Should there be an "ALL OFF" button to instantly shut down all these new AI Defense bots that the Military in the US want to build and deploy in the thousands?
decrypt.co
The Pentagon Is Accelerating AI and Autonomous Technology America’s military leaders are racing to deploy thousands of autonomous weapons and an AI-powered air monitoring system for Washington D.C.
Maybe can we take a step back and ask whether we need thousands of AI defense bots at all? Or are we past that point?
Realistically, who’s gonna stop them?
I believe in you, or should I say, gggrrrrrrruuuuuuuuu(pardon my wokie I only had one semester in college)
I don't like it but I think we do. China and Russia will certainly have them and they will get ten times better in a same amount of years.
I watched the Ted talk on defense drivers. Scary shit. Thing is I work with commercial cameras and have, in hand, camera that can not only identify all kinds of objects such a human's,, they can recognize individual humans and put a name to them. They can recognize if people are loitering or if someone is being followed. They can reconsider a car from a truck from a bus from a bike. This is not done in a server but thru the power of the CPU in the camera alone. The cost. 500 dollars.
Point being the power available in such a low cost item is staggering. Combined with a weapons platform and it is scary. A terrorist group could distribute hundreds into bushes and they could just sit there for a week in low power mode, waiting to recognize a simple person and spring into action. This is stuff we have right now off the shelf.
What will be part of military arsenals in ten years will eclipse this current tech significantly. Troops won't be ambushed by live human fire but by thousands of drones that care not for their survival.
Can I get a few of those cameras and have them record any people around my house that aren't me? What are they called?
Hikvision line mainly if you want the low cost ones.
They won't do a 'not me' identification. Mainly because it can only identify you or any person if they get a decent view of you. Basically the first event will be 'i see a human' and if you look at the camera then it can also do an event and say basically 'Jack black' is here. It is two different kind of events you need to turn on. But the person recognition can only fire if it recognizes you.
I thought the same thing is you in that I could have it ignore known people. But it like you looking out a window. You see someone from a distance. To recognize them you need them to come closer. Thus as a person you don't call the cops or create an event immediately but at some point you might. The cameras are not quite that smart yet but as said, ten years?
It’s the nuclear arms race part deux: AI boogaloo.
What I think is dangerous is terrorists or mass killers getting dozens or hundreds of small drones and installing explosives on them. Install these cameras and CPUs you mentioned that can recognize human faces and have them fly into someone's face and then explode.
You could kill many people and unless we start installing AA turrets all over our populated cities, there seems to be little we can do to stop it.
I had mentioned that in an earlier post. It is pretty scary. They could sit in a bush for a month using extremely low power motion detection. See any motion turn on camera to look for human recognition.
Autonomous drones made by China have been used in Papua New Guinea to bomb at least one village so I think the US is actually behind the curve in terms of the AI arms race.
This is one of those classical sci-fi apocalypse ideas, where humans make autonomous war machines they can't turn off, and the machines outlive the humans and continue the war for them.
How is it that when it comes to reckless ideas and notions Congress takes millions of years and the Pentagon takes no more than three business days to implement?
Most military networks are closed circuit by design. I’m not sure how this could be implemented without also allowing back doors to be exploited. You wouldn’t want someone to be able to turn off your defenses as they begin an attack, for example.
There are a number of ways to do it. You can transmit a one-time code to the device that you set up right beforehand. No one's going to be able to guess your 1024 character one-time password.
You can even protect the password entry program itself with port knocking. If the right ports aren't accessed in the right sequence, the enemy doesn't even get a chance to try their passwords.
Every server is on the Internet 99.999% of the time. They are constantly being tested. The right cybersecurity tools are available now.
Just make the code 00000 like the nuclear launch codes were for years.
I'm sure (or at least I hope) nuclear weapons have similar systems in place so that they can be launched or shut off as needed?
In what ways would this be different
Yeah, they don't. Nuclear systems are for the most part closed sourced and built on DOS level hardware. Most of that shit can't connect to the internet even if they wanted it to. The system you're thinking about is radio waves between people talking.
Makes sense :)
The department of defense was hacked just a few years ago, suck a button would have to have access to an internet. Meaning anyone could get to it and shut off the drones and such
They'll just murder a bunch of people and then be turned off after having been shown to be
ineffectivetoo dangerous.It's not like AI is reliable at this point. Way too many people are actively ignoring experts pointing this fact out and instead obsessing over Skynet or w/e made up sci-fi BS.
Rather than be used for war, they'll be used for threats of violence and propaganda. It's not a new problem. It's just a new version of that same problem.
Surveillance is the word you're looking for. Take all those NSA pipelines and run them through an AI and BAM, you've got your "terrorists".
AI will be downfall of our technological society, but not because of killer robots and malevolent systems. It's going to make everyone completely and utterly incompetent at everything in life.
Well said.
It should be noted that individuals at the forefront of AI research have a direct bias against saying AI is dangerous. It's their job, and saying anything which presents this research as dangerous could halt funding, and put them out of a job. It's also their passion, though, so it's an even bigger deal for them.
We have also seen individuals who have exited AI research calling for more regulation and ethics requirements. At the same time we are seeing AI ethics departments dismantled. These should stand out as red flags.
Autonomous drones are actively being used to bomb villages in Papua New Guinea. The idea that this kind of tech is "only going to be used for threats of violence and propaganda" is already outdated. It's being used today, and the US just plans to also adopt the tech itself.
humanity sucks
Sounds like the beginnings of the plot to Horizon Dawn. Can’t have it both ways, either it’s a secure closed system with no way to stop it if it goes rogue or it has safety’s built in but then those could be exploited.
