The defence press is reporting that an AI shot down a top F16 pilot in a dog fight simulation on 20th August.
“In a 5 to 0 sweep, an AI ‘pilot’ developed by Heron Systems beat one of the Air Force’s top F-16 fighter pilots in DARPA’s simulated aerial dogfight contest today.“
I have been vaguely aware of a few “advances” such as the development of F16 drones, talk of swarming robots, etc. but I’ve not been following it much, until now.
What concerns me is that we are on the cusp of another arms race. In my opinion that in itself is not necessarily a bad thing, depending on your politics. However there is one added factor that should give pause for thought – possession of nuclear weapons.
In all previous wars despite the damage and mayhem, the planetary environment remained relatively intact afterwards and healed, for example you have to look quite hard for traces of WWI, WWII and Vietnam, etc. With the advent of AI we are potentially introducing a major asymmetry. This time though, the losing player may have the option of ending human life on a very large part of the planet for a very long time.
For example, in 1944 as the Russian Army closed in on Berlin, had Germany possessed a few immobile nuclear weapons, would Hitler have decided to take Europe with him to his grave? I think so.
Diplomatic and strategic skills of a very high order are going to be needed to avoid a losing nuclear power from overturning the chess board. I don’t think any country possesses them. What sayeth SST?
I don’t think any of us want to bet the farm on fallible human nature…I know I don’t. Nuclear weapons are a total game changer. They are the all-purpose hammer to a scared, non-creative user…and, yes, I do think Hitler would have used them.
Daniel Elsberg’s book “Doomsday Machine” explains the concept of nuclear winter. He says that the nuclear weapons now possessed by India and Pakistan (not including USA, Russia or any other nation) are enough to cause nuclear winter that would terminate human civilization on earth.
Highly recommended reading.
What I found interesting is that a startup company Heron Systems beat defense giant Lockheed Martin’s entry too!
Says something about where real innovation is taking place. But what Heron lacks is big time lobbying power. Hopefully the DoD will give these young upstarts a real shot.
Correction to my earlier comment on nuclear winter.
Here is what Elsberg actually writes in “Doomsday Machine”:
“There is every likelihood that, for comparable reasons, similar secret
delegation or Dead Hand systems or arrangements exist in every other
nuclear weapons state—China, Britain, France, Israel, India, Pakistan, and
North Korea—meaning that a Hiroshima-size explosion on any one of their
capitals and/or central military headquarters is likely to lead to full-scale
launching of their ready forces. The only difference is that none of these
states could, at present, cause a full-scale nuclear winter, though an
exchange between any two of them (except North Korea) could trigger
enough global reduction in sunlight and loss of harvests for a decade to
cause nuclear famine and the starvation of one to two billion people or
We have the best government can buy, whether you want to talk of military governance, corporate governance or government in general. I like to term where the world is these days as “peak corruption”.
I do not see any ‘white hats’ moving in the ranks – all are shades of gray. This is nothing new, but the dearth of white hats and altruistic people in governing bodies makes me view a new arms race with some trepidation.
Further, an AI arms race contains within it the same plot elements as ‘The Terminator’, ‘Stealth’ and ‘War Games’ – control is ceded to AI and things do no go so well. Trusting expert systems, ALs and such is fraught with millions of lines of code written by teams of programmers all sitting in cubicles doing their part – they have no idea of what the whole is nor what it might be used for.
The current world is hypercomplex. Our complexity is so high it is difficult to even model with much accuracy. There is more data already collected about each of us than can be parsed in the next 20 years. War gaming software is rewritten almost yearly and still is not truly representative.
Removing the human element, empathy and consequences from the equation may win – but the cost may be WW4 – fought as Einstein imagined with sticks and stones.
Isn’t war politics by ‘other means’? The purpose is always what – to get what the other guy has? Territory or people or even some beautiful woman? The rationale is always there for politicians, as it is rarely just the military that takes over and marches to war. Maybe politicians should just fight it out on the floor of the senate, televised for us all. Consequence right there for all to see…
Free download of Doomsday Machine by Daniel Elsberg here:
I am not convinced that this is a major game changer in the field of nuclear deterrent.
