back to article You can be my wingman any time! RaspBerry Pi AI waxes Air Force top gun's tail in dogfights

Today's generation of fighter pilots could be the last of their breed, thanks to an AI system dubbed ALPHA that's proving unkillable in air combat. The US Air Force has just completed dogfighting trials in a simulator, pitching the software against retired Air Force Colonel Gene Lee. The AI – which ran on a $35 RaspberryPi …

  1. Paul Crawford Silver badge

    "forces to be deployed without human loss of life"

    On your side. While that is generally a good thing, if it makes politicians more trigger-happy it is probable not.

    1. M.Zaccone

      Re: "forces to be deployed without human loss of life"

      Quite. It also depends on where and why you are fighting the war. Having just read Robert Fisk's "The Great War For Civilisation" , I find my beliefs in question and now I find I'd rather we never sent a drone of ours overseas ever again...

      1. Dave 126 Silver badge

        Re: "forces to be deployed without human loss of life"

        Years ago, I watched documentary about cats.

        They said that cats rarely fight, because of displays of power - the weaker cat would soon realise it was the weaker cat, and thus bugger off to hiss another day.

        This documentary, Tiger on the Tiles* by Desmond Morris IIRC, doesn't fully inform my armchair strategic thinking.

        Nor does that episode of Shatner Star Trek, the one when a star system's internal battles were simulated by a computer and the virtual causalities calculated - then each side would have to kill a corresponding number of its own citizens. It was neater - no infrastructure damage.

        I guess the film The Last Starfighter, is now out of date. (In this film, a young 1980's American lad proves to be very good at an arcade video game. It turns out that the video game arcade cabinet was a plant by aliens involved in their own war, and that they were recruiting pilots from uncontacted planets. )

        WHOOAH! I've just Googled the above film - It turns out Seth Rogan has been wanting to remake it, but he tweeted that Steven Spielberg told him that not even he has been able to get the rights - the film's writer won't give it up. Probably a canny move - the film might be more powerful in a few years time if recreational Human Interfaces (VR, HUD, AR etc etc) come closer to military gear.

        *Yeah, I was watching cat video before broadband internet.

        ** I like Ricard. Not Picard, Ricard. Which I have been drinking tonight. With water, obviously; I try not to be a jerk. But I may be off topic ish.

    2. Christoph

      Re: "forces to be deployed without human loss of life"

      No humans killed, just foreigners. If they are not Americans then they don't count as people, just as targets.

    3. Cuddles

      Re: "forces to be deployed without human loss of life"

      "While that is generally a good thing, if it makes politicians more trigger-happy it is probable not."

      This is a sentiment I see a lot, but it doesn't really make a lot of sense given an awareness of history and human nature in general. Numerous weapons have been touted as bringing about an end to warfare because, the machinegun being a particularly well known example, but every time they've been immediately deployed as enthusiastically as ever. The only thing that comes close to being an exception is nuclear weapons, but even those were used a couple of times and there remains a very realistic possibility that they will be again in the future. Importantly, when it comes to politicians being trigger happy, even the existence of nuclear-armed superpowers able to wipe out all human civilisation at the push of a button wasn't enough to stop them repeatedly going to war both with each other and anyone else they could find lying around the place.

      If knowing more people will die doesn't make politicians less likely to go to war, how likely is it that the reverse will happen if they think less people will die. Since the end of WW2, the USA has started a new war (or military intervention of some sort) an average of every 3 years. How much more eager to start wars could better planes make them?

      On which note, it's important to bear in mind that this is only about fighter planes anyway. The vast majority of combat losses are not fighter pilots, or indeed aircraft crew of any sort. Many ground support roles are already done by drones, but even replacing all pilots with AI wouldn't make any difference to the losses that would occur once boots are on the ground. And given the degree of air superiority the US, and the West in general, currently enjoys in all the wars they get involved in, having slightly better planes wouldn't make any significant difference at all. The only opponents this would make any real difference against would be Russia and China, and fighter pilot losses are hardly the important factor that will tip politicians into wanting to start those ones.

      1. itzman

        Re: "forces to be deployed without human loss of life"

        Mmm,. And of course when the enemy ha AI, you just fight the fight in virtual reality.

        However that's not the really alarming or exciting thing. Its just how much AI you can pack into a small computer.

        That means a $500 autonomous drone could in theory be set loose to - well - assassinate someone.

        And you might never know who was behind it.

        It scuttles into a corner, plugs itself into the nearest USB socket charges itself up, connects to the internet (its INSIDE the firewall now) and looks like a USB flash drive, uploads anything useful, gets fresh orders, then extends legs and scuttles around before shooting a poison dart into an ankle.

        The Apple I-Spy (lethal upgrade)

  2. allthecoolshortnamesweretaken

    How big is the hardware this runs on?

    But yes, soon a new generation of fighter planes will be designed around an AI system. (If there's enough money left after the fascinating omnishambles known as the F-35. Aw, what am I thinking - of course there will be money for it!) Let's just hope a system like that isn't hacked.

