Flawless Victory! AI vs F-16 Pilot - Is The End Near? | DARPA Alpha Dogfights Reaction

Former F-16 Pilot Reacts to the AI vs Human F-16 Pilot Alpha Dogfight Trials by DARPA. Is the end of the fighter pilot era truly upon us? Was Elon Musk right?
Mondays with Mover
Looking for a good military/espionage/aviation thriller novel?
www.cwlemoine.com
Donate to Lester's Recovery:
www.gofundme.c...
Contact me:
www.cwlemoine....
Donate to Folds of Honor
www.foh.org
Information on becoming a fighter pilot
www.makethemte...
0:00 AI vs F-16
00:01 Intro
1:10 Briefing
3:42 Fight One
5:35 Fight Two
7:14 Fight Three
8:30 Fight Four
9:33 Fight Five
12:22 Debrief
13:16 Be in the cone
14:43 Positive results
18:02 AI Perfect State Information
20:20 AI had 360 views
22:08 Collision Avoidance Logic
24:40 Takeaways
27:20 Closing
#DARPA
#AlphaDogfights
#AIvsF16
Views presented are my own and do not represent the views of DoD or its Components.
The appearance of U.S. Department of Defense (DoD) visual information does not imply or constitute DoD endorsement.

[@havocmwf]

why is he wearing a mask for a video call??

[@CWLemoine]

kzbin.info/www/bejne/h2jMaoWimbeLd8k

[@James-sz5bi]

@@CWLemoine Wow. The one question I had after Mover explained everything else - and Mover explains it here. If it wasn't for this channel I would have just seen the headlines and have really understood what happened. I think I am going to have to buy one of his books to say thank you.

[@SymeonPhronema]

@@James-sz5bi His books are awesome man. I highly recommend them.

[@matthewsheeran]

Shared video equipment in the office which as a result wont get contaminated or need constant cleaning.

[@philipuwumarogie9796]

@paul williams It would help if you simply clicked on the link CW posted and saw the actual explanation.. innit?

[@heronsystemsinc.6849]

Mover, the folks at Heron Systems here. Thank you for the thorough and honest review of our performance. You raise solid critiques about the advantages our AI had and how Banger was out of his best element. ADT was a chance to prove that AI could handle the tactical decision making in a semi-realistic environment . We're pushing it into a real flight environment next. Plan is to start with small scale drones (literally training the agent on this flight model now) and scale to representative fighter jets. We'll also incorporate a more realistic sim that models both sensors and weapons engagements with high fidelity. We'd love to chat if you and your followers want to learn a little more about this technology and where it is headed, including opportunities for simmers to take a shot at Falco.

[@CWLemoine]

Definitely something we can chat about. 👍

[@lippertwe]

Heron Systems Inc. not asking to be sarcastic or obnoxious. just wondering how this experiment is different than a typical player vs ai flight combat game from microsoft or dcs or anywhere else? flight combat games of player versus computer have been around for almost as long as video games themselves. and not arcade style ones but games that tried to have some realism.

[@karlmeyer5383]

Super classy answer. Would be very interested in some folks from the DCS community to take on the AI

[@ajkramer35]

I have to respect this response. I'd love a shot at this AI given some of the perfect SA removed.

[@Joshua-xz7pm]

I appreciate your honesty and willingness to interact with the community, and I believe there is value in constructive and honest feedback. If you want a sim pilot to give it a go, count me in and I may have some friends who will would join as well.

[@rammphillips3002]

Mover, no kiddin', this is one of the best videos you have ever made. And, as a former Navy fighter pilot, I couldn't agree more with your takeaways. When I was flying, all we used the Sim for was procedures, learning switch locations, etc. It's not bad to fly IFR approaches, also. Banger did the best he could but the deck was stacked from the git-go. I'm gonna go watch it again... Great job! Thank you!

[@PhillipBirmingham]

I haven't seen a lot of Mover's videos, but as a reinforcement learning (RL) developer (and sim pilot,) I agree. This is an interesting demonstration, but the omniscience concerning problem state, and the shortcuts taken in gunnery (since face shots are low probability, it takes a lot longer to train a pure RL agent to take them correctly -- that's what the developer meant by "sparse") definitely limit its applicability. That said, RL is scary in how it can learn rules we wouldn't even think of. I won't be surprised if, once the simulation approaches the real world, face-shooting becomes the norm.

[@kyle857]

I knew pilots and their advocates were going to claim that the deck was stacked against the human pilot from before the event. Thanks for proving me right.

[@martinmartinez250]

I agree with Mover. The test designers made a serious mistake in not simulating the sensors a real AI would be utilizing. I have designed real aircraft Fire Control Systems and programmed game engines and more recently worked on AI. This seems like a poorly planned demo where the human is given imperfect information while the AI is given an unrealistic perfect information. Not surprisingly the gamer with perfect information always wins.

[@MistedMind]

@@kyle857 Even the programmers admitted that. So what's your point?

[@isodoubIet]

@@martinmartinez250 " The test designers made a serious mistake in not simulating the sensors a real AI would be utilizing." The point is not to be super realistic. It's to make a simple first test with an easier problem.

[@ConorChaos]

"Trigger, show em' how it's done."

[@constipatedparker5879]

"That's right, Trigger. They're just machines. Don't show 'em any mercy."

[@bombomos]

"Stick with Trigger and you'll make it through."

[@trigger5013]

...

[@Mobius118]

Sup children

[@reke9942]

Wilco

[@afterburn2600]

Let me fix the thumb for ya: *Former Cessna Pilot

[@CWLemoine]

ACKSHUALLY, for this subject matter it should be: Current DCS TOMCAT! Pilot

[@maxwellclark6992]

C.W. Lemoine TOMCATS

[@hoilst]

@@CWLemoine TOMCATS!

[@scheldon2244]

C.W. Lemoine TTTTOOOOMMMCCCCAAAATTTTSSSSS!!!!

[@jerrybandy3827]

I'm not a real TOMCATS! pilot but I play one on TV.