AIs have a big weakness that humans don’t have. Zap the computer core on which they run with an electromagnetic pulse and they are dead. AIs would only provide a major advantage if the location of the computer cores that run the AI can be kept secret to prevent them from being attacked. I doubt this is possible.
The air combat example is more interesting. Modern neural network AIs can learn to play even real-time strategy computer games with imperfect information to a skill level that exceeds the skill of all but the very most advanced human players:
“DeepMind today announced a new milestone for its artificial intelligence agents trained to play the Blizzard Entertainment game StarCraft II. The Google-owned AI lab’s more sophisticated software, still called AlphaStar, is now grandmaster level in the real-time strategy game, capable of besting 99.8 percent of all human players in competition.
It’s also to show the benefits of reinforcement learning, a special brand of machine learning that’s seen massive success in the last few years when combined with huge amounts of computing power and training methods like virtual simulation.
Like OpenAI, DeepMind trains its AI agents against versions of themselves and at an accelerated pace, so that the agents can clock hundreds of years of play time in the span of a few months.”
The problem is: The “AI learning” approach used only works well for “games” that can be simulated in a computer. And in order to be simulated, “games” need to have well-defined rules that can never be broken.
Does that apply to air combat too? The laws of nature cannot be changed, but the rest is not so well-defined.
Has any pilot ever tried to use specific “anti-computer” tactics to exploit the weaknesses that the AI agents have or to confront them with new situations that never came up in the learning simulations?
And how would utilizing an AI pilot it work in a real combat situation and not in a computer-simulated combat? Does the AI remote control the aircraft? Then you can jam the control signals and defeat it with these EW measures. Or is the computer that runs the AI agent built into the aircraft? Then hit it with an electromagnetic pulse.
Walrus, I’m not sure I understand your argument as to how the introduction of an asymmetry in AI (applied in conventional warfare) leads us to the greater risk of a nuclear exchange.
One of the axioms of MAD doctrine is that conventional superiority is rendered moot so long as the weaker adversary retains the potential to overturn the chessboard, as you put it. I therefore don’t see how asymmetric distribution of AI, or any other military technology for that matter, changes this. The one caveat being that the technology in question does not nullify the credibility of the nuclear deterrent.
Computers are replacing humans everywhere and will soon replace fighter pilots. I suspect the most significant impact will be the loss of aggregate ego among the world’s air forces.
I would say AI controlled fuel air drone swarms are the real concern.Nuclear bombs are obsolete due to MAD
neo liberalism is what runs the west and is centered in the usa. russia/china and their allies are creating a different kind of economic and social organizing philosophy and system.
the two are incompatible and only one will prevail in due course.
the west run by the usa appears to be determined to control the entire planet and bring all recalcitrants under neo liberal control.
the russian/chinese alliance will never agree to this and have the economic/military power to thwart any attempt by the west to forcibly convert them.
AI is the not problem at least over the next 10 years or so. what is the problem is the wests unremitting war to bring into its fold those refusing to participate.
the west is neither economically or mlitarily strong enough to force the matter so information war and sanctions are utilized with ever diminishing effect.
there are 2 possibilities going forward. the west backs down and the world has 2 competing organzing ideas for society where one naturally evolves through popularity to over whelm the other or the west fights to the death and one emerges the victor.
if the west chooses to fight that is where the great fear of nuclear war will occur and the west knows with certainty its home turf will be obliterated by russian weaponry even if they manage to destroy russia in the process.
if you watched the democratic convention the one thing that comes through loud and clear is they are a suicide cult not a coherent political/social/economic program
It reminds me of when I was on active duty in the ’80s. All the naval war games we participated in used a rule that it took the USSR 10-12 torpedos to sink an American carrier. Even when you simulated hitting them with a broadside they all miracutlously came back to life, with the air wing, at midnight, the same way Cinderela’s carriage turned back into a pumpkin. It’s election year. DARPA want’s a bigger budget, a contractor is promising the sun, sky, moon and air supremacy over every third world country without an air force, and jobs for procurement professionals approaching retirement. But of course the AI works, in simulation, just like automotive AI, which still can’t get a car to go across country.
