professor at Leiden, said last week. “This way you can make a library of the orbits of asteroids that landed on Earth.”


Astroboffins peeved as SpaceX's Starlink sats block meteor spotting – and could make us miss a killer asteroid


The eggheads also trained the model on data describing hundreds of thousands of real asteroids, taken from NASA's dastcom5 dataset. Unlike the simulated asteroids, these rocks are not expected to crash into Earth, and should help the software determine benign objects.

A paper published [preprint PDF] in the Astronomy & Astrophysics journal this month stated the model looked through a test dataset and identified 13,258 asteroids, previously thought to be benign, as potentially hazardous objects.

"A total of 13,258 asteroids identified by HOI [the AI software] as known impactors are not listed by NASA as potentially hazardous objects," the paper said. However, that's not very useful because these rocks are not expected to visit Earth for thousands if not tens of thousands of years: their paths could change, humankind could have wiped itself out by then anyway, and so on.

More practically, the researchers tested the neural network on nearly 2,000 asteroids listed by NASA as potentially hazardous, and the code managed to correctly identify 90.99 per cent of them without having been explicitly trained on this set of rocks. That also means it missed one in ten that humans previously identified.

What's also intriguing is that it detected eleven non-hazardous asteroids in our Solar System that will come closer than ten times the distance between the Earth and the Moon within this millennium. If any of them were to strike our planet, that would be not pleasant.

"For perspective, Tunguska object which flattened 2,000 square kilometers of forest in Siberia was estimated to have a diameter of between 50 and 80 meters," according to the paper. The identified asteroids – 2005 RV24, 2008 UV99, 2011 BU10, 2011 HH1, 2011 WC44, 2013 AG76, 2014 GL35, 2014 TW57, 2014 WD365, 2017 DQ36, and 2017 JE3 – are bigger than Tunguska.

Don't panic, though: the machine-learning software reckoned the prangs may occur between the years 2131 and 2923.

The academics noted NASA probably listed these 11 asteroids as non-hazardous because their observed orbits are so uncertain, and calculating the probabilities of these hitting Earth is therefore difficult.

NASA's Sentry system, which monitors asteroids to determine whether they are hazardous or not, can only look reliably as far ahead as the next 100 years, bar the special cases of 29075 (1950 DA) and 101955 Bennu (1999 RQ36), which would explain why the 11 are listed as non-hazardous. They are beyond the grasp of NASA's 100-years prediction algorithms, it seems, but not artificial intelligence, perhaps.

“We generated a short list of network identified potential impactors which NASA does not label as potentially hazardous objects, mainly because the observed orbital elements are so uncertain that NASA’s Monte Carlo approach to determine their Earth-striking probability fails,” the paper concluded. ®

Broader topics

Other stories you might like

  • City-killing asteroid won't hit Earth in 2052 after all
    ESA ruins our day with some bad news

    An asteroid predicted to hit Earth in 2052 has, for now, been removed from the European Space Agency's list of rocks to be worried about.

    Asteroid 2021 QM1 was described by ESA as "the riskiest asteroid known to humankind," at least among asteroids discovered in the past year. QM1 was spotted in August 2021 by Arizona-based Mount Lemmon observatory, and additional observations only made its path appear more threatening.

    "We could see its future paths around the Sun, and in 2052 it could come dangerously close to Earth. The more the asteroid was observed, the greater that risk became," said ESA Head of Planetary Defense Richard Moissl. 

    Continue reading
  • Is computer vision the cure for school shootings? Likely not
    Gun-detecting AI outfits want to help while root causes need tackling

    Comment More than 250 mass shootings have occurred in the US so far this year, and AI advocates think they have the solution. Not gun control, but better tech, unsurprisingly.

    Machine-learning biz Kogniz announced on Tuesday it was adding a ready-to-deploy gun detection model to its computer-vision platform. The system, we're told, can detect guns seen by security cameras and send notifications to those at risk, notifying police, locking down buildings, and performing other security tasks. 

    In addition to spotting firearms, Kogniz uses its other computer-vision modules to notice unusual behavior, such as children sprinting down hallways or someone climbing in through a window, which could indicate an active shooter.

    Continue reading
  • Cerebras sets record for 'largest AI model' on a single chip
    Plus: Yandex releases 100-billion-parameter language model for free, and more

    In brief US hardware startup Cerebras claims to have trained the largest AI model on a single device powered by the world's largest Wafer Scale Engine 2 chip the size of a plate.

    "Using the Cerebras Software Platform (CSoft), our customers can easily train state-of-the-art GPT language models (such as GPT-3 and GPT-J) with up to 20 billion parameters on a single CS-2 system," the company claimed this week. "Running on a single CS-2, these models take minutes to set up and users can quickly move between models with just a few keystrokes."

    The CS-2 packs a whopping 850,000 cores, and has 40GB of on-chip memory capable of reaching 20 PB/sec memory bandwidth. The specs on other types of AI accelerators and GPUs pale in comparison, meaning machine learning engineers have to train huge AI models with billions of parameters across more servers.

    Continue reading

Biting the hand that feeds IT © 1998–2022