Wartime use of artificial intelligence

Related spinoff of existing threads of AI, since I don’t want to hijack the other AI threads:
I know AI can be used for a lot of drones or autonomous weapons systems, but what other wartime uses are there for AI? Can it be used to predict enemy behavior, or suggest an optimal attack strategy, or help guess intentions (i.e., what an unidentified ship or aircraft is up to?)

What could AI do for a military that a human brain can’t?

C4ISTAR: Command, control, communications, computers, intelligence, surveillance, target acquisition, reconnaissance. We’ll call it C4ISTAR. AI is potentially great at turning noisy data into useful intel.

AI, especially neural network-type AI, can be used to recognize patterns in the noise. Imagine a rifle scope that searches for and highlights human-like shapes. Or an air defense network which fuses data from hundreds of sensors (radio, radar, IR, visible light, UV, sound, seismic etc), determines where the enemy is most likely to be (“Is that blip just noise, a cheap decoy, a stealth fighter?”) and cues other sensors to take a closer look.

Computers excel at taking many quantifiable factors and complex interactions into account to pick out information faster, more precisely, more reliably than humans. For example, a human in a tank would likely not be attentive or fast enough to recognize a rocket launch, predict its trajectory and take countermeasures in time but a computer can.

I’m less sure its utility when it comes to predicting behavior or suggesting an optimal strategy at the high levels.

Behavior and strategy? Not really. Or rather, not yet.

But other uses are feasible. Moving target indicators, pattern analysis, and synthetic aperture radar rely heavily on machine assistance in both acquisition and interpretation. Search engines have to comb through huge amounts of data, and there are already digital tools for search engines, interpreting ambiguous or variant spellings, and identifying pertinent facts based on format or context. There are tools for digital exploitation, in which the machine is plugged up to a cell phone (for example) and searches it for pertinent data. The AI is programmed to identify things likely to be of interest, and it can use file hashes to distinguish between files that are already known to be irrelevant and those likely to be of interest. And we’re really just getting started on cyber and electronic warfare. A cyber-attack can be perpetrated faster than a human can react to it, so our network defenses are hugely important and will only be more import in the future.

Edit: Forgot to mention facial recognition and biometric tools.

Fiction but I thought it was an interesting conception of the use of AI in a military context. From the Aliens: Colonial Marines manual each squad of marines or soldiers has a ‘synthetic’ attached (basically an android/gynoid for those who haven’t seen the movie) which are physically based on a male or female human in their mid-40’s with a placid Type-B personality. They aren’t allowed to engage directly in combat but act as a medic and ‘kindly aunt/uncle’ for the, mostly young, soldiers, acting as a totally reliable and discreet confidant and counselor or a guaranteed neutral arbiter of disputes. The system is described as noticeably increasing squad cohesiveness, bonding and combat potential.

Also to add they act as advisor to the human in command although not allowed to issue orders themselves.

We’re a very long way from that but I thought it was a rather neat potential use of AI.

During WWII, many times Allied forces were able to anticipate the location & timing of enemy attacks by a sudden increase in radio traffic. Not the content of the (encoded) traffic; just the increased volume to/from enemy units in a certain area. And during the Battle of Britain, increased radio traffic from Nazi planes taking off from french airfields allowed the British to anticipate where & when to send their defending aircraft.

This was done via the human intelligence of the Allied staff in the local area, who had experience to know what the normal volume of radio traffic was, and could recognize an increase.

Presumably, given modern radio monitoring equipment, an AI could be trained to carefully watch the volume of enemy radio traffic between all the various enemy locations, and raise an alarm when it sees a pattern typical of a pending attack in a particular location.

I’d think that stuff like machine learning could be put to great use in logistical applications and in non-combat orchestration type activities - stuff like what units to put in which boats and how to unload them and put them on trains to the staging areas most efficiently. Or in allocating supplies to units maybe.

Another possible application might be to have an AI look at soldiers’ (or sailors, airmen, etc…) service records as well as their physical attributes, ASVAB scores, SAT scores, etc… and come up with a way to fit people to the specialty they’ll be most effective at. I’d imagine by now we have enough data built up to come up with something interesting if nothing else.

Outside of that, I’m not sure the data’s there to really support its use in actually deploying troops, predicting attacks, etc… with the exception noted above about radio traffic. That’s because most modern-day AI pounds through staggering amounts of data looking for patterns and/or learning how to accomplish a task. I don’t think an AI could look at a map of unit positions and figure out where an attack is most likely yet, because we don’t have tens of thousands or millions of maps that would let it pick out that pattern accurately.

It can learn that the only winning move is not to play.

Something that worries me about our militaries seeming reliance on electronics to conduct operations, what happens when these systems stop working.

Our potential major opponents are surly aware of this and are actively developing ways to disrupt such systems. I am no expert but history is full of examples of cutting edge militaries being overcome by an “inferior” force.

I can, for example, forsee a major conventional war opening with an attack on our GPS and observation sattellites combined with a cyber attack. Such an attack could eliminate a large number of weapons and control systems in the opening hours of a conflict.

I did not mean to hijack the thread. The answers given so far cover most of the potential for the use of advanced computing in warfare. I wonder if map and compass training is still given,?

