AI software beats real pilot in simulated dogfight


The Terminator is coming: In a DARPA competition between a number of AI software teams, the finalist AI team, called Heron Systems, went up against a real F-16 pilot in a simulated dogfight and went undefeated, beating him five times in a row.

Heron Systems, a company with just 30 employees, had beaten out Aurora Flight Sciences, EpiSys Science, Georgia Tech Research Institute, Lockheed Martin, Perspecta Labs, PhysicsAI, and SoarTech to claim the top spot in the last of Defense Advanced Research Projects Agency’s (DARPA) AlphaDogfight Trials. This three-day event had started on Aug. 18, 2020.

The software then beat the real pilot.

There is a lot of blather on the website, though there is this interesting analysis by F/A-18 Squadron commander that in the end concludes that we are only a generation or two away from making all fighter pilots obsolete.

Or to look at it from another perspective, we are only a short time away from putting the ability to fight war entirely into the hands of computers and software, with abilities that humans will not be able to match.

Does no one but me see the potential problems with this? Have all of these military experts never seen any science fiction movies or read any science fiction novels?

Hat tip Tom Biggar.

28 comments

  • V-Man

    The problem is if we choose not to do this, future opponents will. Kinda like the Vernor Vinge story, “Bookworm, Run!” where they mull the ethics of a new technology… and then hear the Soviets have a similar prototype already active.

    Damned if you do, damned if you don’t.

  • Andrew_W

    Former Fighter pilot C.W. Lemoine isn’t convinced it was a fair fight, and makes some good points about the flaws in the simulation. But in the long term, if the issues he raises are overcome, I think it’s certain that in the longer term it can’t be anything but pilotless fighters reigning supreme.

    https://www.youtube.com/watch?v=ziCQqmEllZo
    [28:47]

  • V-Man: No one remembers that after World War I, everyone from all sides resolved to never use gas in warfare again, and thus even Hitler and Germans did not do it in World War II.

    These things can be worked out, though it seems humans are only willing to do so after making some horrible mistakes and suffering horribly for it.

  • Andrew_W

    . . . all sides resolved to never use gas in warfare again,. .

    Why?

  • pzatchok

    I can see semi autonomous military equipment as the future of a modern warfare.

    We can now push the limit of fighters to the materials limit and not the limit of the pilot.
    Or get rid of all fighters and go with missiles in their place.
    We will soon only need aircraft as transports.

  • Jerry Greenwood

    When was the last combat dog fight?

  • commodude

    Patton: Wonder weapons? My God, I don’t see the wonder in them. Killing without heroics. Nothing is glorified, nothing is reaffirmed. No heroes, no cowards, no troops. No generals. Only those that are left alive and those that are left… dead.

    Without knowing the parameters of the exercise, it’s almost impossible to know whether this is a meaningful trial.

    Given that the limits of G-loading on modern warbirds are mostly involved with the wetware, however, I’m not surprised.

  • John

    This particular simulation was just a basic exercise for the public and really isn’t applicable to modern aerial combat. I’d be surprised if there hasn’t been much more classified work on the subject.

    Semi-Autonomous UAVs are very likely the future. If you can eliminate the cost in money, weight, and performance of keeping a meat sack alive, the aircraft become that much more capable. The real AI adversary will not be another viper, it will have superior range, maneuverability, and/or payload.

    As long as humans are kept in the decision loop then it’s really no different than current stand-off weapons. The cruise missile decides how and where to fly and detonate, according to the humans who programmed it.

  • Col Beausabre

    About 10 years ago, a speaker at the Air Force Association’s annual meeting stunned the audience by proclaiming, “The world’s last fighter pilot has already been born.”

    This year, Elon Musk ground the USAF’s face in it.

    “With six words, Elon Musk sent a shudder throughout the great hall in which thousands of Airmen and aerospace industry officials had gathered to hear his comments at the conclusion of AFA’s annual Air Warfare Symposium in February. “The fighter jet era has passed.”

    https://www.airforcemag.com/article/the-fighter-jet-era-has-passed/

    The response to Musk has been the greatest mass howling of military professionals since the US Army declared the horse cavalry obsolete in 1942

    A good primer on the issue https://www.popsci.com/last-fighter-pilot/

    In part answer to the question “When was the last combat dog fight?”

    “As networked warfare has risen, incidents of aerial combat have decreased. Since 1990, only 54 fighter jets have been shot down globally, says John Stillion, a senior fellow at the Center for Strategic and Budgetary Assessments, and a former Air Force officer, who put together a database on all confirmed aerial victories between 1965 to 2013.”

    From

    https://csbaonline.org/uploads/documents/Air-to-Air-Report-.pdf

  • Alex Andrite

    “Only the dead have seen the end of War.”

  • pzatchok

    The last gun to gun dog fight was back in the Vietnam war.
    Air to air missiles had a less than 30% chance of hitting let alone killing a target.

