AI-Piloted Fighter Jets

News from Georgetown’s Center for Security and Emerging Technology:

China Claims Its AI Can Beat Human Pilots in Battle: Chinese state media reported that an AI system had successfully defeated human pilots during simulated dogfights. According to the Global Times report, the system had shot down several PLA pilots during a handful of virtual exercises in recent years. Observers outside China noted that while reports coming out of state-controlled media outlets should be taken with a grain of salt, the capabilities described in the report are not outside the realm of possibility. Last year, for example, an AI agent defeated a U.S. Air Force F-16 pilot five times out of five as part of DARPA’s AlphaDogfight Trial (which we covered at the time). While the Global Times report indicated plans to incorporate AI into future fighter planes, it is not clear how far away the system is from real-world testing. At the moment, the system appears to be used only for training human pilots. DARPA, for its part, is aiming to test dogfights with AI-piloted subscale jets later this year and with full-scale jets in 2023 and 2024.

Tags: , , ,

Posted on June 25, 2021 at 8:53 AM • 24 Comments

Comments

Skynet • June 25, 2021 9:36 AM

Yes, please make fighter jets computer controlled. Nothing could possibly go wrong, I promise!

tomb • June 25, 2021 10:11 AM

I wouldn’t even question the superiority of AI controlled jets over human pilots in combat judgment with publicly available technology. I’d imagine you could even program the AI using a database of human to human combat interactions in DCS (a popular combat flight simulation game) for guidance.

As noted in one of FX Holden’s fictional future-war books, AI could also readily beat human pilots if you had access to their characteristic flight profiles and pilot-tendencies/personality characteristics.

Pau Amma • June 25, 2021 10:25 AM

Prediction: no US-developed version of this will ever be deployed successfully, as the AI will be unable to spot Black or Brown people to shoot at.

Hedo • June 25, 2021 10:30 AM

Chinese state media reported that an AI system had successfully defeated human pilots during simulated dogfights.

When there is a Democrat President in the WH, one should watch Fox News, and when a Republican presides over the USA,
one should watch CNN – this, in order to get a more critical(accurate) view, and less toeing the partyline.
What clouds your judgement is precisely this: toeing the partyline, no matter what.

With all that being said – Chinese state-run media, or RT in the case of Russia – EVERYTHING you hear/see
coming from them, take it with a Mount Everest of salt (not a grain).

Now, with setting the politics garbage straight, on to the realistic approach of Human vs. Machine.
It will always depend on how good/smart the human is BEHIND the Machine(AI).
Ultimately, it always comes down to this question: is the human who developed it smarter
than the human(s) that it’s being used on/against?

If
the AI developer is some very high IQ mofo
then
as a human fighting against it – you’re screwed
because
there are no emotions involved (Logic, boolean-true/false, binary -1/0, yes/no, on/off).
No hesitations, no second thoughts, no regrets, no ethical/moral contemplation
about civilian/human collateral damages.
No split-second hesitations, all of the above is not a part of the equation anymore.
It’s out the window.

Duh, AI has been around for a very long time, it just wasn’t called that back then.
There was no way, decades ago (or today, or tomorrow), for ANY human being to detect an “incoming” very
high up in the sky had it not been for various heat-sensors/detectors and other
sensing triggers being built into most fighter jets’.

So, the conclusion is – nothing new, nothing sensational.
Typical Commie self esteem, moral boosting propaganda.
Let’s say I’m wrong and that this is not Commie propaganda,
it’s still nothing new, nothing special that Logic vs. Man
wins due to not being “infested” with feelings, moods, tiredness,
and a million other “obstacles”.
Not that these “obstacles” such as LOVE, compassion, and many
others that make us what we are – HUMAN, are bad. Not at all.
It’s just cold hard truth.

Ted • June 25, 2021 10:59 AM

Add to the above mentioned capabilities of AI the fact that a computer can withstand far more g-forces than a human and the meat-based pilots are screwed. Drones are the way of the future and the US military knows this just as well as China. Perhaps they fell less need to brag about their accomplishments.

Ollie Jones • June 25, 2021 12:07 PM

Wild thought: if AI planes from opposing countries must do battle, maybe the UN or somebody should set up a neutral simulated-combat platform so these battles can happen without burning fuel or throwing bent and burning sheet metal into the ocean.

It would save a bundle of money and carbon.

Nah. Too much like Ender’s Game.

wumpus • June 25, 2021 12:33 PM

Anyone know how much g-force a fighter jet can sustain in flight (without a pilot)? I’d guess that not only is the AI likely to win in a tactical battle (where they excel, assuming they understand the state of the battle: in an air duel it should be pretty clear. Your height&velocity vs. enemy height&velocity.

