The AI War Machine Is Much Scarier Than We Realized
The more I learn about AI, the more it scares me about our future. The public has no idea, none whatsoever, about what could be coming. Elon Musk seems to be the only one alarmed:
Not even Hollywood has anticipated the functionalities that could exist. The methods of discovery can be discussed, but what is found is actually unknowable today until it is discovered by our future AI systems. I'll explain why we can't even imagine the weaponized AI in the future.
Unknowable Sensory Systems:
Humans have been proud of our discovery of electromagnetic insight such as RGB cameras, thermal, infrared, radar, x-ray, and millimeter vision systems at airports. All of these are impressive and they allow us to see through bodies, walls, and clothing for hidden weapons. With radar you can even look through space, water, and walls if needed. Bigger vision nerds will talk about hyper-spectral imaging where you merge N-dimensional images from multiple wavelengths into a single image. All of these technologies are childish compared to the future AI sensory processing systems.
AI vision systems of the future will develop into full spectrum scanning systems. AI will comprehend it's environment across all wavelengths in real-time. A wide range of electrodynamic wavelengths will be broadcast between radio and gamma ray and sensitive receivers will analyze the feedback from that data. Human's can't comprehend something this complicated, only AI can. If you want to nerd-it-up you can add audio and magnetic pulsing/receiver sensory systems as well. The ability to scan and collect signals is not innovative, but the ability to digest and infer outcomes from that data (e.g. person behind a wall, gender, heart beating, etc..) using AI is the key point here.
So what if AI is ingesting thousands of frequencies while it walks or flies around? So what?!? At least with Hollywood movies you can pour gasoline on yourself, roll in the mud, or hide in the freezer. With this type of sensory system not only would the AI know you were in the closet it would know your vitals, hell it could comprehend your age and kidney function before it enters your home. There is no such thing as hiding anymore, maybe in a Faraday cage, but it could probably detect the presence of such a masking object. It may even discover that one of these scan frequencies can be weaponized, dark indeed.
The World's Most Terrible Metric: KDR
Today we have AI metrics we applaud, like deaths-per-100-million-miles-driven. The metric sounds dark, but it is going down thanks to self-driving cars. The media loves to put every Tesla injury on the front page of the news, but Tesla has already gone significantly below the death rate of regular human drivers. This number will continue to go down and we will all celebrate.
The darkest metric will be discussed in private, away from the public eye. As AI war vendors compete on billion dollar contracts they will use kill-death-ratio (KDR) as a selling point. This drone/humanoid is capable of killing 11.563 AK-47 enabled human fighters before it is destroyed based on the latest field testing. That number will begin to climb higher and higher driven by money, power, and fear. Think of it as a body count ROI. Eventually, you may hit numbers as high as 1,000s or even 10,000s. For humans to stay relevant they will have to level up on their armor and weaponry, but the AI will adapt and continue to push impressive stats regardless of the human attempts. It is impossible to know what an upper bound would be for a KDR ratio for a 150mph flying drone capable of seeing objects at 1000fps.
ZettaFLOP Wars: Beyond Our Imagination
A big break though with robots and AI is the ability to train them in a virtual environment. Humanoids can learn to jog, jump, and even backflip in a virtual environment without having to learn and fail in the real-world. This allows for much faster learning and iteration cycles. That also means they can practice killing virtually as well. Any AI expert would know that the best AI weapons of the future will fight millions of wars in a virtual environment. Countries might even invite their citizens to fight in these virtual wars on gaming systems to help the drone/droid armies learn. Not to be cheesy, but we all know that young kids will be the most useful teachers here because of their superior hand/eye coordination (this isn't just an Ender's Game plug). Now before you invade a target or other country you will do so having fought the war on that geography millions of times already. With the pre-war simulation information the AI army may even bomb blind targets in a city, knowing that these strategic points cripple the opponents probability of winning the war game. You will have a very good estimate of AI resources needed to win. Any country with ZetaFLOP supercomputing could calculate the droid/drone cost of a world domination battle to check if they have sufficient material resources on hand.
Ready for the brain melt?
These simulated wars with virtual drone/droid models will naturally realize it makes sense to allow for very flexible mutation capabilities. Almost unbounded mutations where some drones/droids would be disabled. This means your droids could grow longer limbs, extra arms, and new weapons. I'm not trying to be gross, but these monster droid/drones would actually mate digitally in the computer (not joking one bit) and produce offspring for future wars. We can't imagine their intelligent designs, we just know they will not be inspired by humans or nature, but bred for maximum performance. Meet your new aliens. The fittest fighters with the highest KDRs will have their digital genetics sent to manufacturing in the real-world where they will be 3D printed in 24hrs for battle. It is actually wise to allow for a genetically diverse droid army. That way the droids that survive and come home get to mate digitally based on real-world KDRs. The AI war hero returning from battle with the highest KDRs would contribute the most digital DNA back into the population.
The scariest drones/droids are the ones that got away because they broadcast their updated kill counts back to base where their DNA is on file. Let that sink in.
The Manhattan project of AI could be the physical modeling of the human body in these war simulations. The medical research efforts might accidentally enable a darker use case for the military. The more accurately you can simulate a real human body model the more you can learn by interrogating it. Think of this as virtual torture. A country with fewer droids could win a global war just by simulating more attacks on a simulated human body. They may discover a new weakness like commotio cordis or an electromagnetic pulse that causes aneurysms or seizures. New insight like this could be a major breakthrough for a country fighting in a conflict.
So to recap the future AI sensory systems will be unknowable, their kill efficiencies will be unknowable, their physical design and appearance will be unknowable, and future human body vulnerabilities are unknowable. Sleep tight!