[mathinpens]

my only comment is that the opportunity to stop weaponized AI is long long long gone. the barriers to entry for modern machine learning methods are just so low and the military benefits are there. Near peer states are going to do this and we frankly cannot stop them. the train has left the station...

sidenote: my impression is the next block of the small diameter bomb has a semi-autonomous feature where it can glide over a target area without a pre-sighted target-use some computer vision-recognize a tank- adjust its trajectory to hit the tank. all without a man-in-the-loop. pretty nifty.

edit: also these jobs are literally hiring right now. you can polish up your cv and send it to a DOD lab and in a few months you will be using deep learning to kill. SICK!

[gwenzek]

How is using unmanned drone to automatically kill people around the world different from blind terrorism? It's just a matter of precision.

I agree it's inevitable that someone ill intentioned use ML + drone to create deadly weapons, I don't understand why USA is doing it, nor what are they trying to accomplish like that.

Drone strick is already a main argument in terrorist propaganda, I don't see how increasing the number of attacks and decreasing human intervention is going to help that.

[mathinpens]

Well currently the unmanned drones do not automatically kill anything. Each drone has `pilot' who runs that drone and I believe only that drone. that pilot is in arizona somewhere instead of in the plane.

look I am pretty critical of american foreign policy and our current state of imperial dominance but I mean...calling it blind terrorism misses a huge amount of nuance.

most drone strikes are very targeted in that they try to kill specific groups of people in very specific locations. However, in a counter insurgency war, from 20,000 feet + the fog of war it is quite easy to drone a wedding for instance.

> I agree it's inevitable that someone ill intentioned use ML + drone to create deadly weapons, I don't understand why USA is doing it, nor what are they trying to accomplish like that.

Do you really not understand it? People in government think this technology will have substantial military benefits and they don't see any compelling reason not to develop the technology.

> Drone strick is already a main argument in terrorist propaganda, I don't see how increasing the number of attacks and decreasing human intervention is going to help that.

I don't really know if more autonomous weapons will result in more air strikes/drone strikes or less. I doubt it would have any huge effect on the number.

Moreover I don't really see increasing automation will have much bearing on propaganda. I think people in the middle east don't like drones because their family, friends, and neighbors are killed. The degree of automation in the kill chain is probably the last thing they are thinking about as they grieve.

[DuskStar]

Regarding that SDB improvement... Isn't that something that's been done for a while? Not with image recognition, I mean, but the general "fire weapon at area and let terminal search + guidance deal with the rest" methodology. HARM missiles, Anti-Ship missiles, that sort of thing. Granted, it's probably a little bit harder to misidentify a radar installation or warship than a stationary vehicle...

[mathinpens]

I am actually not sure. before posting this comment I spent like 20 minutes reading the defense literature/media to get a more precise understanding of the technical details. The commentary is pretty vague but I think the difference between HARM/Anti-Ship is you fire-and-forget but before you fire, the missile does have a TARGET that the missile then is tracked onto while in this case you release the bomb to a general area and then it either identifies a target and locks on or it destroys itself in the air. I think that is somewhat different from previous smart weapons.

warning: this could be apocryphal since I can't find the documentation that made this distinction explicit but i remember reading about this a few weeks ago.

[DuskStar]

Hmm. My recollection was that the later Anti-Radiation Missile systems would do a form of target persistence to deal with "target radar switched off during flight" issues, but that wasn't a feature of the original models, such as the AGM-45 Shrike [0]. So with that, you might have a target in mind when you fired... But the target you hit would be the one that was radiating when the missile got close enough.

Automatic visual targeting is still something rather new, though.

[0] https://en.wikipedia.org/wiki/AGM-45_Shrike

[mathinpens]

> The weapon gives pilots the ability to destroy moving targets on the battlefield. Its seeker detects, classifies, tracks and destroys targets, even in adverse weather conditions from standoff ranges. "We call SDB II a game changer because the weapon doesn't just hit GPS coordinates; it finds and engages targets," said Mike Jarrett, Raytheon Air Warfare Systems vice president. "SDB II can eliminate a wider range of targets with fewer aircraft, reducing the pilot's time in harm's way."

I think it is pretty hard to parse just how autonomous these are.

'https://www.prnewswire.com/news-releases/developmental-testi...

edit: muahaha I found a good reference

> Once launched, the SDB-II relies on a sophisticated package of internal computing and algorithms that are designed to get the most out of its tri-mode sensors, and make the process of launch and targeting as simple and flexible as possible for the pilot. The GPS/INS system or datalink messages guide the bomb toward the target during the initial search phase, while the tri-mode seeker gathers initial data. A revisit phase combines information from all of its sensor modes to classify targets. That’s especially useful because the SDB-II can be told to prioritize certain types of targets, for example by distinguishing between tracked and wheeled vehicles, or by giving laser “painted” targets priority.

so some amount of classification/targeting. again I am unclear if the bomb can re-prioritize targets without confirmation from the pilot but it seems like technically very possible.

https://www.defenseindustrydaily.com/raytheon-wins-usas-gbu-...

[mirimir]

That's true, I'm sure.

But even so, those who contribute to weaponized AI still share moral responsibility.

[mathinpens]

I am sorta mixed on this. Is it better to be killed by a person or a computer? does it make a difference? what if the computer system kills less `innocent' people than a dumb human? I don't think it is that black and white.

It is very troubling though.

[mirimir]

In the short term, it means that high-tech nations can start wars with less concern about citizen backlash over troop deaths and injuries. That's already evident for the US, given modern weaponry. They're still limited by costs of pacification, however. And AI will help reduce those costs.

In the long term, Skynet.

[mathinpens]

meh we already are kinda in that world. as for skynet...throws hands in the air

[more_corn]

Picture yourself on the receiving end. You don't want fast, efficient, perfect targeting. That's the stuff of nightmares. The Italian fascists were every bit as ideologically scary as the German ones, but the German ones were ruthlessly efficient and capable. Let's not make the weapons the Nazis would have wanted.

[mathinpens]

lol. obviously if I am on the receiving end I want bad targeting!

[dbuder]

They already have that.