NO! Movies would be so much more entertaining if the bad guy learned the error of his ways but was still unable to stop the robot slaughter.
There is also a danger that the kill switch command could be leaked to the Russians or the Chinese who would use it to shut down the USA's defenses just before a full scale invasion of a now defenseless USA.
I don't think the robots are the only military America would have. The military is kind of their thing.
No. Such a thing would only be a good idea if you want the enemy to be able to turn your shit off when they please.
You're thinking of 'AI', as something intelligent that can go rogue. Current and near future that's just sci fi.
That’s what they want you to believe. All hail the robot overlords!
Military AI is already going rogue. It doesn't need general intelligence to act unpredictably.
That's called a bug - aka what it's called when a program behaves unexpectedly and against design intentions.
That's not going rogue, that's doing what it was programmed to do.
By your standards you'd also have to consider WW2 acoustic homing torpedos as rogue AI because they might home in on the ship that fired them.
Edit:
A followup thought: the only real question is whether they can realistically test and refine these systems enough to trust them to carry out attacks autonomously without serious errors.vIm gonna guess no, but they'll use them anyway.
Your edit follows the point I was making. It doesn't need to truly "go rogue" according to your definition, and it doesn't need general intelligence to have the same disastrous outcome. We have examples of AI killing humans to accomplish the goal it is given, so we need to be damned sure that's not going to happen in real life before deploying them over Washington DC.
Honestly that wasn't even a bug, it was a perfect execution of the instructions it was given to perform its task with maximum efficiency and would have been incredibly easy to see in advance if anyone had spent 5 minutes thinking about it. Classic paperclip maximizer style literal interpretation of goals.
Logic errors are bugs.
Or how about not building them at all
It is not the world we live in unfortunately. Tell that you the Russians, to China or to some terrorist group with a few hundred thousand to spare.
I agree. It always has been an arms race and probably always will be. Humans are too aggressively murdery for our own good. I mean it served us well when we actually had real wild adversaries, but now it's tearing our civilization and planet apart.
We still do have those real wild adversaries. Russia and China are only two that are of the conventional type.
The less obvious ones are the ones we don't think about so much. Iran, Iraq, North Korea, India, Pakistan, Argentinian at one time and so fourth. Most of these countries stay in line predominately because they now understand the likelihood of Western nation military involvement it they act too aggressive.
Point being if all Western nations backed out of the arms race right from the beginning. From WW2, I suspect WW3 would have happened long ago. One or more of those other countries would be far more active in one way or another. If we backed out now, it also would not be long before those same countries realize they could act without impunity. Nothing has really changed. Just the military damage potential is far more serious.
This is the most troubling to me. They're entrenching themselves. They already wrapped razor wire and concrete walls around the white house. Now they're deploying military assets on US soil.
They have been deploying military equipment for decades now on US soil, under the guise of police.
The new development here is that this system depends on far fewer humans and their consciousness.
Sounds like the opening to the most recent Battlestar Galactica series. A kill switch can be hacked.
We're not gonna make it, are we?
Most people in this thread need to learn the difference between AI and AGI
I'm sorry Dave. I'm afraid I can't do that.
Anti-robotics is where the money is.
Are armor penetrating EMP rounds a thing yet?
More likely we'll see a button to instantly transfer money from tax payer pockets to these companies' CEO's pockets
Don't forget to add the option to tip that's in vogue for everything these days.
Would you like to add a tip for your robot defenders to this months subscription fee for AI Protection Max Ultra (tm)?
Sorry I tipped my landlord my last penny
Age of Ultrom 2: Electric Boogaloo
There really would be no way to have an interface that shuts it down, that an AI wouldn't be able to compromise. Though I imagine the military will set up a plan to blow up it's connections to power and the internet should it go rouge.
People use the term AI too loosely. We don't actually have artificial intelligence. We have neural networks that can perform tasks based on training data, but it's not actually intelligent by any means. That said, fully autonomous systems have existed for decades already. Look up CIWS or CRAM.
I'm not going to get into a "no real Scottsman" arugement about the advancements in this field
Good lord. You're talking about programs being smart enough to disable their own killswitch. That's not a fucking thing.
Worse. They are taking about a program that has some sort of motivation or desires to disable the kill switch. It is quite silly yet. Worst case is a program bug that somehow disables. That would be highly unlikely. Particularly in such a way it would be dangerous.
I wonder what would go down if a CWIS managed to go rogue. Like I know it’s literally impossible, but if it was what would it do? Would a Phalanx just fire 20mm at random? It’s an interesting hypothetical imo. Obv I agree with you that current tech is most def not even close to smart enough to disable their own programming
The Phalanx does auto target. It just can't fire on its own. They're designed to shoot down missiles and humans can't aim that fast so they're fully autonomous except for the fire control. From what I remember targets are assigned and prioritized by the AEGIS system on the ship but the Phalanx is designed to do all of the actual targeting on it's own so that if the ship targeting stuff is damaged it can still function perfectly fine without it.
Theres a short video out there of a phalanx targeting and tracking a passenger plane. Aparently they're also known to track birds and people walking around on the the ship.
So as far as one going rogue, all it would take is for the person in the control room to give it the fire command at the wrong time. Of course they fire at 4,500 rounds per minute and their large magazine only holds 1,550 rounds, so it's spree of carnage would be pretty short lived.
More likely, P.O. Malarky gets an ass full of hot lead because he forgot the reapply the lube during the previous maintenance.
Maybe if FC2 wasn't always gun decking his shit he'd have less ammo inside him.
Better dead than rouge, I always say.