In my setting there exists some AI based weaponry. Though it's not sentient/AGI level. Rather it's a fanciful extension of what we have currently in terms of practical and theoretical. One of the militaries in this setting has a primarily automated military with minimal troops.
An airstrike for example starts with a drone loading up an aerial drone with weapons, this drone is then given a signal by a base computer to taxi to the runway. By this point, information is loaded into the drone such as target location, target profile, way point navigation etc. The drone flies off, and then autonomously kills its targets, returns back and lands. At no point is there human involvement, especially once the drone leaves over the horizon from the airstrip.
This automated warfighting technology isn't just relegated to the air. It exists across all facets of warfare. From automated sentries, to aerial, ground, surface fleet and undersea drones. All independently fighting and returning back to base on an interval to get new orders.
This is because my world is inherently hostile to long range radio communications. A particle cloud layer does a variety of things to inhibit long range radio control. The quick rundown is that it prevents things from going past the upper mesosphere and bounces radio signals inconsistently in the ionosphere. All this means the following: No GPS, no stellar navigation, no ionosphere bouncing (sky wave).
This means that autonomously controlled vehicles have a big advantage over semi active guided systems since they can guide themselves to a target, perform an operation and return all by themselves. Taking the human out of the loop also means less logistical strains. Humans are mainly used at bases for debugging purposes or updating orders.
A faction has setup these types of air, ground and naval bases all over a continent. Using their autonomous army to exert power and force multiply their military. This faction has never had the human population numbers to fight toe to toe with other factions and rely heavily on their autonomous army.
In an attempt to turn the tide against their autonomous enemy, another faction, has decided to weaponize the very same particles that wreak havoc on their atmosphere. This artificial particle/dust cloud can be deployed on command to significantly interfere with an autonomous vehicle's sensors or outright cause physical damage.
What properties does my artificial dust cloud have such that it affects autonomous combat vehicles, but leaves crew served weapon's platforms relatively combat capable?
Any answer doesn't have to outright destroy the machine. It just has to make its job harder to the point that a human operated vehicle isn't vaporized in mere seconds. Ideally it should be good enough conceptually to cause mayhem in an enemy's automated lines though.
- There are a variety of particles with some weird behaviors. One can essentially write in a property to help an answer. In terms of wild capability, a misty cloud when charged has the capability to slug off multiple long rod penetrators.
- Controlled particle shielding exists. So, particle/dust cloud shape/structure can be controlled into specific geometric patterns.
- There is a limited form of human brain computer interface. Using a lab grown organism, it can convert human thoughts into either analog or discrete signals (haven't decided yet). These signals are sent to other lab grown computers which handle subroutines. Essentially a biological CPU of sorts but distributed. It does have true parallel/multi-processing unlike computers which utilize threads and locks to simulate parallel processing. Requires a person to be hooked up to it at all times to work. Otherwise, the whole machine goes dark logic wise.
- The size of the autonomous vehicles in question are full fledge war machines. As in tanks, jets, bombers etc. They might be smaller because they don't need a crew, but ultimately larger machines with big weapons require big engines.
- Something like a cruise missile or a smart bomb doesn't fall under this AI weapon since they are preprogrammed with at best some minimal evasion patterns. The automated weapon's platforms for this question are inherently capable of learning using neural nets, adapting based on previous combat data, evading weapon's fire, protecting itself etc. That said, I have no issue with this cloud's property affecting them as a byproduct.
Some things that I considered but ultimately don't work as answers.
- Sensory/visualization: The cloud mainly affects sensors. Thus, one could very easily blind a machine quickly. It would be the equivalent of throwing chaff and flares right at the sensors but constantly. This relies on the notion that AI systems don't recognize images the same way humans do. In my novice understanding of AI visual classification systems, they process images based more on how closely they resemble something based on certain parameters, features and geometric arrangements. An arrangement of pixels and bitmaps are compared to vectors and certain weights to create a confidence interval in identifying an object. Whereas human based sight is more pattern based and learned through biology. My understanding in this area in general is poor so I'm not sure if what I'm saying even makes sense. The idea here would be that the particle would create a lot of noise for a drone's image processor and classification algorithm, thus reducing effectiveness.
Why this doesn't work: Noise reduction algorithms already exist and have existed for a while. Similarly, anything that would completely blind a drone's sensors, would more than likely completely blind a crewed vehicle's sensors. Thick smoke for example would completely blind a driver or pilot from maneuvering. The situation is even worse if it's broad spectrum. Disabling radar, optical sensors and even eyes for a human pilot would leave them completely blind.
- It interferes with the speed of electronics inside semiconductors. This relies heavily on the assumption that an autonomous platform requires significant voltage/power usage for computing. Existing military hardware for something as complicated as mission computers on an F16 are fairly distributed and rely on are in the 300-400 Mhz range. My assumption is that autonomous vehicles will be doing a lot more thinking, thus they need more computing power. Existing AI systems require a lot of computing power, even with advances in computing techniques and algorithms, I'm assuming that you aren't going to get classification, navigation and combat algorithms to all run in the 300-400Mhz range.
Why this doesn't work: This only works if technology is forever stagnant, which makes little sense in a war spanning years by the start of this question. Even currently, advances in computer science have allowed some lighter weight processors execute some impressive AI/ML simulations and train. ASICs and FPGAs can be utilized even further to decrease the power requirement for computing. Lastly, if it can affect a drone this much, there's a good chance that one's own machine can be affected severely as well. If your fly by wire system stops working all of a sudden, and you have no mechanical back up (which involves most modern and next gen platforms) you will quite literally fall out of the sky because of your own weapon. If you're a tanker, having your active protection system or electrical drive system shut off mid battle would be a practical death sentence.
Lastly the sci fi tag is very much in effect considering things like particle shielding exists alongside early military grade bio computers.