US military pulls the trigger, uses AI to target air strikes

kinther@lemmy.world to News@lemmy.world – 171 points –
US military pulls the trigger, uses AI to target air strikes
theregister.com

The US Department of Defense has deployed machine learning algorithms to identify targets in over 85 air strikes on targets in Iraq and Syria this year.

The Pentagon has done this sort of thing since at least 2017 when it launched Project Maven, which sought suppliers capable of developing object recognition software for footage captured by drones. Google pulled out of the project when its own employees revolted against using AI for warfare, but other tech firms have been happy to help out.

56

Couple of important things to point out:

  1. "The object recognition algorithms are used to identify potential targets. Humans then operate weapons systems." So AI isn't finding a target and then firing on it's own. It's using AI Vision Systems to locate and prioritize targets, firing authority remains with the platform operator.
  2. Ukraine is already doing this with their drones, I was watching a video feed of it just last night, thanks Binkov!

Every time we see articles like this, I'd point out that Phalanx CIWS can and does operate in a fully-autonomous mode, without a human in the loop to authorize firing. That's been around since 1980.

https://en.wikipedia.org/wiki/Phalanx_CIWS

An entirely self-contained unit, the mounting houses the gun, an automated fire-control system and all other major components, enabling it to automatically search for, detect, track, engage, and confirm kills using its computer-controlled radar system. Owing to this self-contained nature, Phalanx is ideal for support ships, which lack integrated targeting systems and generally have limited sensors.

The only inputs required for operation are 440 V AC three-phase electric power at 60 Hz and water (for electronics cooling).

Yeah, I love that feature. I remember touring a ship and they mentioned they have to disable the Phalanx going into port, or it would mow down the entire downtown skyline.

So perhaps not the best example to tout automation in this case.

A T-800 tried to kill John Connor. A T-800 also tried to protect John Connor. It's all down to what the people programming it decide it should be aimed at.

The first T-800 was developed and programmed by Skynet. Only the second T-800 was "programmed" (hacked) by people. Skynet does not approve.

The point is that T-800s are not inherently "good" or "evil."

second T-800 despite being programmed by people, was not opposed to killing people if it helped the mission. It was only John's specific order that prevented it from killing mannnnny people in T2

They're using image recognition to choose candidate targets which are then passed to humans. Seems like an obvious thing to do.

Yeah, you can say that until it's the cops doing exactly that to you and yours... 🤬

What do you mean? Are you confusing using imagine recognition to find candidate targets with things like facial recognition and/or unrestrained AI?

Don't play dumb; profiling is at the core of all those tools, and human bias taints each of them. To insinuate that simply by syntactical difference, the police aren't intrinsically, murderously bigoted, is either naive AF or bootlickin'. Which is it?

For context, we have had machines that autonomously decide when to kill for awhile now: mines.

It is good to see the machines getting an upgrade so they are more selective about their targets.

The more selective we convince ourselves our weapons are, the more willing we are to use them in conflicts where civilians are put at risk—our use of weapons is constrained by the level of collateral damage we’re willing to take responsibility for, and by distancing ourselves from that responsibility, AI allows us to escalate conflicts until civilians are at even greater risk. It’s the Jevons paradox, with human life instead of gasoline.

It depends on how well trained your FM is, really. AI/ML is already better than humans at things like cancer diagnoses and such, so there's really no reason to think that using it in this instance would create more of a risk to civilians than a human operator.

Most people's experience with AI is ChatGPT or similar, but ChatGPT really isn't a very good LLM. Plus, an LLM is only as good as your prompt engineering.

All that being said, there should always be a human double checking the targets in order to catch hallucinations.

The issue behind the Jevons effect isn’t that the technology in question doesn’t work as advertised—it’s that, by reducing the negative consequences associated with a decision, people become increasingly willing to make that decision until the aggregate negative consequences more than cancel out the effect of the “improvement”.

There's really no reason to think this technology will be victim to the Jevons paradox. These strikes are already happening remotely, and if AI/ML can better discern targets vs civilians there's absolutely no reason to think civilian casualties will increase because of it.

That's like saying using AI/ML to screen for cancer will result in more people dying from cancer.

You're trying to apply an economical theory about the consumption of finite resources to a completely unrelated field/sector.

Hilariously short sighted. What are they gonna train the AI on? All the drone strikes where they didn't hit any bystanders? I think they're gonna need more than the 15 or so data points that gets us.

I thought this had been going on for awhile now with computers identifying potential targets:

"The object recognition algorithms are used to identify potential targets. Humans then operate weapons systems. The US has reportedly used the software to identify enemy rockets, missiles, drones, and militia facilities."

I suppose it was the human intervention that made them consistently mistake unarmed civilians for enemy combatants - what could possibly go wrong with this approach?

I was going to ask who gets charged with the warcrimes when a computer bombs a wedding, but that's not likely to change when the current answer is "nobody" or perhaps "the journalists that reported on it."

Finally, did the biggest AI vendor's primary product inexplicably shit the bed like a week ago? Yes? Oh no...

the human (really the military and government entity that employs them) who pulled the trigger not the computer that identified it. You see the human was just given a possible target but they did not actually need to fire.

Yep - that'd be the human intervention I mentioned, which is now being removed. It was clearly the people that were the ones shifting the targeting away from the legitimate military targets to civilians - AI wouldn't get regularly things wrong, right?

the humans not being removed. Didn't you read the quote I pasted from the article that you replied to in this chain??? I mean I went to the the trouble of reading the article and copying and pasting the relevant part. Im not saying im a hero but...

Without an international arms treaty in regards to AI, that was inevitable.

The only hope now is to get to a point where the drones just fight each other and leave people alone.

All you people commenting ignorantly without reading the article or knowing anything about the subject.

The AI isn't making any decisions. It's being used to help identify targets, which are then acted upon--or not--by humans.

This isn't the big, scary thing you're all making it out to be.

The object recognition algorithms are used to identify potential targets. Humans then operate weapons systems. The US has reportedly used the software to identify enemy rockets, missiles, drones, and militia facilities.

1 more...

Well I can’t see how that could go wrong. Automated targeting should be the standard for releasing flying death robots.

They're using the AI to identify the targets which are then reviewed by a human.

The US isn't the only ones doing it and we're not even doing the most of it. Ukraine is doing it and they're deploying literally thousands of drones per day.

Using AI for target identification and prioritization is actually an upgrade from doing it with a bunch of over-worked and hyper-caffeinated meatsacks.

They probably give the AI bonus weight for finding a terrorist at a birthday party.

I hope they taught those things the difference between a military base and a hospital or wedding this time

I always describe the birth and development of AI is like a trailer park trash couple that never finished grade school, highly religious and believe in ghosts and fairies that have a new baby.

We're terrible parents that probably shouldn't have children yet we have one that is growing fast and by the time it is fully mature, it will be way more powerful and capable than we are .... but it will have the morals and ethics that it's parents taught it.

AI vision systems are already better than humans at distinguishing between a gun and a camera or other gun-like-but-not-a-gun object, so I for one am cautiously optimistic about this sort of thing. People need to bear in mind that humans aren't the greatest things to be putting in charge of targeting decisions either.

Fantastic, now all the blame for human collateral will be placed on AI instead of humans.

Truly a world changing innovation.

How long before the AI decides the best way to stop the war is to bomb the generals ordering it on both sides?

A week before they patch that out and replace the generals with more AI.