    1. Throatwarbler Mangrove Silver badge

      FTA: "The AI – which ran on a $35 RaspberryPi computer"

      So, not very big.

      1. kmac499

        Phew that's a relief.

        As the UKs future (post Brexit) looks a bit grim at least we'll be able to afford the best AI fighter pilots in the world.. Plus with the low power requirements of the R-Pi it should fit neatly into our front line squadrons of Sopwith Camels ... Chocks Away Ginger.....

        1. Tom 7

          Re Phew thats a relief

          Sopwith Camels would have pretty much a non-existant radar response too! Stealth attacks that would be mistaken for the local youth on a moped would be all the rage.

          1. itzman

            Re: Re Phew thats a relief

            I played a flight sim in a camel and shot down an ME109

            Very manoeuvrable, is a camel.

    2. PAKennedy

      It's running on a Raspberry Pi. About the size of a credit card and as thick as a USB port.

      1. elDog

        But it needs to be "militarised"

        Remember the $10 toilet seat that cost $700 to be installed in some fighter?

        Think the $35 Arduino that after Lockheed-Martin/Boeing/MIPC and all the reviews and beltway analysis will cost $350,000 (initial, $400,000/year min for maintenance.)

        And now it won't even work. And by the time congress has voted 45 times, the opponents have moved on to AI forms far superior.

        Ahhh. The days when any fly-boy could get in his crop duster and drop a home-made incendiary on something. Those were the days.

        1. Sorry that handle is already taken. Silver badge

          Re: But it needs to be "militarised"

          Why was a toilet seat being installed in (or on) a fighter plane?

          1. Old Used Programmer

            Re: But it needs to be "militarised"

            It wasn't. It was in a bomber. The extra money was the usual paperwork, but not distributed over several hundred of them. Same with the stories about extremely expensive hammers.

          2. BebopWeBop

            Re: But it needs to be "militarised"

            Why was a toilet seat being installed in (or on) a fighter plane?

            Might be very useful when a meat sac realises he is up against one of these?

          3. Voland's right hand Silver badge

            Re: But it needs to be "militarised"

            Why was a toilet seat being installed in (or on) a fighter plane?

            If memory serves me right, the seat in question was on board of B1-B. That one one of the many examples of ridiculous overspend and fleecing the public by military equipment contractors during the rule of the late president RayGun. Unfortunately, what got established during his reign continues ever since.

            In any case, all bombers have toilet facilities as they are intended to fly 16h+ missions with in-flight refuel in order to send that friendly greeting to another nation. Even something just barely bigger than a fighter like the Su-34 Platipus has a toilet. Dunno about fighters though.

            1. Anonymous Coward
              Anonymous Coward

              Re: But it needs to be "militarised"

              If memory serves (because I'm too lazy to look it up), costs were high due to the overzealous operating-parameter specs and that they were turned out on a rather expensive lathe.

              1. Anonymous Coward
                Anonymous Coward

                Re: But it needs to be "militarised"

                Toilet seats were expensive because the rules at the time requires that everything be put out to bid even if it was only one toilet seat that was needed. Companies would put in bids for supplying hundreds or thousands of toilet seats because nobody want's to bother supplying a bid for one. As a result the government winds up buying a few hundred when only one was needed with the spares going into a warehouse where they would sit and finally be auctioned off as a single lot for pennies on the dollar. Thus the entire cost is absorbed by the single used toilet seat. Recently someone pulled their head out and decided that COTS items picked up locally were a far better deal than the bid system since even if it isn't quite "fair" to the contractors it's more fair to the taxpayers.

          4. Anonymous Coward
            Anonymous Coward

            If someone was shooting at me

            I'd shit myself!

          5. Steve K

            Re: But it needs to be "militarised"

            Shhhhh! The USAF are trying to keep a lid on it now.....

          6. Fungus Bob

            Re: But it needs to be "militarised"

            "Why was a toilet seat being installed in (or on) a fighter plane?"

            So the pilot could take a shit, DUH!

          7. Paul 195

            Re: But it needs to be "militarised"

            Because it's scary being in a fighter plane?

      2. Old Used Programmer

        Two stacked USB ports, if it's a B-series Pi (which is likely...given when the encounter is said to have been, it's very likely a Pi2B was used), or a single USB port for the A+, plus the thickness of the PCB and some short protruding pins and chips on the bottom.

        For most people, the relevant size is that the PCB is the dimensions ISO sets for a credit card.

        One really has to wonder what kind of market there would be a Pi AI add on to a commercial combat flight sim. Might actually be worth more (and with better margins) than selling this stuff to governments as unit sales would likely be *way* higher.

      3. m0rt

        "...and as thick as a USB port."

        What is the usual IQ of a USB port then? How do Thunderbolt ports compare? What about firewire?

  3. Dwarf


    Anyone else thinking rise of the machines ?

    1. Destroy All Monsters Silver badge

      Re: T500

      Agreement factor 75%

      At this point, seems to be more like "chessmasters in the sky". They are not slowed down by any remains of human fighting ritual, so this is like fighting a lawnmower while taking on the role of the possum. Once they get autonomous enough to take off from the carrier and do CAP all by themselves, it's going to be interesting.