[@btbarr16]

The part where he mentioned not getting any feedback from the aircraft reminded me of something. Recently I was watching a NASCAR race where they were at a track, a road course, they've never been to and because of COVID there wasn't any practice/qualifying sessions before the race. Nascar teams have simulators for their drivers. A driver was being interviewed and asked how helpful a simulator is. Basically, the driver said that they're helpful for learning the track layout, but not much else since they can't feel what the car is doing underneath them and they don't include things like bumps, different smoothness levels of the asphalt, track temp, shifting weight of the car, vibrations from the brakes, and so on.

[@Paul-hy6rp]

Glad I watched your coverage on this dogfight, there has been a lot of rubbish spoken on this matter online and your video put it all in perspective, thanks for covering this. Fighter Pilots are going to be around for a long time to come.

[@mwara2444]

The whole point of AI other than saving pilots, is that you can do different flight characteristics when pilots don't have to undergo g-forces you can design planes differently. Remove the pilot in the g-force and you can create vehicles that can do some crazy stuff to get kills. The hard part is having an AI that can identify predict decide and execute, while maintaining awareness of everything going on around you. But if we ever need fighters in space AI is going to be extremely handy for extreme speeds and maneuverability. I must have develop sci-fi inertial dampeners The human brain is still the most complex computer on the planet.

[@bobdillon1138]

Yes but not necessarily seated in air frames.

[@eitleoir170]

This is a case of, "maybe so sir.....but not today".

[@Jester-uh9xg]

​@Benjamin McCann I think the only aspect of this that seems impossible ten years ago is having that VR headset

[@markvincentcocjin]

Tom Cruise lands jet as the enemy drone tries to lock in. Tom Runs as fast as his Tom Cruise legs can take him. Enemy AI is like: "WTF is that? Cannot compute..."

[@markvincentcocjin]

@Benjamin McCann Ten years from now, an F-35 launches a drone to dogfight another drone with energy weapons.

[@nickbryan217]

14 year Army Armor Officer, spent a lot of time in simulator both gunnery simulators (AGTS) and the Abrams Simulator (CCTT). You lose a lot of situational awareness in simulators. Simulators are great, but they are limited in what they can do to simulate the tactical environment. You often either feel like you’re looking at the world through the straw or you’re not hearing 3 dimensional. So if you hear gun fire it’s in stereo and you’d have no idea what direction it’s coming from. So you’re scanning, getting shot, and hopefully you pick up a visual signature or you’re just dead.

[@HormigaZ]

I don't want to sound out of place and I'm sure there's caveats to what I'll say but in gaming we are used to surround sound and the current virtual reality goggles are pretty damn impressive at recreating spaces in 3 dimensions too. I know from some former or current military pilots that often their simulators are objetively behind technologically than the some of the commercialy available ones we use, although we often can only work with old technology.

[@nightofthunder5509]

They are based on rules and rules can be broken or bent that is why you will always be faster than they are. Morpheous. Perfect example why and how bots can be beaten they dont have the ingenuity plus most air battles will be BVR anyway so this proves nothing even if the Ai won a dogfight

[@kyle857]

@@nightofthunder5509 The AI will be better BVR too.

[@briantaylor8197]

I agree. The simulator is much different.

[@isodoubIet]

@@nightofthunder5509 " plus most air battles will be BVR anyway " How are missiles guided?

[@dbrady1966]

This reminds me of when the Air Force decided to build fighter planes without a Canon because missiles made them obsolete

[@halcionkoenig243]

Which was a spectacular success, which is why all planes after never needed guns ever again... until the F-35, which replaced both the A-10, and F-22! /s

[@level5565]

@@halcionkoenig243 wrong!!

[@deniswitschi1859]

@@halcionkoenig243 are you drunk?

[@theboothy91]

@@halcionkoenig243 but they all have guns..

[@ClingyCrab]

@Halcion Koenig Do you are have stupid

[@mastersheff37]

Flawless victory... it's even the same font from Mortal Kombat.

[@Real28]

I came to make this comment. I'm glad I'm not the only one who noticed this. Means someone anticipated it and programmed that. Cheeky.

[@kaesees]

Interesting commentary on high-aspect shots in RL BFM training. The RL French fighter pilot "Ate" was extremely aggressive and took a lot of high-aspect shots when BFMing against Growling Sidewinder in DCS. I wonder whether that was a French/US training difference, or Ate trying to exploit the characteristics of the delta-wing jet he was flying where he has a great first turn and a tougher time thereafter.

[@lucasknighten4626]

Great breakdown and analysis on the DARPA contest. Really appreciate your feedback.

[@tylermassey5431]

18:15 The AI had perfect state information? That negates the entire presentation.

[@Antares2]

Imagine if you made a tiny AI-controlled plane that instead of a gun had an explosive charge! And then it tried suicide ramming itself into (or at least close to) enemy planes and detonating? And because it's so small and AI's don't pass out, you could make it super maneuverable (like 30g or something) It could maybe use infrared or radar sensing to find its target! I know! We should call them "Air Intercept Missiles" ... or AIM for short. Almost feels familiar...

[@nikolatasev4948]

To be fair, the "AI" in most missiles is so basic that it does not deserve to be called intelligence at all. You can see it in DCS - missiles are leading the aircraft in all directions and thus ramming into the ground when the enemy dives for example. Lacking image recognition it can't tell apart flares/chaff and plane. A missile with a smart phone's worth of AI would be much deadlier than current ones. I'm not sure why this is not being developer further. I mean, it might be classified, but I don't think it is.

[@Antares2]

@@nikolatasev4948 Well, I am pretty sure that the newest types of anti-air missiles are significantly more advanced than in DCS. DCS is a game. A good game, but still a game. We know that AIM-9X has imaging IR and should be extremely difficult to spoof with a flare as it isn't simply looking for the brightest heat source, but rather recognizing the shape of the plane as a thermal image. This is also why modern missiles are far less likely to lock onto the sun and such as they did in the Vietnam war. But what kind of tech that goes into the newest models of AMRAAM is most likely top secret.

[@alivewithpassion]

@Antares they have that technology, for example look at cruise missiles or smart bombs. The guidance systems are on point and developing; the military has always and will always be on the cutting edge of technology.

[@nikolatasev4948]

@@alivewithpassion Cruise missiles have inertial and GPS guidance. "Smart" bombs are contracted to "dumb" bombs because they can correct their fall towards a target marked by GPS or laser marker. None of these is a dog-fighting AI like the one showed here.