For the moment, the only country with anything approaching a capability of taking out multiple satellites would be Russia or the EU, and they’d have to take out quite a few before the GPS system would be non-functional, although they could degrade it by taking out satellites. Even then, actually destroying the satellites would take a series of rocket launches- that’s hard to hide.

Plus, GPS satellites are in medium earth orbit (~12500 miles), which would take a while for any kind of satellite interceptor to reach; certainly enough time to spot the launches and figure out what was going on. It’s hard to sneak up on something when you have to go that far.

I’d imagine the bigger threat would be some sort of jamming or confusion of the signals making GPS less accurate or significantly distorted.

AFAIK, they do teach map & compass to soldiers still; I believe they still teach sailors how to navigate using sextants and clocks and what-not as well.

Agreed—and I love that manual. :cool:

I also love the bits where it’s stated that nuclear, biological (pathogen and beastial) and chemical weapons are all in the stockpile and fairly fair game to use…but that equipping androids with weapons or “uninhibited combat capabilities” is against the Geneva Convention!

Although perhaps understandably—while the Nostromo’s Ash is said to be a “Hyperdyne” android, the original draft of the script explicitly says he was built by Cyberdyne, the latter even being mentioned offhand in the USCM Manual.

Or just get ride of the people that start wars.

Back when I was a kid in the 1980s, I read that the most advanced robots in the world were cruise missiles. They didn’t have GPS yet, and so navigated to their targets largely by terrain recognition.

Ha – it wasn’t until reading your post that I recalled that I actually worked on such a system :smack:

In a combat zone, resupplies can take months, even years for some equipment. So you need AI not just to figure out how to transport goods, but even just to anticipate what’s going to be needed well in advance (1 year+), taking into account short and long term shifts in the situation.
I worked on a system like this about 20 years ago.

Oh and also countermeasures (but I think this has been alluded to). Missiles need to track targets, but those targets want to disrupt that tracking / hide. So there’s AI at both ends of this.

In principle it could be used for a form of control that human generals are incapable of. Assuming you can’t or haven’t replaced the individual soldiers/combat capable units with robots, the AI could individually micromanage every single aircraft, tank, and squad on the battlefield. (I am assuming the receivers would be somewhat bulky, what with all the encryption and displays and batteries, and thus 1 per squad).

That is, it would give everyone orders as to where to move, which targets to focus on, and where it anticipates the enemy response to be.

So it would be issuing thousands of orders per second, at a minimum, and each order takes into account the engagement and vision of each and every unit in the entire army group.

Essentially all the other layers of officers in between would be eliminated. This AI would not be sentient, there would be a core staff of officers who can see numerous diagnostics feeds including the overlapping probability grids the system is using to calculate these orders. They would also be able to adjust things in real time.

Yes, the system would not be infallible, and it would sometimes fail to correctly predict the enemy responses or an enemy ambush, etc. But unlike humans, when these events happen, the machine won’t feel guilty about the losses or feel surprise and assuming a fast enough software/hardware architecture, would immediately issue out orders taking this new information into account.

Still do infact. GPS is pretty damn vulnerable.
To the OP:
AI is already used in Counter Intelligence and Counter Terror, its even rumoured that some of the biggest advances have been due to those fields, which have migrated to the civilian world.

Most military use of AI is classifed, but from open sources apparently AI research is apparently being used in fighters jets, newer guided missiles which can detect that they are being targeted, and take evasive action, RADARs and comms.

Yes, they still train with maps and compass for precisely this reason. We have already seen wars in which both US (Iraq) and Russians (Georgia) used offensive cyber attack to disrupt enemy systems. And large-scale exercises now include things like quadcopters, cyber intrusions, and GPS jamming.

The idea of shooting down satellites is terrifying, not just for the immediate effects but also the potential for Kessler syndrome.

We’ve been looking at some solutions here at work that include machine learning, and most of the vendors’ sales material centers around the applications of machine learning for predictive analysis and optimization, so when I read the OP’s question, it jumped straight to mind!

As for some sort of AI commanding individual troops; we’re firmly in the range of science fiction here; modern day “AI” is much more limited than that and are more along the lines of being able to create a model of a solution and test it versus a colossal data set, and then apply that model to current/future data. So predicting what ammunition will be needed and where to send it is right in its wheelhouse, but figuring out what Bravo Company of the 3/143 infantry should be doing is probably beyond its capability, much less lower units.

How useful are AIs at the operational level? Can it prepare a mission analysis, tell me when and where to assault (and likely developments), etc?

Today, that isn’t feasible. Algorithms are being worked on that show this is ultimately achievable but a system that does a task this sophisticated reliably and at least as well as humans would require further development in algorithms as well as it would need to be a complex architecture using many separate subsystems.

Indeed, I also liked the bit where the state of the art sensor technology in the 22nd century had a resolution of 640 x 480. :wink:

Edited to add: Also sci-fi but the Yukikaze novels had an interesting explanation of why a super-advanced AI might keep humans around, at one stage a massive assault is taking place on the facilities with both real physical and simulated attacks. The military AI asks humans to go and do reconnaissance to determine what attacks are real and which aren’t because it was unable to tell itself and biological humans were unaffected.