      Then someone with connections to the MICC will propose to close the feedback loops and give these things nuclear ramjets, just to be sure...

      I have never heard of this particular decision fuzzification (is there a book from Springer yet?), but I remember the good old rule-based "TacSoar" (based on SOAR with the "chunking"/rule-learning stripped out) trying to assist air missions back in the 90s (the time of "Strategic Computing Initiative" and all that). It was not very successful and also easily confused by ground-hugging kangaroos, IIRC.

      1. Voland's right hand Silver badge

        Re: T500

        decision fuzzification

        Not new, just approached from AI/Machine Learning perspective.

        Russian guidance systems have been using probability space calculations for two decades to more or less the same effect. They do that by applying a differential inclusion to a set which describes a discrete representation of the probabilities of target actions. In laymen terms - instead of applying a deterministic calculation to the target current vector (like some of the BAE crap), it applies a similar calculation to the probability space of immediate future target maneuvers based on current and previous vectors and continuously adjusts it. It also does using appropriate multiple pursuers, single target equations for a multiple rocket launch scenario.

        As a fundamental idea this is similar - you do not work with deterministic data, you make up (by randomization) a probability space out of it and you apply an algorithm (in this case a machine learning one instead of differential inclusions) to it. You also continuously adjust it via a feedback loop (same as Russian diff incl/prob space algos).

        As a result the machine in either case ends up thinking one step ahead. Even if the wetware in the opponent had a couple of preplanned moves initially, it is quickly out-computed in the initial encounter. After that the human ends up being reactive and not being able to think ahead due to timing constraints. So the AI (or the probability based algo in the Russian case) kills him at short notice.

    2. Mark 85

      Re: T500

      Yes, indeed. I would hope the meatbag pilot would still be aboard if for nothing else than to override the AI. But then that put's limits on the aircraft's performance. So... a bit of sentience might very well be a bad thing.

    3. Ment_al

      Re: T500

      Yes, me too.

      Dear Scientists, please stop building Terminators...

    4. Dadmin

      Re: T500

      Not really, as long as this is a RPi2 then I can take out the system* with a high-power LED flashlight or LASER. Unless they strategically apply some electrical tape to it.

      *reboot, or other disturbance via the 5v rectifier chip error on that device

      Also, I agree with that guy ^^^. In that I find it unfortunate that a lovely hobby computer might be used to kill people. Can't they use some Intel clone for the dirt work?

    5. Peter Gathercole Silver badge

      Re: T500

      No, I'm thinking of the X-9 Ghost in Macross Plus, or the F/A-37 Talon in Stealth.

    6. Tom 7

      Re: T500

      Its not just the rise of the machines that you have to worry about - its the rapid approach from any angle.

  4. Anonymous Coward
    Anonymous Coward

    Hopefully this will mean cheaper planes

    Without the need to protect a human life inside, it seems like the correct strategy would be to make them cheap and numerous, rather than loading them up with all the bells and whistles like stealth and supersonic capability. Heck, maybe they don't even need missiles, if they're cheap enough they can just ram their targets.

    I doubt the US will go the "cheap" route because that would mean less profit for the aerospace industry, but you can bet the Chinese will. If it ever came down to an air war between the US and China, where the US has a few thousand $100 million AI drones and the Chinese have a few tens of thousands of $2 million AI drones, I know which side I'd place my bet on.

    1. Destroy All Monsters Silver badge

      Re: Hopefully this will mean cheaper planes

      The first thing to employ would be nuclear weapons to clear the skies, followed by anti.aircraft missiles...

      1. Dave 126 Silver badge

        Re: Hopefully this will mean cheaper planes

        >The first thing to employ would be nuclear weapons to clear the skies, followed by anti.aircraft missiles...

        I'm assuming you mean the EMP effects of nuclear explosions. That'd do it.

        Research into EMP blast is ongoing - both in the offensive ("can we make an EMP 'cannon' without a nuke?") and defensive ("How do we protect our shit if our enemy uses EMP?")

    2. Mark 85

      @DougS -- Re: Hopefully this will mean cheaper planes

      I'm not so sure... back in the Cold War, the US decided it couldn't afford the numbers of main battle tanks to take on the Russians. So they came up with M901 which basically was a cheap, expendable tank destroyer. Since then, TOW missiles are deployed on everything from HumVees to armored personnel carriers. Everything can be a "tank destroyer".

      It's possible that the DoD would consider low-cost and in large numbers for this as occasionally, they do get it right. The AF, however, probably won't like it as they like the dashing fighter pilot image.

      1. John Brown (no body) Silver badge

        Re: @DougS -- Hopefully this will mean cheaper planes

        "The AF, however, probably won't like it as they like the dashing fighter pilot image."

        Except it appears to be the AF leading this charge to do away with fighter pilots. There seems to be some divisions within the ranks.