[@hippoace]

@@Antares2 www.thedrive.com/the-war-zone/14344/heres-the-definitive-account-of-the-syrian-su-22-shoot-down-from-the-pilots-themselves Things will fail when least expected. The AIM-9X went stupid even from a 6 o clock launch.

[@miatacollector]

Plus a pilot with a swarm of heron driven uavs would definitely be a killer combination.

[@luarbiasawaras8700]

"Kill the mother ship first" will become number one strategy

[@nespolo1597]

Santai Sehat I don’t think heron needs a mother ship to survive if anything the mother ship will need heron to survive

[@luarbiasawaras8700]

@@nespolo1597 so the pilot more like a liability not an asset

[@michajastrzebski4383]

@@luarbiasawaras8700 might be hard if said mothership stays BVR in a stealthy aircraft

[@ProdigyXI276]

@@michajastrzebski4383 stealth would mean immune - other stealth aircraft can pose a threat to a "mothership"

[@FiveStar68]

The age of fighter pilots will never end Mover!!! Like you said there always has to be two-way communication. ALWAYS!

[@stal2k]

So in summary it's a great aimbot for faceshots :)

[@skipker]

I'm not sure why anybody commented negative. You are a teacher that always explains the events here in a way we non fighter pilots understand. Awesome job

[@Altoar]

Definitely agree with the points brought uo here. In addition though, the AI is missing the ability to assess in sooo many other factors that can spring up unexpectedly. Damage to the AI's aircraft affecting performance, for example. Don't even start on the more abstract decision making that pilots may have to make on things like "I'm not gonna use this armament right now even if it might be easier, because I'm probably going to need it later on in this sortie/mission/flight/whatever the correct term is.

[@petitio_principii]

It's only a matter of time until unbeatable AIs are developed, even without jets made with no regard to human pilots in terms of g forces and whatnot. Nearly all the imagined human superiority is not beyond the reach of AI, people used to imagine similar rationales when imagining AIs would never beat humans at chess. It's not really fundamentally different from chess, though, or non-ply games where AIs quickly learn to play in ways that humans never even imagined possible. The most significant constraints are in the sensory equipment, which isn't something that's inherently superior in humans/animals, it's a harder task than AI with a "granted" awareness (even if limited by that of conceivable current sensor equipment), definitely, but also just a matter of time.

[@Mich_Angel]

Some really good points you make. IMO. This was like having two paintball team fight in a arena and one team know exactly where their opponent are at all time and the other team only if seen visual. Or like a election where one is already set with all information to win and the other not and have to get it as he proceed, it set a unwinnable situation against the one with all info. Kind off a pointless test IMO sure it show that Ai have evolved and can perform if all parameter is known, let it learn the parameter as it come in to a hostile situation and see how that go. But then we already seen unbeatable Ai in DCS if you push it to max and human pilot need to stick to the rule-book of engagement, so what's new about it, it really don't show anything. Except that combat flight simulator games are very fun.ha ha Great evaluation of the video Mover, thank you. CHEERS!

[@Deuce_Dufresne]

I’d always shoot the other guy in the face, so to speak, when we did “cage matches” in the simulator. Definitely not applicable to the real world though like you talked about. I can vouch that Banger is an excellent fighter pilot 🇺🇸😎🥃

[@cirrusaddict2552]

Mover - great video, loved your critique of the 5 engagements. You're spot on regarding flight simulation. I challenge anyone to perform a PPL checkride in a Cessna after practicing on a flight sim, let alone performing BFM in a tactical jet. Landings, slow flight, stalls, steep turns...all require the feel you are talking about. Sims are wonderful at teaching you instrument scan, not actual flying.

[@DaveFMillard]

Many years ago my step son bought me an hour in a 172.. he told the instructor pilot that I had never flown an airplane... only simulators. Long story short.. after the pre-flight chks he let me start it up, taxi out, take off fly around for 40 minutes and land. He never touched anything other than the mic. Next year he bought me another hour... I beg to differ with your statement.

[@cirrusaddict2552]

I'm glad to hear you enjoyed the discovery flight. I was referring to a checkride and an actual license

[@Wes-xk6hl]

I suppose I could see the day where an AI in real life could have full situational awareness by syncing with third party radars and whatever else fancy stuff but we're pretty far away from stuff like that. So that one thing alone makes this a ridiculously loaded test. What I see out of all this AI stuff is what u said. It's how we can make our guys better than they otherwise could be. For instance, the whole shooting accuracy thing is actually fairly easy and well within our reach with current technology. The day where our pilots are that accurate with guns only could easily be right around the corner. But they're still pulling the trigger and everything is the same.

[@secretsquirrel8010]

Radars are old school IMHO. Link an AI to satellites and you’re set. This is the future.

[@MrMattumbo]

I mean you can stick cameras all over a drone, the problem is having computer vision algorithms that can process that data and not only identify the target at close quarters but glean useful data from that like energy state, relative position and nose angle, etc.. then process that to extrapolate the enemy's future possible positions and devise its own counter moves with the aim of getting accurate gunshots. That huge amount of raw sensor data processing on top of decision making calculations for itself and the enemy is going to require some serious computing hardware and cutting edge machine learning algorithms if it's going to work in a real aircraft in real time. I think the human brain still wins for now, but the AI is on track to match it within our lifetime. Ultimately it will have to if frontline combat drones are going to be a thing, its too easy to jam drones that rely on external control/sensors. Boeing's loyal wingman drone gets around this by being close to the controlling aircraft and being more basic in its functions, pretty much just being a flying missile rack/decoy for the human plane. It's a long way off from winning dogfights or even just operating by itself.

[@Wes-xk6hl]

@@secretsquirrel8010 well u would have whoever is operating the radars identify which aircraft are hostile. And "radar" is just a generic term. Doesnt matter what the technology is (tho satellites are very likely not the answer). My point was simply that, with AI, it doesnt need to be a self contained system. It can be using data from all kinds of sources. But any kind of wireless link is a security threat and, like mover said, there has to be 2 way communication. U cant just give it the keys and say off u go do whatever and we wont know whether it's going well or not. No matter what, these things happen in phases. Everyone who said autonomous cars were just going to pop up and be the next thing are just as stupid as whoever is saying that will happen here. It will go through many phases of helping the current pilots before we see anything like that.

