According To Artificial Intelligence, The Movie Armageddon Was Not Too Far Off


Aunt Spray/Shutterstock

If you’re over the age of 28, chances are you’ve seen the 1998 disaster film Armageddon. In Michael Bay’s cinematic masterpiece, ordinary Joes played by Bruce Willis, Ben Affleck, and company are sent into space with a mission to set up and detonate a nuclear weapon on a giant asteroid that is on track to destroy the planet.

Twenty years later, an advanced, self-learning artificial intelligence algorithm has decided that the controversial director’s penchant for solving problems with explosions really is the best way to deal with an incoming celestial object.


Obviously, we have not yet needed to defend the Earth from an extinction-level impact, and experts at NASA’s Jet Propulsion Laboratory keep careful tabs on any large objects that could come within striking distance in the future. According to their reports, nothing is likely to threaten Earth within the next several hundred years.

Yet because the stakes are quite high, it might be nice to be prepared.


To that end, a team of astrophysicists led by Erika R. Nesvold developed and trained a computer program to analyze the current options for neutralizing an asteroid or cometThe charmingly named “Deflector Selector” looked at 6 million hypothetical scenarios of an object approaching Earth. In each instance, the program compared how effectively the object's trajectory could be altered using one of three options: 1) a projectile called a kinetic impactor that is crashed into the object, 2) a large craft called a gravity tractor that gets close enough to the object to disrupt its path, and 3) a nuclear weapon to blow it to pieces.


According to New Scientist, the simulation training took 40 hours on a cluster of 100 computers. This process and the algorithm’s results are detailed in an article in Earth and Planetary Astrophysics.

Machine learning algorithms differ from traditional AI because it can solve problems or complete tasks without human input after being trained on sample data sets. The way the program teaches itself closely mirrors how our brains do the same task – just much, much faster.

The freshly educated Deflector Selector then looked at three types of incoming objects: near-Earth asteroids, comets, and debris clusters. It determined that nuclear weapons could successfully prevent impact in about 50 percent of situations. Kinetic impactors and gravity tractors, on the other hand, are less likely to work. How soon in advance the objects could be detected were factored into the calculations, and the scenarios were based on the assumption that a powerful load-bearing rocket such as the Boeing Delta 4 Heavy could carry the defender of choice into space. (At the time of the research, the SpaceX Falcon Heavy had not yet been tested.)

Despite its usefulness, the authors emphasize that the Deflector Selector was generated to help agencies decide which of the prospective methods is worth exploring further.


“Developing every proposed technology is currently prohibitively expensive, so determining now which technologies are most likely to be effective would allow us to prioritize a subset of proposed deflection technologies for funding and development,” they write.

“We are absolutely not advocating putting the algorithm in charge of asteroid defense,” Nesvold told New Scientist.


[H/T: New Scientist]

  • tag
  • comet,

  • asteroid,

  • extinction,

  • impact,

  • artificial intelligence,

  • physics,

  • rocket,

  • astrophysics,

  • armageddon,

  • nuclear weapon,

  • bomb,

  • machine learning