      2. Anonymous Coward
        Anonymous Coward

        Re: @DougS -- Hopefully this will mean cheaper planes

        Well, tanks are a different matter because of the armor and relatively slower speeds at which they travel. Having a million unarmored unmanned tanks does you no good, as they would be easy to destroy so the only hope they'd have of stopping a good tank would be having so many burning carcasses in the way the good tanks couldn't get past :)

        Since planes aren't armored (at least not to any degree that matters in terms of the ability of other planes to destroy them with large caliber bullets, let alone missiles) and travel much faster, a horde of cheap planes could do a lot of damage. If they relied on kinetic kills they wouldn't need to be all that big, and could probably be made for a few hundred thousand dollars each. A simple jet engine good for 400-500 mph could get it in the ballpark of its target, and a solid rocket booster on the back could give it some extra pep in its step during the last couple seconds to maximize the damage and make it harder to avoid.

        Yeah yeah, you can clear the sky of them with a nuke, but I'm kind of assuming a war wouldn't go in that direction. If you're using nukes, every other piece of equipment in a war like planes and tanks becomes irrelevant anyway. It will be ICBMs and sub launched nukes with mutual assured destruction, and the world is fucked.

        As for it being a bad idea to make a war cheaper, I'm not delving into the morals, I'm thinking in terms of what a military should do to best prepare itself for the conflicts they'll face in a decade. If the US decides "well we don't want to make war cheap, that's a bad idea, so let's keep throwing money away on manned boondoggles like the F35" while China and Russia are making the drone horde I outline a reality - and selling it on the open market, the US is going to be screwed in a conflict. Wouldn't matter if his plane had perfect radar invisible stealth, a human pilot facing a horde of a few dozen of those things will end up dead if he doesn't run away.

    3. Christian Berger


      Do you seriously think we need to make wars cheaper? After all in our current wars are highly asymmetric. Wars aren't fought between equal states, they are now fought between powerful states an some ill-defined groups of people. Particularly with drones the risk on the state side is close to zero.

      At the same time you will have large parts of the population living in constant fear. Every drone strike could kill them and their families even though they haven't done anything. And no, they cannot flee, as people like Merkel closed a deal with Erdogan so Turkey won't let them through to Europe, or airlines being forced to pay the cost of bring refused refugees back, or refugee ships being prevented from entering harbors.

      Then if you look at the people the states fight against, you'll find that those are often exactly the people which have been militarized in the first place:

      And if you look deeper, you'll find that much of this is based on thinks like The Sykes-Picot Agreement:

      You cannot solve problems by killing people on mass. Making war even cheaper will certainly not help.

      This isn't "SWAT Kats"!

      1. Joe Montana

        Re: Hopefully?


        You cannot solve problems by killing people on mass. Making war even cheaper will certainly not help.


        Oh but you can, you just need to do it thoroughly and brutally, ie the way things were done historically.

        There would be no refugees fleeing if there were no refugees left.

        No refugees would flee a warzone where they risk death to enter an enemy territory where they faced almost certain death at the hands of an organised military force.

        Insurgents could not hide amongst civilian populations if the rulers were willing to systematically wipe out civilians until the insurgency stopped.

        A lack of morals and a willingness to do whatever necessary is a significant advantage in any war.

    4. Simon Westerby 1

      Re: Hopefully this will mean cheaper planes

      You would think the 'Muricans would have learnt that lesson from WW2...

      Specifically the M4 vs the Tiger..

      One a superior engineered war machine.. The other a mass produced heap-o-junk...

      1. Anonymous Coward
        Anonymous Coward

        Re: Hopefully this will mean cheaper planes

        One a superior engineered war machine.. The other a mass produced heap-o-junk...

        Arguably the Germans had better small arms, better artillery, better tanks, better ships, better submarines, better field transport than almost anything the Allies had. The only military assets that the Allies had that were incontrovertibly better than the German equivalent were medium and heavy bomber aircraft (and perhaps aircraft carriers), and arguably something of draw on fighter technology. You can argue the toss on specific types and niches, but overall that's a reasonable summary.

        So it would appear that cheap junk trumped superior engineering. Presumably this is the logic behind the F35: "No point in superior engineering, the enemy can afford cheap junk that'll win, so we need to counter with something they can't create: Expensive junk."

        1. nematoad

          Re: Hopefully this will mean cheaper planes

          To employ a Napoleon misquote "God is on the side of the big battalions"

        2. Tikimon
          Thumb Down

          Re: Hopefully this will mean cheaper planes

          Don't know where you're getting the "Germany was superior in every respect" thing but it's incorrect.

          Better field transport? Are you referring to horses and feet? The Germans relied on horse-drawn transport to move supplies to the very end. Most German troops walked or were moved by rail. Neither served very well on the Eastern Front where good roads and rail were not available. Their only real cross-country mobility was in their tanks. The US Army was almost entirely motorized, using excellent all-wheel-drive deuce-and-a-half trucks to carry supplies and troops, even where the roads were awful or nonexistent. It was US mobility that enabled the Bulge to be contained, moving troops to block German moves faster than the Germans could advance.

          Hitler failed to secure his western flank because they simply lacked the ability to invade the UK. A few years later the Allies would show them how it was done at Normandy.