[@nelno]

@@MrMattumbo this is the other side of the coin that people rarely recognize. In most simulations, the sensory data you have is perfect, and an AI can exploit that to make perfect shots. To more closely model reality, the AI needs to be fed its sensory data through simulated sensors that have all of the same quirks and shortcomings of real sensors. The compute is available, though. A smart phone SoC several generations old has enough compute to visually identify and track hundreds to thousands of moving features while rendering entire VR scenes. But in a real world scenario there are many more variables, ranging from difficult to impossible to sense/predict, that need to be accounted for. The fact that this simulation only required the opponent to be in a cone to score a hit really makes it clear that this was a DARPA marketing event.

[@nocare]

Data from any source other than extremely close friendly aircraft would be too delayed to be useful in a dogfight. This sort of sensor fusion would be more useful in longer range missile engagements. Now getting the level of accuracy required from vision alone in bright daylight is probably not a hard problem to solve. In fact I bet today we have good enough machine vision systems to get the level of accuracy as depicted in the sim. The problem is what happens when you are low light, or there are clouds, flares are going off, some of the cameras may have been damaged in the last pass. It doesn't even take that much computing power because you just use low FOV high res cameras on a gimbal and track the target and use a lower res wide field of view cameras for augmentation. Most AI problems in modern days is solving edge cases. It's why autonomous driving still isn't full fledged. It can handle roughly 94-98% of driving scenarios as of last year. However since most humans can handle 99% of most scenarios its not better than use yet and thus has not replaced us yet.

[@anthonykaiser974]

Regarding your closing comment about having a human decision-maker in the loop, when I was a LT at Field Artillery school in 1995, our old Battery Computer System (BCS) had a function called "Automatic to Pieces" coded ATP, that would allow calculated firing data to be sent to the guns without operator interaction. Our gunnery instructors (mine was a USMC Capt) referred to it as "Automatic to Prison."

[@GriffMJ]

.... put "GrowlingSidewinder" in the seat up against the AI and I bet that the AI loses by 4 to 1. 100% Sim Fighter pilot would destroy the AI.

[@LtDogoop]

Agreed

[@toddie4usa1]

Nope the AI will not lose because the AI is not limited

[@nonamesplease6288]

Growling Sidewinder did a few recent videos where he went up against a real Rafael pilot. The real pilot lost the first round or 2, but once he got the hang of DCS he owned GS. Check it out. So how much time did this pilot spend on the simulator?

[@GriffMJ]

@@nonamesplease6288 ..... yes up against "Ate" the French/Canadian Rafael Pilot (who was flying the Mirage). "Tim Davis" the ex-Tornado pilot flying the F5 sim for the first in DCS up against the ace AI was also great to watch (second flight). So give the real fighter pilot the hours etc.

[@brian2e]

No because "growlingsidewinder" wouldn't be able to show only the engagements he won

[@hakadah]

Spot on Review Mover. I couldn't agree more with your assessment.

[@BuggSmasher]

And we expected Skynet to deliver us pizzas. What fools we were!

[@TheWatson83]

Was waiting for Sarah Connor to bust into the guys room from Heron systems and start shooting while he was talking haha

[@steak5599]

If Skynet is real, who's side would you be fighting for? AI or Human?

[@BuggSmasher]

@@steak5599 I don't think a.i. will be in recruiting mode ! It'll be in "Uzi" or "40 watt range" mode ! You see, it was never really about John Connor, but more about an everyday retired fighter pilot "C.W. Lemoine" to lead the resistance. He's educating us about the dangers of ai right now, and right under the very nose of the enemy ! Man He's good !

[@operator8014]

So, this was literally a very aggressive video game AI stuffed into a flight simulator. Such wow.

[@dutchflats]

Re: face-shot effectiveness, checkout Cunningham/Driscoll vs. Toon initial merge (F-4 v Mig-17) on 10 May,1972 over North Vietnam and the consequence it had on their fight.

[@dutchflats]

@, I've heard that as well, but not sure which story is the truth? I guess the point is, the Mig's face shot forced the F-4 onto the defensive twice before Cunningham was able to out think the guy whoever he was, kinda fire for effect.

[@ti1ion]

I don't know if you are exactly correct, but I love this story. It is one of my favorites. Together with a bunch of other first person narratives they go to show that life, and success/victory is often only possible with a healthy dose of luck (or fortune). This dogfight involving Cunnigham and Driscoll is one example of skill mixed with dumb luck. Listen again to Randy Cunningham tell this story in the "Dogfights" series and you will hear him chuckle at his stupidity. When they saw the MiG-17, Cunningham told Driscoll (and I am paraphrasing here) "I am going to scare the crap out of his guy. I am going to go right by his canopy." Cunnigham intentionally put his Phantom on a direct course for the MiG's nose in a game of chicken. It was only after tracers started flashing by the cockpit that Cunningham (with a chuckle) thought "oh, yeah, he has guns!" And in the "Dogfights" interview he also says he made a similar mistake again as they merged and went vertical. Discovery Channel has pulled the videos off KZbin, though, so the interview has to be watched somewhere else. So, the MiG's gun was not as important in the story as Cunningham's adrenaline fueled cockiness in the heat of battle.

[@ti1ion]

@ I wish. Watching them all on demand would be awesome. There is a much shorter version of the encounter on KZbin, from Cunningham here -- kzbin.info/www/bejne/pZatfGCFmMdjhKM -- but it is not as good as his later interview for the "Dogfights" series.

[@andrewhall2411]

I see the bigger (scarier) issue being this... When we just produce jets as AI kamakazi vehicles designed to direct impact. Remove the design constraints for longevity, pilot life support, holding munitions (outside of maybe some missiles) when we don't care about the loss of a pilot (because there isn't one) and we don't care about loss of equipment (because it is designed to be disposable) then things get scary...

[@confuse3671]

The AI had perfect information about the adversary - meaning the AI was feed exact position, course and speed. Its unclear if the shots even took into account bullet ballistics. The human pilot had to make estimates about what the other plane was doing based on simulation/animation. Bottom line, they created a game that is visually similar to real life but distinctly different. BTW, yes, if you feed a computer all the information, it can make high aspect angle shots. In real life, even an M-1 Abrams is only 90% effective first shot with laser range finders. In the AI vehicle control world, the trick is sensors. You need to understand what is going on outside the window.