But design the plane without the assumption of a squishy human pilot and the thing becomes even more terrifying. Although you probably want lots of cheap AI aircraft vs. an expensive jet trying to keep multiple millions of pilot training alive, so a human in a F22 should beat an AI fighter in a dogfight. But in reality it would face a swarm of AI fighters.

benjamin shropshire • June 25, 2021 12:57 PM

One huge advantage that AI has over humans in a dogfight: a computer won’t black out in a 20 Gs turn.

OTOH: A long range missile engagement is more or less already an attack piloted by an AI. I’m not sure how relevant an AI in a dogfight will be.

John • June 25, 2021 3:32 PM

@benjamin shropshire
AI fighter ought to be capable of taking down more than one enemy.
Missile – destroys target. But is also destroyed in the process. So only practical if the missile cost < target cost.

Clive Robinson • June 25, 2021 3:59 PM

@ benjamin shropshire,

One huge advantage that AI has over humans in a dogfight: a computer won’t black out in a 20 Gs turn.

I would not be too sure on that, avionics has a history of going wrong in high G situations.

At the end of the day a computer is a mechanical object and subject to the laws of physics. The glue that holds the PCB tracks/traces down is a chemical-mechanical bond. They are nowhere near as strong as people think they are and degrade under all sorts of circumstances sometimes in the oddest of ways.

I used to “wear the green” and part of my duties was to repair radio and crypto kit that “squadies” had broken in one way or another “in use” in the field, so no blackout G-Forces (I got the job because I was to bl@@dy good at fixing on the bench rather than “return to stores”).

A frequent failure was “intermittent faults”, which were often –incorectly– called “dry joints”. The actuall cause was mechanical shock causing soft metal to fracture be it solder joints or copper traces thus creating a very very weak contact connection… So even computers go funny and recover in even quite moderate G-Forces, humans would not black out at. Because we are by and large “self healing” which computers are in no way capable of, G-Force has a cumulative effect on computers but not so much on us.

@ TimH,

There are drones and missiles. What is the point of a fighter jet anyway?

Simple answer two reasons,

1, To get the petro dollars back.
2, To launder tax dollars into the pockets of the chosen few.

Drones are cheap, because they don’t carry large humans and the even larger life support systems they need. Also humans like to come back but drones care not if they do or not. That is they don’t give a fig if they get splashed, it in no way effects their abilities in the combat zone.

Oh fighter jets have another issue “limited range” most drones have two to three times the range… So how about drones carrying missiles?

Well the maths says 1-billion fighter jet -v- 0.5million drone with 0.25million missile… sunk cost of 1billion to the side not playing with drones. It’s a numbers game and USAF Generals darn well know it, hence those shipping containers outside of Vegas.

The Chines and Russian’s know it if you look at their weapons development.

The only use for fighter jets in war now is “multi-roll” where they are in reality “small bombers” against ground forces and the like of second and third world nations where US Hawks want to “Bomb them back to the stone ages” for having the temerity of not wanting their natural resources or way of life taken from then by US Corporations hell bent on destroying the planet as fast as they can.

Blue4 • June 25, 2021 4:09 PM

….have you noticed all the computer-contolled, fully automated, driverless automobiles dominating U.S., China and world steets & highways ??

Me neither, but such vehicles have long been predicted — and actually claimed to be a practical reality in last few years.

All these alleged “driverless” vehicles failed dangerously when used in actual real world operational conditions.

And combat jet fighter operations are an order of magnitude more complex than driving a car in city traffic.

JonKnowsNothing • June 25, 2021 7:24 PM

@All

Last week (06 18 2021) there was a report of an attack on “computer-vision based object detection systems” stabilizer sensors, creating what used to be known as Camera Shake.

If the attack works as described and is that easy to carry out, you won’t need armaments to down the AI Fighter, just a blast of R&R or Heavy Metal or maybe the leitmotif from Die Walküre:

  • “I love the smell of napalm in the morning”

The attack works by disrupting the camera stabilizers sensors, the images is distorted and the object recognition program has significant “failure to detect”.

The researches claim that the sample attack can use different methods to achieve the same distortion:

  • audio signals, ultrasound, visible light, infrared, lasers, radio, magnetic fields, heat, fluid

The same attack method can be used against multiple types of sensors.

I’m not sure the reference I posted last week made it past the Ladezudu or if there was another problem with the post (happens and happens).

Search:

  • Poltergeist [PG], Ubiquitous System Security Lab of Zhejiang University, the University of Michigan’s Security and Privacy Research Group, AMpLe, “injecting physics into adversarial machine learning”

===

ht tps://en.wikipedia.org/wiki/Image_stabilization

Image stabilization (IS) is a family of techniques that reduce blurring associated with the motion of a camera or other imaging device during exposure.