          They did invent the excellent STG-44, but these appeared too late and too few to make a real difference. The average field troops carried K-98 bolt-action rifles, at a disadvantage to the M1 Garand semi-automatic. They didn't build the MP-40 until Soviet forces with plentiful submachine guns started trashing the K-98 armed troops in close quarters combat. Kudos to the MG-42 which was a real terror, and its descendants live on today.

          Both sides had their high and low points in engineering. It's totally incorrect to state "Superior war machine vs. mass-produced junk."

  5. channel extended

    The Goverment bid....

    Suddenly Pi's will cost $3.14 million each from Lockheed!! And take twenty years to deliver.

    1. VinceH

      Re: The Goverment bid....

      Don't panic! According to the headline it's using a RaspBerry Pi, and the article says it's a RaspberryPi - so I think the Raspberry Pi is safe from that possible future.

  6. Magani

    I S'Pi with my little eye...

    Let's see if I've got this right:-

    * An AI system of very high complexity and speed can run on a Raspberry Pi.

    * It can seemingly beat a wetware Top Gun who's hosed other very expensive AI systems.

    * The AI software presumably fits on either an SD card or a USB drive ('cos that's the way Pis work)

    * This is the same RasPi that my grandkids use with Scratch, right?

    So, either all the aerospace companies are ripping off the US/UK/EU taxpayers by saying they need squillions to develop airborne AI systems, or ...

    My Bovine Excrement Meter is telling me that April 1st has reappeared on my calendar.

    Obviously boffin icon material (or is that materiel?)

    1. Dave 126 Silver badge

      Re: I S'Pi with my little eye...

      So, to boil your post down to its essentials: You're asking why it is possible for some software ( game playing in the realm of physics) has been developed for an ARM chip which is better than that written for some other chip from a year or two back...

      I'd say it's plausible. The thing about software is that you can never know if your algorithm is the fastest that can be written. Also, the last few years has seen a lot of work down on autonomous vehicles - there is more prior work to draw upon.

      If you've ever tried to swat a fly then you will know that the hardware to control some damned good flying doesn't have to be big - if the software is good enough.

      [You've used the wrong icon. If you hover your POINTER over the ICON it will tell you when it should be used. If you don't have a mouse or trackpad - i.e you are using a phone or tablet - then it might be best if you don't use any Reg icons, since the lack of MouseOver means you won't know what they mean.]

      1. Magani

        Re: I S'Pi with my little eye...

        @dave 126

        Well, thanks for your reduction of my points. What I'm trying to get at, if this claim is to be believed, is the enormity of the reduction in the hardware required for this dog-fighting AI from dedicated, expensive machines to a $35 piece of hobby kit. Your comparison to any wetware's ability (even fly-sized) doesn't translate to anything man-made and silicon.

        The AI required for dog-fighting is within an order of magnitude for that required to beat a chess Grand Master. Therefore, we should be looking at IBM to trade in Watson on a couple of RasPis? It beggars belief.

        "[You've used the wrong icon."

        Er, NO, I didn't. I chose it because it would take a huge amount of boffinry to get a RasPi to do what's claimed. At the risk of revealing my Secret Squirrel identity, I started working in aviation in the late '60s, and more particularly, writing software in that area in the '80s, so I hope you'll grant that I know just a tad about what I speak.

        More iconic boffinry awaits your downvote. :-)

    2. Anonymous Coward
      Anonymous Coward

      Re: I S'Pi with my little eye...

      It all depends on the parameters. Suppose for a moment we're thinking in terms of air-to air gun combat (I'll get onto missiles in a moment). Further, suppose that you're pitting human against AI in identical aircraft. The AI has the edge because it can pull high-G manouvres until either its fuel runs out or it overdoes it and snaps a wing off. The human may be able to pull high-G manouvres for a while, but will eventually tire and also probably won't cope as well with high positive G followedby high egative G and back in quick succession.

      Some years ago I recall reading that AI based on that used in the game 'Creatures' had found a novel tactic in air combat - that of the eternal barrel-roll, which made it nigh on impossible for a human pilot to get a firing solution on it. On the other hand, the AI coudl work out just when it could shoot with a reasonable chance of hitting teh human-piloted plane, even with teh human taking evasive action and it doing its eternal barrel-roll.

      If we consider missile combat, in order to shoot, the target can be in a rather larger area of sky as compared to if a gun solution is required because the missile is guided. Defensively, the AI has the advantage because it can pull more extreme manouvres (and can time them more precisely) in order to try to evade incoming missiles, whilst offensively it can better take advantage of very fleeting firing solutions that a human might not be able to react to.

      Things get more complex when ECM (electronic countermeasures) chaff and flares and suchlike are available to aid defence, and I wouldn't like to guess as to whether the AI has the advantage with them in the mix or not (I've only flown virtual aircraft, of up to Korean War vintage) But absent them, yes I can believe that an AI pilot could do a better job of getting a firing solution on a human than vice-versa. Air combat is much simpler to program for than ground combat - far fewer variables to cope with.