My suggestions: 1. Make the communications secure so you can remotely pilot these, then eliminate the pilot and all supporting infrastructure in the aircraft and you can change the operating parameters significantly. 2. Practice combat that way.3. Make an actual evaluation of potential enemies and their air-to-air capabilities and make it public so we can know just what we our spending money (of which we now have a couple trillion dollars less available) on.
A losing nuclear power will overturn the chessboard. Case closed; and AI will have nothing to say about it. The doctrine of unconditional surrender insured that that would be the case. Harry Truman authorized the use of two atomic bombs to bring the war in a pacific to a close, presumably in good faith. There is a continuiing debate 60 years later whether or not whether it was necessary. We needn’t get into one side or another of the sides of that debate to ask the question. Would the Japanese, if they had the ability to retaliate in kind, hesitate to retaliate iout of some abstact concern for the future of the human race. Reverse the coin: the Japanese got to the table first and dropped atomic weapons on Los Angeles and San Francisco. Would Harry Truman have reasoned that we had better not drop atomic bombs on Hiroshima and Nagasaki, even if someone in the situation room had told him that if you go ahead and do that you run the risk of nuclear winter. The question answers itself. With mushroom clouds rising over Los Angeles and San Francisco, concerns about nuclear winter, even if he was aware there was such a thing, would be the last thing on Harry’s mind. Ditto Donald, Putin, Xi, Boris, Macron, Bibi and whoever it is who is running the show in Pakistan and India. That cat is not going back in the bag and all the AI developments in the world will avail nothing in getting it back in.
That does not mean that war is inevitable.
you should look into what Imperial Japan did to the people of China, the Phillipines, and all the other places they occupied.
Most of the recent analyses say that ~30 nuclear detonations are sufficient to reduce average global temperature by 1-2 degrees for 10 years. Essentially a pause in global warming. So maybe localized famine but not full-on Nuclear Winter for all.
It looks like Ellsberg is talking about irradiated Indian farmland rather than global Nuclear Winter.
30 detonations is about average in an open-source India-Pakistan scenario.
If we are talking about Dead-Hand systems, they will not be limited to bilateral nuclear scenarios. India expects to fight 2 fronts; DPRK ROK & Japan; etc. Almost all of them probably will trigger worldwide escalation.
You seem to be unaware of recent developments in AI, GPT-3
(must click on “Show this thread” for full text)
Part of the MAD problem and the use of AI within the various Dead Hands, is the generation of nuclear weapons held by the various parties. Those countries who have not had the scientific and computing resources are stuck with early generation weapons. Those that have have continued to develop them so that now they have the ability to ‘dial a yeild’ it seems in most cases on launch.
An example is the USN publicly loading mini or micro warheads on their missiles. The Russians, in their role of obvious target, responding by saying that they don’t know what the warhead is so if anything is launched then they have to assume the worst.
That might not change MAD but smaller warheads certainly change things when attacking a non nuclear armed adversary. Such that when the yield can be in the 100-1000 ton category with potentially no or little fallout there are even people claiming that blasts like Beirut could be one. It leads to the possibility of an almost undetectable nuclear war, no mushroom clouds and no nuclear winter. The obvious first target being Iran.
This would be a war by ‘tactical’ weapons potentially used strategically as well. An ICBM launch will always be a ‘doomsday’ weapon whilst a SRBM or cruise missile will have an element of doubt until it reveals itself on impact. Then, even if AI is on the job, was it nuclear or not? Bearing in mind that our expectations may have been softened up over the past few years by nuclear blasts being called very big conventional bombs whilst the lack of fallout is used as a ‘therefor it can’t be one’ hiding the fact that it could now be the result of a design feature.
Regardless of AI we no longer have the security of a proper black and white MAD and shades of grey can be a very uncomfortable place.
An interesting thread, thanks for the link.
Gwern’s description of the scaling hypothesis, popular again with deep learning researchers, is prophetic I think:
Silicon, All Too Silicon no more. One day, soon, one of these things will pass the Turing Test convincingly. Will we have the
intelligencewisdom to control a creation smarter than ourselves? I am not at all confident.