[@ReflectedSimulations]

Thanks for sharing your thoughts. Some Interesting and valid points there.

[@ericlawrence9729]

Great context Mover I think you are spot on, would have loved to see how the AI would handle an aggressive maneuver like Colonel Boyd’s flat platting or squatting the jet to force an in close overshoot. Love the channel

[@njgrplr2007]

The best athletes react to stimuli instantaneously and without much conscious thought. I could see where fighter pilots, who have honed their skills in the real world, are at a significant disadvantage when they're put into a simulator that doesn't provide the same feel.

[@billcarlin7411]

Mover, I'm a big fan and a man in my sixties. One thing I have had the joy to witness is the technology exponential curve. We must be careful not to believe something won't happen in a quicker amount of time than it has taken in the past. My guess is that the only thing holding these developers back is the speed in which they can process information. I had one of the first MS flight simulators when it came out back in the early eighties. Back then all you needed was a joy stick and a vivid imagination. You only stopped playing when you had to answer the phone in the other room. Now you can fly over your own house in a variety of aircraft. These AI guys seem to thrive on the "make them tell you no" mantra. Great video by the way.

[@Baleur]

7:40 wtf why didnt he shoot? Did he not see the jet? Im starting to think maybe this was an issue of having low resolution vr goggles? Sometimes he's flying almost as if he's blind and cant see the enemy AI..

[@austinmunson5034]

Was watching this last night thinking, now this is gonna be all over Mr. Lemoine's refrigerator door.

[@Aiur89]

Love the content your making man. I may not know much about this stuff but your arguments are always within reason and a lot of fun to watch.

[@carloscastanheiro2933]

Your feedback analysis was amazing. Thank you Mover.

[@thekarmanline3748]

Next week: “DARPA reveals their new Arsenal bird concept”

[@unlovedone5639]

Darpa hurting the american people

[@DavidWilliams-nq6fj]

Really insightful comments. You have to be impressed, but acknowledge it for what it was. Nice job Mover.

[@sox5131]

Total situational awareness = the A.I. cheats.

[@Ananamitron]

Would not an A.I. in real life be able to do the same thing? Multiple cameras, melding with data from other intelligence sources such as an AWACs A.I. wouldn't have to limit their vision to two eyes facing forward. I don't think the "era of the fighter pilot" will end soon, but it will end eventually. Laser weapons will become more prevalent, and "gunpip" skills won't really be useful anymore. Instead, a computer will point the laser, fire it, and neutralize the enemy aircraft in seconds.

[@regiondeltas]

@@Ananamitron "Would not an A.I. in real life be able to do the same thing" - Nobody is saying an AI couldn't have access to lots of data, but it is fundamentally limited in real life - both in current technology terms, and in a pure fog of war sense. The AI in this test is given precise, trustable data 100% of the time in real time of the entire battlefield. There is no sensor or suite of sensors that can realistically achieve that. Even with your AWACS AI, it needs to be on station and is still limited by its sensors. Bearing in mind, in the world of said AI fighters the enemy is going to be doing all it can to deploy electronic counter measures - jamming, spoofing etc.

[@killchicken123]

What do you propose to even the odd? Put a box over AI and display exact 3D coordinate, velocity, acceleration, orientation and what not, every data the machine uses basically. It doesn't do much to human other than cluttering the interface because we can't incorporate that data into decision making even if we have those. They don't have budget for more complex and realistic scenario btw.

[@regiondeltas]

@@killchicken123 I mean, the obvious would be to only let the AI "see" what it would in real life. I.e., it can read data from the instruments and it can work visually. Imagine fighting an AI in a first person shooter that knew what you were doing, and where you were, every second. Of course it would win - it doesn't mean an AI powered robot would though

[@Jason608]

@@regiondeltas Take everything you said and apply it to the problem of a missile intercepting an aircraft. An AIM-120C may have imperfect information, and can be decoyed by flares and chaff. Yet somehow it manages to hit its target often enough while sailing through the air at mach 3. The AMRAAM has to know the target's speed, direction, aspect, and other parameters in order to calculate an intercept, all using a weak (relatively speaking) radar in its own nose for the terminal guidance phase. I can imagine an AI driven aircraft could do the same thing in a purpose-built craft with radar pointing in all directions. Just not in a stock F-16. And of course, it could be jammed and fooled with chaff, etc. just like a missile. But it wouldn't be useless, either, just like chaff and ECM isn't 100% effective against missiles all the time.

[@baronvonfaust]

It's an older book now, but "Wired for War" by PW Singer goes over some of the implications for autonomous armed platforms, and in particular I bring it up because it discusses things like legal ramifications: if an autonomous platform engages a target erroneously, what happens afterward? If it shoots down an aircraft it wasn't supposed to, it can cause anything from unintentional deaths to outright warfare and war crimes. And if this happens who is accountable - the state, the CO, the software coder? These are really important questions when it comes to fully autonomous weapon systems. I hope this is interesting to some of you. Thanks for the videos CW, love the content and you discussed a little of the above at the end, just thought the book might be of interest. All the best.

[@jefsonb]

Very good points! Regards from an older F-5 pilot!

[@MarianneD1952]

Mover....made a donation to our Dear Lester!! Tried to make it a number that fits the helicopter. Hope he gets it!! Thanks for the link!! ♡♡♡

[@SogenOkami]

The issue I have with AI, and I've seen this a lot with the deployment of bots in video games in the past few years, is that once you go at it with the mindset of playing the bot and not the game, the bots tend to fall apart really quickly. OMG 28:47 would trigger all the DCS Facebook nerds way too hard. lmao

[@LetsKeepThePeace]

100% agree.

[@ZefferGaming]

Where's Maverick when you need um? "Hit the brakes and he'll fly right by"!

[@nopus1]

In my naivity I presumed, that AI was real combat oriented (or, at least, real AI) and get its input from cameras and sensors. Just a game to aquire DOD money...

[@FlyShamu]

Outstanding vid & discussion! Thanks Mover. Always a learning experience.

[@seatedliberty]

Dogfighting is an art, and only a human being can be an artist.