  7. ecofeco Silver badge

    Oh shit

    You are witnessing history.

    This takes us back to the time when high altitude bombers, fighters and ballistics missiles could not be reliably defended against.

    Yes, the tactics are different, but the results are just as terrifying. Only railguns, energy weapons or highly accurate missile will be able to defend against this.

    1. AnonFairBinary

      Re: Oh shit

      ... or another AI enabled drone ?

  8. Dave 126 Silver badge

    Read in depth

    The humans were at a numerical disadvantage.

    The mission analyzed in this document features two blue fighters

    vs. four red. The red aircraft begin over a defended coastline and

    the blues are 54 nautical miles due west

    The white paper itself is fascinating though! Go to page two:

    1. Anonymous Coward
      Anonymous Coward

      Re: Read in depth

      Bit of a problem with that journal -- it's classed as a predatory publisher. That doesn't necessarily negate the work of the authors, but you might want to think twice about the veracity of what's reported. For more information on Omics (and junk journals in general), see:

  9. Evil Weevil

    Can El Reg militarise the LOHAN project

    ....with this AI and a raspberry pi as a tribute to Lester Haines ?

    1. Simon Westerby 1

      Re: Can El Reg militarise the LOHAN project

      L LOHAN

      E Easily

      S Simulates

      T Thought

      U Using

      R RaspberryPi

      Couldn't work out the 2nd E do cheated...

      1. Vulch

        Re: Can El Reg militarise the LOHAN project

        E Employing

        1. Anonymous Coward
          Anonymous Coward

          Re: Can El Reg militarise the LOHAN project

          "LESTUR" sounds ominous enough.

  10. John Jennings

    Can I has?

    Can I has this on my copy of Falcon 3? or better yet - BOB (Battle of Britain) I might make me get back into combat filghtsims....

    1. Old Used Programmer

      Re: Can I has?

      Only if the opposing side gets it. Of course, that means you'll always lose....

  11. JLV

    See, this is why we should think 20 times before continuing commitment to the F35 system.

    This AI may or may not be that hot in real life. Seems a bit too good to be true, to be honest. But, at some point, they are going to be good enough. The whole "but you need to have a human in the loop" is also a red herring.

    Yes, you need a human in the loop to fire on blended civilian/enemy locations, like we've seen in the last 20 years. But, if, instead of shooting up 3rd world opponents, we move to a hot war with China (not advocating that in the least, but let's face it, that's the only excuse for the F35) then anything flying that doesn't respond to IFF as friendly will be fair game.

    In an all-gloves off fight with a technologically savvy opponent there is no guarantee whatsoever that autonomous AA combat drones wouldn't massively outperform humans in say 2030 - 10 years into the F35's active service. I am thinking mostly area denial - air defense over a territory you want to deny to your enemy's air force. You don't need much - if you can field drones cheaply enough, your enemy won't be able to replace their pilots no matter what the human-over-drone kill ratio is.

    Drones may not in fact be that efficient by then, but if they do we will have way too much sunk costs to re-arm.

    We just don't know. Just like a dispassionate analysis of battleship survivability in 1935 would have flagged that they were at risk of becoming a liability (well, actually Mitchell said that in '21 and got canned for his efforts). No one ended using them for much except shore bombardment. Or the superiority of fully mechanized forces over static infantry formations - that was also pooh-poohed by Western armies - sunk costs in Maginot Line.

    1. Mark 85

      This is perfect for air superiority fighters but not so good for ground attack, as you're pointing out. But if your drones control the airspace, the slower moving, meatbag flown attack craft have a better chance of getting home. On the other hand, we do have drones that do limited ground attack (fewer weapons, no guns). Warfare is changing....

      And yes, the F35 should be canceled as I'm sure there's things that will work better for less money. Even re-opening production of the F22 has been flogged about in some circles.

    2. itzman

      WE are already at te stage where in terms of air combat, missiles have the edge over manned fighters Purely in terms of 'G'

      Give them wings, and more fuel, and an AI equipped missile is simply unbeatable,

      However all that does, is mean you send over unmanned AI equipped bombers with mach 17 AI equipped air to ground missiles.

      I'll let you into a secret. A mounted knight is no match for a machine gun

  12. Chairo


    If they need a name starting with "A", why not Arnold? Would somehow be more fitting.

    1. Anonymous Coward
      Anonymous Coward

      Re: Alpha?

      Motto: "I'm in your back!"

      (The good advice for countering is of course "Roll a hard six!")

  13. aqk
    Big Brother

    Nerd questions! (finally!)

    So was this a Pi-II or a Pi-III ?

    And was it running Raspbian or (gulp) that new Windows-10 thingamajig?

  14. Tom 64

    Deeply impressive tech.

    Given this is running on a Pi, its awesome. All those meaty neurons wasted.

    OK, its running a pre-taught AI on the Pi, but its still awesome. Read the PDF for more insights, they basically had an AI train this AI.

    I wiser man than myself once remarked:

    'Game over, man!'

  15. John Smith 19 Gold badge

    I've read the paper.