Generally, it compensates for pan and tilt (angular movement, equivalent to yaw and pitch) of the imaging device, though electronic image stabilization can also compensate for rotation.[1] It is mainly used in high-end image-stabilized binoculars, still and video cameras, astronomical telescopes, and also smartphones. With still cameras, camera shake is a particular problem at slow shutter speeds or with long focal length lenses (telephoto or zoom). With video cameras, camera shake causes visible frame-to-frame jitter in the recorded video. In astronomy, the problem of lens shake is added to variation in the atmosphere, which changes the apparent positions of objects over time.

ht tps://en.wikipedia.org/wiki/Ride_of_the_Valkyries

  • The “Ride of the Valkyries” (German: Walkürenritt or Ritt der Walküren) refers to the beginning of act 3 of Die Walküre, the second of the four operas constituting Richard Wagner’s Der Ring des Nibelungen.

ht tps://en.wikipedia.org/wiki/Ladies_Who_Do

  • Ladies Who Do is a 1963 British comedy film

(url fractured to prevent autorun)

ADFGVX • June 25, 2021 7:29 PM

Blue4 • June 25, 2021 4:09 PM

…
And combat jet fighter operations are an order of magnitude more complex than driving a car in city traffic.

Weather • June 25, 2021 5:41 PM

Its 12-14 ges with a g suit, I thinks its down to 6-8 with out one.

Air Force commies.

Fake • June 25, 2021 7:36 PM

Yeah well when a 22b usd plane is shot out of the air how much down time do you have before your 14 years of flight school is reloaded into a cockpit.

Uplink wins, nobody said it was 100% onboard.

I deleted some significant responses here there are some delightful takes but if turkey can fly 100usd drones the only question is

How do you get into a theatre when the other person’s military industrial complex is bigger?

This is why our politicians have it all wrong,

Deny deny deny is about theatre not men’s bathrooms.

RealFakeNews • June 26, 2021 5:57 AM

Simulations? In that case, it’s actually unlikely to win all the time.

The AI makes decisions in large part on probability. In any given BFM engagement, there are only so many moves you can make.

As I understand these systems, they use optical input (scene processing) to make a determination as to the maneuver that has been started by the adversary. The AI has the benefit of not only seeing the long(er) term trend (sudden high g pull-up) but the preceding event of which way and how much the elevons moved before the aircraft even starts pitching.

This buys the AI some 2 seconds to counter a move.

Where the AI has a weakness, is it may react to a condition, but then be unable to spot/counter an unexpected move, such as a pilot making a calculated decision to do something stupid, like dump all their speed, to force an overshoot.

If the human now does something else, like spamming flares, the AI is going to process this as an attempt to spoof IR missiles, but in so doing, would this be enough to cause the AI to make a mistake?

Is there a BFM equivalent to sticking white tape on the road making it think there is a wall ahead?

NoJoy • June 26, 2021 8:31 AM

@RealFakeNews —

… the relevant fighter pilot jargon/acronym here is ACM, not BFM.

and real world combat operations rarely involve just one jet fighter against one other — think multiple good guys against multiple bad guys — and how that severely complicates the AI problem.

MikeB • June 26, 2021 10:39 AM

@john:
“Missile – destroys target. But is also destroyed in the process. So only practical if the missile cost < target cost."

… or if the value of what the target might destroy (perhaps the target is an attack aircraft heading toward the White House or Pentagon) is greater than the missile cost. In such a scenario, it would be better to waste a dozen missiles, regardless of cost, that to fire one missile because of cost considerations, and have it miss.

Jon • June 27, 2021 11:02 PM

Real-world example: Look at how good “computer-assisted players” (aka cheaters) do in online shoot’em’up games.

And I’m not talking about “transparent wall hacks” or anything like that – just aimbots and their ilk – and see that they do so fantastically well that humans literally can’t compete, and so get banned.

In fact, those cheaters who don’t want to get caught will often deliberately make their bots miss sometimes, to make them look worse: aka human.

The first army that puts that into the battlefield: Well, the humans won’t be able to compete.

Yeah, there are technical issues: Not all hardware is built to handle 20G’s of acceleration*, and there’s decision time: “Is that 737 an innocent passenger jet or a refueling tanker?” but they’re just development. The research has already been done.

AI is going to win this one.

J.

  • Incidentally, I was peripherally involved in a project that had to withstand 20,000G – Fin-steered tank ammunition. The acceleration in the barrel is roughly 20kG (and, for grins, the round de-celerates at about 2kG when leaving the barrel!)

ADFGVX • June 27, 2021 11:49 PM

In fact, those cheaters who don’t want to get caught will often deliberately make their bots miss sometimes, to make them look worse: aka human.

That’s what happens when Air Force commies try to railroad the whole system of local state and federal courts in the same town.

Leave a comment

Login

Allowed HTML <a href="URL"> • <em> <cite> <i> • <strong> <b> • <sub> <sup> • <ul> <ol> <li> • <blockquote> <pre> Markdown Extra syntax via https://michelf.ca/projects/php-markdown/extra/

Sidebar photo of Bruce Schneier by Joe MacInnis.