The US Military Desperately Wants To Weaponize AI (technologyreview.com)

Artificial intelligence is a transformative technology, and US generals already see it as the next big weapon in their arsenal. From a report: War-machine learning: Michael Griffin, Undersecretary of Defense for Research and Engineering, signaled how keen the military is to make use of AI at the Future of War 2018 conference held in Washington, DC, yesterday. Saber rattling: "There might be an artificial intelligence arms race, but we're not yet in it," Griffin said. In reference to China and Russia, he added, "I think our adversaries -- and they are our adversaries -- understand very well the possible future utility of machine learning, and I think it's time we did as well."

Comments

Oh come on

[Anonymous Coward]

There is not a thing the US military does not want to weaponize. It's what they do.

The rationale is just there to clothe the thing a bit, the gist is always the same. "Can we weaponize this?" And yes, yes they did, including boring old business machinery mounted in trucks for battlefield use back in the 50s.

Doesn't make the headline less superfluous though. This state of affairs is a given.

Re:Oh come on

[Rei]

What do you do when weapons without AI go south? Same thing. Innocent people die. The point of AI would be to decrease the odds of innocent people dying, not increase it.

First off, all "skynet" stuff aside, we're not talking about "generalized AI". We're talking about stuff like image recognition. Not learning neural nets, just pretrained ones. Secondly, this would be most desirable serving as a backup, not primary, control mechanism. One of the big problems faced by drones is jamming. If a drone gets jammed and you're lucky, it has to return to base autonomously without completing its mission. If you're unlucky, bad things happen. For example, US-made surveilance drones given to Ukraine met with bad reviews because the Russians - oh, I'm sorry, "rebels" - were using advanced electronic warfare systems to down them, and then downloading the images from memory to see where the Ukrainian positions were.

You want a drone that does whatever you want tell it to so long as you can maintain communications with it, but if someone jams your communications, you want it to carry out whatever mission you assigned to it as best as it can.

Beyond drones, AI would be very useful in missiles / guided artillery / etc. You find a target, you lock onto the target, and you want to ensure that the missile hits the target - whether the target is a tank, a plane, a group of soldiers, or whatnot. Your "targets", in turn, deploy a variety of countermeasures to try and make sure that the thing you shot gets confused and can't find it. Simple human-created algorithms describing how to determine what's your target and what isn't can only take you so far; having a neural net which is trained to recognize various types of targets - and various things that aren't targets - can make the weapon much more effective, while reducing the risk of it impacting a non-target object.

Then there's just general surveillance / monitoring. The more visual data you collect, the more manpower it takes to analyze it. But if you train neural nets to recognize weapons systems - from camouflaged tanks to guns in people's hands (trained *against* non-gun items which people might carry) - it can narrow down the imagery that humans need to look at when determining what should be targeted and what shouldn't.

To reiterate: in none of this are we talking about "generalized AI". We're not even talking about "learning AI". We're talking about "AI pre-trained to specific tasks". Which are things we already use in our everyday lives.

Re:stupid headline

[AutodidactLabrat]

Given the amount of money, yes it is desperately seeking a conscience free killing machine