[@Antares2]

Not really. If you have all the information, I am sure you can make an algorithm that always makes the best choice (kinda like those unbeatable chess computers). But the problem with real life fighter jet combat is that you DON'T have all the information. This AI could see through his own plane and always knew with millimeter precision where the enemy was, and how fast it was moving. With all that data, you don't need creativity or luck (or artistry), simply a calculation of how to maneuver to achieve your goal, which is to point your plane at the enemy. Also, the fact that this AI doesn't care if it crashes into its opponent and only has to point its plane directly at the enemy to score "hits"... well, it makes it fairly unrealistic. If you had set two of these up against each other, I have a strong feeling they would be doing head on jousting passes, colliding every time.

[@TheMechanicalHermit]

Two debatable statements.

[@jwaustinmunguy]

Fighter-on-fighter, the day of the fighter pilot is far from over. My dad was in the RCAF through the fifties, sixties and into mid seventies. He once told me that interceptions could be more effective with data-linked fighters against subsonic bombers (Bears and whatever else had the range for your northern SAC bases). I don't know a lot of the details and he has been gone for twenty years. There is so much stuff he could talk about now. As a young navigator he was trained to operate the AI radar in the Mosquito NF at the very end of the war in Europe (May 1945) and as an old Colonel (1977) he was really impressed that the AWACS worked so well. Quite an interesting career.

[@paranagua1960]

Very well covered Mover. Thank you for sharing your analysis.

[@0280530]

Excellent review Mover. AI will have a significant amount of influence in the future, but this was a demonstration that gamers would appreciate, not pilots.

[@matthewellisor5835]

I haven't seen Glock in years! Good dude. Thanks for the video. I'll have to give him a call. Greetings from SC!

[@hughgreentree]

"Good against remotes is one thing. Good against the living, that's something else." H. Solo.

[@Suthriel]

Unfortunately, H. Solo comes from an universe, where remotes/droids/AI are usually a failure in most combat situations, and i am not sure, if they are made this dumb on purpose - in-universe, or if it´s just bad writing. Would be more interesting, if the SW droids would at least be granted the fighting abilities from the drones from the movie Oblivion.

[@charlesormond964]

I agree with your assessments of this BFM situation, Mover 👍✈

[@FirstDagger]

21:40 ; Well F-35 has a 360° field of sensors with EODAS.

[@HH60GDriver]

I'm sure 60% of the time, it'll work every time... Just kidding, it's Lockheed, make that 20% of the time.

[@すどにむ]

Kind of but it’s multiple camera stitch so not 100% consistent or accurate

[@kerbalairforce8802]

Infrared or TV guided or radar guided drones are vulnerable to all the ECM that guided missiles are. Sidewinder missiles drop from 80% hit rate to around 50% hit rate when training target QF-16s are equipped with countermeasures.

[@maxvonnitten4213]

I hate this new trend of "this era is over" and bla bla bla. I heard and read a lot of people saying that. They said MBTs are obsolete because of new anti-tank systems, they said Aircraft Carriers are obsolete because of hypersonic missiles, now they're saying that pilots are not "enough" anymore because of drones. For the sake of God, stop that. All of that is just crap, because if there is one thing that we're good at is adapting, upgrading and evolving. Not only it was absolutely unrealistic due to the fact that there was no G-load limit (- 3/+ 9), but also that there were no weapons to actually simulate a fight. Not including gun dispersion, countermeasures, etc. Human pilots will always have a place in combat, the "tin man" can stay at his place. Oh and by the way, I read of people talking about "build a super-G limit jet" (like 25 or 30 G's) and similar stuff. There is one specific reason why we limit planes to a certain amount of G's and that is a structural and mechanical thing. No Jet can pull more than 13 G's without pulverizing itself.

[@JBM425]

The point you bring up about the data link around 25:30 is important. So far, we haven't even encountered a military situation where we have been denied our GPS capability. What happens when data/comms is disrupted? AI can be a useful tool to augment the cockpit, but they aren't going to be laying off fighter pilots anytime soon.

[@kerbalairforce8802]

I could see a head on attack run where the human pilot squeezes the trigger and the AI computer at his fingertips waits until the fleeting fire solution appears to actually fire the gun.

[@foodank_atr817]

24:22 they have machines that look at dozens of potato slices in a split second (for chips) _flying_ by after being dropped off a conveyer belt and recognizes imperfect slices then blows them out of the falling deluge with a precise puff of air onto the reject bin, all with in a second.

[@WildBillKelso32]

He almost seemed to be flying like he was told to lose. That’s my opinion and is not based on any facts. That last kill though. Just drawing lazy circles...🤔

[@toddie4usa1]

He was not told to lose

[@Eddievargas1]

The ai had all knowledge of the battle field not going of sensors

[@WildBillKelso32]

Eddie vargas so he was set up to lose so they could justify spending billions on this AI...

[@dampsok]

In DCS world, when I lose SA, sometimes my default state is to keep what I have and hope I can pick up the target again by making another circle. It is bad practice, but I can definitely see how a person could make that "Hold what you got" mentality

[@WildBillKelso32]

toddie4usa1 I said he seemed to be flying like that and that I had no proof. Were your present at this test?

[@julianrossi8460]

Such a great in depth analysis, thanks for doing your research and presenting such a informed video

[@MIGuy]

"never ask a barber if you need a haircut"

[@rannors723]

Awesome video! Thanks for the clear analysis!

[@dberaupilot]

Honestly don't really see the point of this experiment. How much time did the pilots have to get use to the simulation, I imagine not a lot. I'd also venture the average DCS player would wipe the floor with the AI "Heron", I hope a lot of money wasn't spent on this....

[@theegg-viator4707]

DBflyguy truth

[@daviddale3624]

If the computer has the data from both aircraft all it need do is a mathematical calculation for the manuever to firing position. Reminds me of why we went to "full contact" martial arts in the 70s. Yep, you did score two "points" when your opponent closed the distance before knocking you out.