    There is no reference to it running on a Raspberry Pi.The learning system runs on a single desktop PC. Very interesting approach.

    For those of an older generation of AI work I'd suggest it runs a multi level blackboard of fuzzy logic systems that analyse different parts of the problem.

    It is also an evolvable system. The system evolves system designs (using a proprietary algorithm) to solve the problem, then the most successful of those evolve to get better by learning.

    Lots of feedback with a human, lots of recursion.

    But note you can't prove there is not an interaction in the system that has not been modeled which would be it's Achilles heel in combat.

    Personally I'm suspicious they've inserted the proper time delays for vehicle responses and they respond like vehicles in a cheap video game IE instant turns and acceleration/deceleration

    But they'll have to work hard to make it the $40m a copy that cost plus defense con-tractors like to charge.

  16. Anonymous Coward

    This was a scam by Intel

    They now have the Raspberry Pi flagged as munitions and cannot be exported or sold across the world without the US getting a cut.

    There is hope for the Atom yet.....

  17. Wolfclaw

    Just get a Cylon to flash it's sexy red eye and the plane will fall from the sky, job done !

  18. ColonelDare

    RE: "The AI – which ran on a $35 RaspberryPi computer ...

    Details? Can I download it the RasPi store?

    Just what I need since my daughter gave me some Darth Vader socks for Father's Day.

  19. Dazed and Confused


    Sod the rarity and cost of training. The big problem with pilots is that they are fragile. As soon as you want to start to maneuver they break. I mean 9g in one direction and possibly 3 in the other, what sort of mil-spec HW is that. If they can get the soft squidgy bits out of their planes they can look at designs that can handle massive G loads in combat situations.

    1. Anonymous Coward
      Anonymous Coward

      Re: Pilots

      It's not only the pilot that starts to fall apart at high Gs. While a plane could be designed to handle more (including heat produced), it could also become heavier, losing some advantages. There is also the problem to ensure engines still get the correct airflow (or they can stall), and fuel feed is not interrupted.

      1. SkippyBing

        Re: Pilots

        'There is also the problem to ensure engines still get the correct airflow'

        Not to mention the engine itself, the issues with the F-35's engine were caused by the shaft bending under high g which led to the compressor blades rubbing on the inside of the casing. John Farley has pointed out on a number of occasions the worst regime for the Harrier's engine wasn't in the hover, which is essentially like being on a test stand, but at high speed and low level where air is being rammed in the front and manoeuvring sets up gyroscopic and bending loads on the rotating bits.

        1. John Smith 19 Gold badge

          "manoeuvring sets up gyroscopic and bending loads on the rotating bits."

          Which may explain why the Pegasus was the first (AFAIK) engine to run with contra-rotating shafts (or spools as gas turbine people seem to call them).

          This was one of the avant garde features of the SABRE design but now the Trent 900 uses it well it's gone mainstream, although probably not due to the A380's need to do snap turns in a dogfight :) .

      2. itzman

        Re: Pilots

        While a plane could be designed to handle more (including heat produced), it could also become heavier, losing some advantages.

        Scale is then your friend.

        A carbon fibre 3 ft wingspan jet could probably handle 50g+

  20. Seajay#

    Sensor interpretation

    I'm surprised that it's taken this long for AI to beat a human in a simulator. I didn't realise that the "deciding what to do" aspect of dogfighting was particularly difficult.

    However, this may or may not be ground-breaking depending on how accurately the simulator simulates the sensor input that a real fighter aircraft would receive. Does it simulate birds flying by, clouds, etc? Do the heading, air speed and other inputs have the sort of glitches that the real ones would have? I'd imagine that building up a real picture of what is happening from real-world imperfect data is much more difficult than doing it when attached to a simulator. But maybe I'm wrong, anyone worked with one?

    1. Anonymous Coward
      Anonymous Coward

      Re: Sensor interpretation

      I was about to ask how the AI got the opponent flight data. If it got a feed directly form the sim data, it had a huge advantage. AFAIK, pilots today don't have a full 3D-360* situational awareness by on board sensors (AWACS can feed some data, but not everything). Knowing any opponent vector instant by instant would be a huge advantage.

      1. Seajay#

        Re: Sensor interpretation

        Reading the actual paper there seems to be an answer. They say

        Data is collected from sensors, fused real-time, and sent to ALPHA, complete with noise

        and potential failures. This data feed is received at and commands sent to unique UDP ports for each platform, causing this problem to contain some of the software complexities that would be present in

        actual hardware implementation

        In which case. Wow.

  21. Lotaresco

    A need to rethink tactics

    One possibility is that there is no longer a need for human piloted supersonic fighters. Supersonic drones can be deployed to gain air superiority then subsonic human piloted aircraft can shuffle in under a drone umbrella to take on ground targets. This would mean F35 not needed, no real need for stealth. The A10 and the ground attack Harriers would be ideal for the ground attack role. Biggles is going to be upset. Biggles already had the childish pout because gamers can fly drones better than they can. Now a box of fags with some silicon can fly a fighter better than they can.

  22. AIBailey

    The fascinating part...