“Alexa, should we put you in charge of the nuclear arsenal’s dead hand mechanism?”
Fred – I am well aware of the barbaric Japanese transgressions in the countries they occupied. My two older brother-in-laws fought in the Pacific as enlisted men, one was in a foxhole in the Phillipines while the Battle of Leyte gulf was being waged; he was medically evacuted from the Phillipines shortly before his unit was sent to Okinawa where it was decimated; the other was a marine who was decorated for action in Bouganville and wounded in action on the second day of Iwo Jima. Neither had much use for Japanese cars.
There is a very useful symposium in the current issue of Chonicles magazine on whether the use of atomic weapons against Japan in retrospect could be justified. It ranges from absolutely with the usual reasons being advanced, to no, the Japanese had been defeated and it was a matter of coming to terms. There is also a novel twist discussed of which I had not myself been aware. Apparently a Japanese historian has advanced the argument that it was not the bombs which induced the Japanese to surrender unconditionally – leaving in place the Emperor – but the Soviet entry into the war. The Japanese did not want to be occupied by Soviet troops. Given the experience of Eastern Europe, this seems in retrospect to have been a prudent decision regardless of what in actuality brought the war to an end.
War seems to be an intrinsic part of the human condition. We hope that nuclear war is not; for the very good reason that it would be absurd to entertain the belief that a nuclear power would sustain a nuclear attack on itself without instantly retaliating in kind. I believe that I read recently somewhere that it is a part of Russian war doctrine that it will not fight another war on its own soil.
Hoping that our neighbors in the universe decide to step in and save us from ourselves. We are such a primitive species.
I am fascinated by the potential for AI in air-to-air missiles. My undestanding is that traditional air-to-air missiles are useless once they lose a lock on a target, or if they miss a target. This seems to me to be an artifact of rocket motors that accelerate the missile to top speed, then burn out and can no longer provide thrust. If instead a missile is powered by a ramjet engine (eg the Meteor), then it can be something more akin to a hunter killer drone than a simple missile.
Hunter killer ATGMs are also possible. I think this is where AI can provide a lot of value – more so than AI powered fighters. If the missiles are smart, the vehicle that gets them to be “within range” can be relatively dumb.
Re: Hiroshima and Nagasaki
Elsberg writes that the firebombing of Tokyo killed more people than Hiroshima. It melted the pavement. It boiled the water in the canals/ditches. There was no hesitation by Truman to bomb Hiroshima, none whatever.
Elsberg also points out the differences between Hiroshima and Nagasaki, he says that most of the reasons for Hiroshima no longer existed for Nagasaki.
And yes the Russians were coming, as Stalin and FDR had agreed that USSR would enter the war against Japan 90 days after Germany surrendered. So USSR secretly sent major forces into Japanese occupied Manchuria and succeeded in a surprise attack on Japanese forces there.
At that time the Japanese envoys were in contact with Stalin and attempting to agree on surrender terms. When USSR declared war on Japan (which it had NOT done heretofore) the Japanese entreaty to Stalin was over.
Also worth noting that nuclear winter was not a known concept until 1983. For many years USA was attempting to build a nuclear force great enough for first strike against USSR (and also China was to be attacked in the same plan). These USA planners did not realize that their attack would destroy civilization of planet earth.
Please I urge you to read the Elsberg book which covers all this ground, and a lot more. The link to the free download is in my prior post.
Indeed Gwern is the guy to go for all things techno.
He even has a (free) monthly newsletter.
There is nothing novel about a Japanese historian arguing that the Soviets are the reason Japan surrendered, that arguement was being made decades ago by the anti-American left. The Imperial Government, or more precisely the militarists within the government, tried to stage a coup rather than actually surrender. The mindset is alive an well in America, see Michigan:
““A declared statewide emergency only ends upon the governor’s declaration that the emergency no longer exists.”
@ “… you have to look quite hard for traces of WWI, WWII and Vietnam, etc …”
Not so for Vietnam: https://www.nytimes.com/2018/03/20/opinion/vietnam-war-agent-orange-bombs.html