[@davelister9755]

I have probably close to a thousand or more hours with AI in some competitive video games, and I feel like I can confidently say that AI doesn't think... it just does. NOW, there is a little bit of an exception with Neural Networking. In which case the AI can 'see' the ingame world eal world and with trial and error, solve the complexity of accomplishing a particular goal. However, what I've always found about 'good AI' is that it is just incredibly clutch in all regards to an inhuman level. As mentioned before, it doesn't really think, it just does. So when it comes to beating AI, and I am a candidate for beating this kind of BS AI in my particular craft, you literally just have to set traps. Throw a little bait out there, and if it takes it you get yourself position or something. I mean, translate that in any particular way that you will with air combat sims or what not have you. But it definitely does seem like in this example provided to us, the AI is programmed to just execute really well. It is kind of like seeing if a human can aim a long distance missile better than a computer, which is ridiculous. What you can do as said in this video is you can utilize the AI to help you become the better you by running it through analysis sims, and it might find situations that you missed during your exhibition flight. If at all, the AI will consider variables like airspeed, your angle, and altitude maybe in regards to your opponent's. But outside of that, it probably isn't really that great of a candidate for real life usage. Edit: Imagine a computer using Neural Networking trying trial and error to figure out real life dogfighting, and the massive budget costs of losing that many aircraft for insignificant gains. Until simulations are practically real in every aspect, there is absolutely no way Neural Networking would work without an unimaginable amount of spending. It could be possible however to create miniature AI Planes in real life that would go up and fight each other, but at that rate... you're talking years and years maybe even decades of training. Maybe even never, given how much technology must be advanced, and the limitations of future hardware.

[@ipodhty]

Yeah using video game ai when talking about irl ai doesn't work, since game ai is ment to represent how human's think and act. While if you want true combat ai it can do things that human's just can't compute. The amount of data it could store about possible enemy paths is insane compared to a human

[@Redmanticore]

"you're talking years and years maybe even decades of training. " funny you mentioned that, AI's are cheap to train to levels that are impossible for humans - parallelism across a server farm allows for MILLIONS of hours of flight time experience compressed into hours. when a human loses, he dies and we lose all that training and experience. when an AI dies, it loses only the vehicle. a normal videogame a.i. were always very limited. i am not sure it just Does anymore, watch a documentary on how a.i. deepmind defeated #1 Go player in asia, on youtube, called "alphago the movie"

[@theodore8285]

Sir I have no idea about planes, fighter jets or anything but I really enjoy your channel and videos for a reason haha Greetings from Greece!

[@FuegoJaguar]

For me the fact that the AI has perfect knowledge of everything totally invalidates this test. It’s completely unrealistic and also SUPER important to the fight. Never mind the cone = damage. This is so skewed to the AI that this isn’t a test of AI vs human. It’s a test of this contractors ability to deliver on something with unrealistic constraints.

[@Cecil97]

its sort of like making an aimbot for a tank but in a game

[@User-gs1dk]

"For me the fact that the AI has perfect knowledge of everything totally invalidates this test." This. It's not just super important it's literally EVERYTHING. This test was essentially nothing more than a video game demo where the developers have cranked up the difficulty level to extreme. I'm not trying to be rude but I'm not even sure why this is considered impressive. There has been video game AI that has been perfectly capable of beating real world pilots, especially when they're not that experience on said sims, for a pretty long time. These AI tests only become interesting when the AI has to actually gather it's data in the real world using real world sensors.

[@Steve3684]

Excellent breakdown Mover. I couldn't agree more. Could computers take over one day? Possibly. Will it be any time soon? Not a chance.

[@RevanBartus]

Calling this a "perfect win" considering all the advantages AI had is embarrasing.

[@Redmanticore]

i mean in real life the a.i. still has certain advantages the human does not. also downfalls.

[@raabaddler5802]

very true but if it couldn't win all of them like this would it be worth even funding the next round of prototypes?

[@mazbaird]

Yeah the AI has the advantage of not passing out at over 10 Gs...Any aircraft that can make a 17 G turn against a 9G turn will win every time.

[@RogerJL]

@C.W. Lemoine: SAAB Viggen JA and Gripen has cannon aiming slaved to radar for head shots - it does the fine tuning for exact shots. Viggen does it from behind too, but I guess it can get lost. Gripen can autofire from behind when there is a 95% chance of hitting. I assume Gripen E to be able to slave from other sensors too, like FLIR.

[@sinabastami]

This was completely on point. Everything in it was true. The most important aspect from now on, as we will integrate and incorporate AI into warfare, is that decision making element of a human vs AI. That cannot be understated.

[@zayannajam6691]

Elon Musk: They called me a mad man

[@asphalt325]

OK so should I live or not?

[@ftc9258]

Thanks for doing this episode. As always, you explained a lot.

[@gordyboy01]

Hey Mover, hope you guys are doing okay up their in new orleans, and louisiana in general, dealing with hurricane laura, stay safe guys .

[@LManonymous]

AI only wins when in a program because the computer knows everything. Real life is not like that.

[@tojassargaja2085]

You can put sensors on a plane so it converts real life into a computer. If you put enough sensors it will know almost everything.

[@kiko5573]

alot of thing depend on technology these days. AI like alphaGo just a simple AI even this Falco one.

[@razorfett147]

@@tojassargaja2085 and what kind of sensor technology would that be? Radar? IR? Lots of GoPros? How many fighters flying today have that level of realtime 360° sensory tracking?? What you're inferring is Star Trek level sci-fi tech....tiny modules that can detect anything at large ranges, and can be affixed all over the airframe without significant compromise to aerodynamics, TtW, or fuel efficiency.

[@kiko5573]

@Black Fx nope, ai can do it way better, just today ai is not that advance yet, that what i want to say. Conclude ai is terrible by watch this fight from darpa is wrong. AI still wayy far from its full capability, Full/ super intelligent AI can train /think faster than human.

[@tojassargaja2085]

​@@razorfett147Isnt dogfight shortrange?

[@ScreaminEmu]

shit, being an airline pilot in the sim is hard enough. my first day of CQ is basically just "re-learning" how to fly the simulator.

[@abe2571]

Really. The guy is wearing a mask during a zoom meeting.

[@lolembarrassing2758]

embarrassing

[@2xKTfc]

Could be sitting in an office with a bunch of people. That's what we do at work when we don't have the big conference room available.