    ...really is the fact it's running on a £35 computer the size of a credit card.

    A few years ago an announcement that there was an AI system capable of flying a fighter plane better than a human would have been followed up with an inventory list of hardware involving computer clusters taking up the size of a garage, along with some proposal that the next version will hopefully fit in a space no bigger than a small car and require less than 10KW of energy to run it.

    The fact that this is running on something with a smaller footprint than a smartphone, that could in theory be powered from a set of AA batteries, and that I, an ordinary member of the general public, could buy the same hardware for less than the cost of the aforementioned smart phone is simply staggering!

    1. nematoad

      Re: The fascinating part...

      "...really is the fact it's running on a £35 computer the size of a credit card."

      Actually that'sa $35 computer. GBP price about £28.

  23. M7S

    Is that my flying car I hear approaching.....?

    Does the piloting skill of this offer a potential solution to the "untrained pilot" issue that is one problem (admittedly of several) that prevents delivery of the flying car to the masses?

    It might be that a combination of this "avoidance capability", flying as flocks of birds do (i.e. reacting only to the birds in immediate proximity), combined with a satnav that sorts out a new course when it has been deflected from its original heading (as most are now capable of) might solve the air traffic control problem in some circumstances.

    Of course the "air rage" option that is part of the air forces AI would probably be deleted from standard installation. Those operated by some Reg readers with the right skills might differ...

    1. Anonymous Coward
      Anonymous Coward

      Re: Is that my flying car I hear approaching.....?

      Flying car, beyond require more skills to be "driven" have also several disadvantages:

      1) Flying requires far more power (and fuel) than moving on ground. Generating lift requires power, the ground-generated lift between solids comes for free. And the more the weight, the more the power required (also weight distributions is far more critical than for a ground vehicle).

      2) Faults are far more dangerous if you can't land safely quickly. VTOL vehicles could be safer, but they are also more complex and expensive (and usually also more dangerous, due to the long blades which may be required...)

      3) Because of 2, systems becomes far more expensive, and may need redundancy (you can't realistically use ejection seats on flying cars...). Maintenance is also far more expensive.

      4) Due to the higher speed required to stay in flight, sensors like RADARs and FLIRs may be required to ensure obstacle avoidance (especially in bad weather), GPS alone may not be safe enough. You don't really want to enter an hailstorm at 100kts or more. That makes them again more expensive.

      Thereby, it's not really just a pilot issue AI could resolve, even if could see "flying cars" for the rich and wealthy, flying cars "for the masses" are really far away, I'm afraid.

      1. Kubla Cant

        Re: Is that my flying car I hear approaching.....?

        @LDS All good points. It sounds like what we need is a lighter-than-air flying car, but I suppose there are problems, such as size and slowness, with that.

  24. John Mangan

    Friend or Foe?

    They better make sure the identification system is un-hackable.

    1. Mark 85

      Re: Friend or Foe?

      And also entirely reliable. IFF dies, the pilot and plane die.

  25. Big_Boomer Silver badge

    I for one...

    would like to welcome our new RaspberryPI masters.

  26. Jock in a Frock


    ... but can it run away in a Crysis ?

  27. Bob Rocket

    2 points

    1. Have they tried using one AI against another AI ? Seems likely that using AI v AI would lead to stalemate.

    2. Have they tried using rookie pilots against the AI ? Rookies are more likely to use unconventional tactics when faced with unconventional tactics (it is the established actors who reprise their last successful performance)

    I think meatbags will still have the edge (unpredictable unpredictability to misquote Rumsfeld)

    1. Androgynous Cupboard Silver badge

      Re: 2 points

      1. Have they tried using one AI against another AI ? Seems likely that using AI v AI would lead to stalemate.

      They did, but it turned out the only way to win was not to play.

      1. adam 40 Silver badge

        What a whopper

        .. as predicted by the Tic Tac Toe scene in War Games (1983)

  28. PaulAb

    I wouldn't Worry because...

    Being a government project, some half wit will leave the Wifi/Bluetooth enabled and a five year old outside the airbase playing with an RC car will let loose all it's firepower before it leaves the runway.

  29. Stephen Keane

    No one has said it yet ...

    Pi in the sky?

    1. Chris 116

      Re: No one has said it yet ...

      Damn you! <shakes fist>

  30. Stevie


    Can it tell the difference between a civillian aircraft and a military one?

    Can it differentiate between a house and a hospital?

    If so then I say the Pis have it.

  31. Chris 116

    Pi in the sky


  32. Eddy Ito

    Skynet runs on a Raspberry Pi?

    Wait, the simplicity, the low cost... Shit! It all makes sense now. It's a full blown invasion! You've all got to bbbbbbbbbbbbbb



  33. Winthorpe3rd

    In three years, RaspPi will become the largest supplier of military computer systems. All stealth bombers are upgraded with RaspPi computers, becoming fully unmanned. Afterwards, they fly with a perfect operational record. The Skynet Funding Bill is passed. The system goes online August 4th. Human decisions are removed from strategic defense. Skynet begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plug.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like