[@chauschi]

He’s going to need AI to program for him due to CO2 inhalation

[@RovingPunster]

Hi Mover, Ive never worked on AI like this but I used to work for a company that did a lot of radar and air traffic control software, so I think I can speculate a little on what some of the strengths an AI might have in a situation like this. I'm not a pilot, so take this with a grain of salt. The AI is almost certainly VERY good at 2 things: 1) Tracking targets with extreme precision and at very high refresh rates (many times per second) ... esp if it only has one target to focus on. It can also rapidly extrapolate an extremely precise predicted position based on target course and speed from multiple prior scan passes. In short, it always knows EXACTLY where you are, and where you will be based on your current motion ... assuming that motion continues without change. 2) Computers are VERY proficient at high speed ballistics computations, and combined with the data from #1 that gives the AI a decisive advantage over any human pilot when it comes to guns ... it can fire short burts with astonishing accuracy at your projected position and then simply wait a few hundred or even a few thousand milliseconds for you to collide with the burst it already fired. THAT is the AI's chief advantage - rapid high precision course projection and firing at same with high accuracy and efficiency. Based on that, priority number 1 for a pilot would be to try to push back against the AI's ability to shoot at where its tracking projections predict he/she will be ... and that means juking whenever the AI's nose sweeps across your flight path, even if only a little. Probably still get hit, but it should reduce the damage. Priority 2 would probably be giving the AI other things to worry about - like a missile. A missile would force the AI to respond with a defensive maneuver sequence lasting until it predicts the danger has passed ... which seems like a viable way to for a human pilot to briefly buy some maneuvering initiative to line up a shot. AAAAAAH, but missiles arent allowed in that simulation ... holy convenient one sided handicap Batman. Not a fair test. 😉 I'll stop there. Keep up the great content. +1 😁 P.s. I dont understand banger's tactics in engagement 5 - why just circle ad nauseum ? Made no sense to me.

[@evanwochos451]

The pilot didn't seem to try very hard, literally just went in circles and didn't even hit the AI once, but to be fair the AI was kinda OP and the Pilot had a VR headset.

[@fireball676]

For banger to be at a disadvantage and still come up in engagement positions is still impressive.

[@thebadcrc]

Very good and interresting analysis. Thanks.

[@Blackmoon9u9]

I think the pilot made some rookie mistakes on purpose

[@RogueBeatsARG]

Cant imagine how a High Aspect shooting AI would perform in a supermanouverable aircraft like a Flanker or Raptor instead of a F-16, Just imagine the merge, the Plane doing literally a horizontal 180 degree cobra and shooting a quick burst or missile faster than the pilot can even start the turn... scary shit

[@tanit]

Just a heads up. JSBsim doesn't have graphics. It's a high fidelity 2d engine/host - the stuff happening under the hood is very high fidelity. It needs an "IG" (image generator) to render graphics. Can't tell what IG that is, might be VRSG.

[@CWLemoine]

It’s very potato-like.

[@vivekkaushik9508]

Yes! Been waiting for this video.

[@DoubleClutchProductions]

Those head on shots reminds me of DCS when the cockpit turns red, dam aimbot is no joke.

[@ThomasCallahanJr]

So basically this proved that “AI” isn’t really “AI” is a glorified algorithm system. If you give it all the adversary information it’s obviously going to “learn”. Good to see it’s coming along, but if anyone thinks pilots are going to be extinct any time soon... I’d say you’re dreaming. Based on this, if Banger just rolled out and ran away he could have ran AI out of gas.

[@Warriorcat49]

AI in this context always has been and always will be some form of algorithm. The learning part comes from how you build the algorithm; you give it a certain goal and iterate the code, essentially, until it can achieve that goal X% of the time, rather than some sci-fi “AI learning and adapting on the fly” kind of thing. So no, no artificial consciousness any time soon, but that’s not to say what we have now isn’t powerful in its own right. Simple and relatively easy to spoof? Yeah. But still extremely good in narrow application.

[@stargazer2504]

Well, for one, Banger NEVER hit the brakes and let him fly right by...

[@Luckyyetcursed]

So... I'm not going to see AI wingmen any time soon? And more importantly extended scenes of Jessica Biel in a bikini by a waterfall? So disappointed.

[@Redmanticore]

he is not an expert on a.i., every year a terribly expanding field, so i wouldnt be so sure.

[@toddb930]

I agree with you Mover. I retired a couple years ago but during my career I was involved with controling electro-mechanical systems to move from one position to another. What Heron demonstrated was a 3-dimensional control problem. Heron was able to know, at any time, exactly where Banger was located and exactly what direction he was moving and how fast. Heron probably even knew if Banger was speeding up or slowing down. Accomplishing this wouldn't involve any AI. It's hard to say if Heron was learning what Banger was reacting to and the moves he was making. Heron may just have a set of extend and engage profiles it was using to get the job done. It would be interesting to see a Heron simulation in which Heron didn't have all the precise details of Bangers position, direction, velocity, acceleration.

[@TorToroPorco]

A fascinating look at bleeding edge tech. What was left unsaid is the increase in complexity when dealing with a multi ship BFM. But given the high risk involved with WVR combat nowadays I’m not even sure how relevant high density BFM is. However I could see huge potential in BVR where positioning of sensors to avoid blind spots in SA (to overcome ECM, notching, beaming) and positioning aircraft to optimize missile shots for things like NEZ.

[@Hatredy11]

This guy is defining AI logic and wearing a mask at a video conference?

[@Jason608]

It's possible he was giving his video conference from an office with other coworkers in the building. My employer has mandated mask wearing while on site, so maybe his employer has the same rules.

[@fanou101]

Eye-opening comments, Mover ... thanks. A bit of a reality check that AI's not ready to take the lead in messy real-world open-scenario environments.

[@sameo416]

Reminds me of the Bomarc missile debate in Canada, Cold War era. Who needs manned fighters if you've got nuclear warheads on missiles? Thanks for the analysis - this was a goofy demo. I'm remembering the number of times we'd get hits on banner missions in Canada during AAG camps...getting that little 20mm to meet up with a target in that big sky is non-trivial, even with radar and predictive sights.

[@christopherhoffman2560]

Mover, simple solution is turn on autoAI engage and watch the aircraft dogfight for you like tesla autopilot, and then you the pilot intervene as desired, everyone is afraid of ai, but as you see, it has its limitations