    Since then everything has been stand off missiles.

    It has gotten so bad for fighter pilots that the aircraft can now tell when the pilot has lost consciousness and will finish a pre programed maneuver to bring the plane to a safe flight attitude until the pilot wakes back up.(at least at the last I knew)

  • commodude

    pzatchok,

    For the US, maybe, but there have been dogfights in every conflict up to and including the current situation in Syria.
    Sidewinders are used within visual range.

    Just because it’s a missile doesn’t mean there wasn’t close in maneuvering involved.

  • MadRocketSci

    So many dog and pony shows are *just* dog and pony shows. They indicate something that might be possible, but there’s a lot less there under the surface in terms of development than is presented breathlessly in the press.

    I think the real sci-fi scenario that we need to be worried about hasn’t really been belabored in sci-fi. (Maybe the Hunger Games.) Rather than worrying about what the computers want, we should be worried about what the people commanding the computers want. The ability of a small group of high-capital people to field a force that can crush a much larger conventional army in a one-sided slaughter is worrisome. So far, such forces have been under the control of democracies (though increasingly those democracies are looking like unaccountable oligarchies.) Historically, democracies grew when the bulk of the citizens were dangerous to centralized authority, and necessary to field a competitive force. The absence of those two things means there’s no “selective pressure” for democratic governments that respect the rights of the majority.

    I think in order to keep freedom alive in a world like that, we need to democratize the ability to manufacture and field these robotic weapons.

  • pzatchok

    Have you seen the videos of the weaponized drone?

    Someone strapped a 9mm glock to it and aimed with the camera.

  • LocalFluff

    Machine learning (a statistical method) is imitation. All superficial human behavior can be imitated by and outperformed by machine learning. What cannot be imitated is understanding and creativity and motivation.

  • wayne

    Star Trek (original series):
    “The Ultimate Computer”
    https://youtu.be/Q-FNPRfkOg8
    3:05

  • J Fincannon

    To Robert Zimmerman:
    “No one remembers that after World War I, everyone from all sides resolved to never use gas in warfare again, and thus even Hitler and Germans did not do it in World War II.”

    Not quite. https://en.wikipedia.org/wiki/History_of_chemical_warfare Quoted excerpt below:

    The Nazis did use chemical weapons in combat on several occasions along the Black Sea, notably in Sevastopol, where they used toxic smoke to force Russian resistance fighters out of caverns below the city, in violation of the 1925 Geneva Protocol. The Nazis also used asphyxiating gas in the catacombs of Odessa in November 1941, following their capture of the city, and in late May 1942 during the Battle of the Kerch Peninsula in eastern Crimea. Victor Israelyan, a Soviet ambassador, reported that the latter incident was perpetrated by the Wehrmacht’s Chemical Forces and organized by a special detail of SS troops with the help of a field engineer battalion. Chemical Forces General Ochsner reported to German command in June 1942 that a chemical unit had taken part in the battle. After the battle in mid-May 1942, roughly 3,000 Red Army soldiers and Soviet civilians not evacuated by sea were besieged in a series of caves and tunnels in the nearby Adzhimushkay quarry. After holding out for approximately three months, “poison gas was released into the tunnels, killing all but a few score of the Soviet defenders.” Thousands of those killed around Adzhimushkay were documented to have been killed by asphyxiation from gas.

  • wayne

    J Fincannon–
    Interesting stuff.

    The only chemical weapon ‘incident’ I’m aware of—during the Allied invasion of Italy in ’43, they feared the Germans might use gas, the Liberty Ship John Harvey was alleged to be carrying “2,000 M47A1 mustard gas bombs, with 30kgs of agent in each bomb.” A German raid on Bari sunk the John Harvey and 83 people died.

  • LocalFluff

    @J Fincannon
    Well “toxic smoke” and “asphyxiating gas” is commonplace in war situation, not least against millions of civilians in fire bombed cities. How is filling a cave with smoke different from filling it with burning gasoline? They didn’t use mustard gas or nerve gas. The US actually used phosphoric fire bombs that were formally prohibited as chemical warfare. But on the whole mutual deterrence against WMD actually worked in ww2, in Europe. The Japs used mustard gas in China but didn’t dare turn it on the Americans although it should’ve been efficient against beachheads on islands, even using it only in artillery grenades.

    The use of gas against millions of captured civilians, the nazis didn’t consider “warfare”, but rather a “medical” measure for purpose of “population control” administered by MDs… (Who had dangerously unchallenged authority in Germany since decades by then.) Perhaps that was what made Hitler abstain from using, in the field of war, the nerve gas that he was alone to have? Pretending to be against gas usage in order to try to get away with his mass murder in case of a ceasefire. Himself having been injured by chlorine gas in ww1 and the allied air superiority were surely also very important reasons.

  • Spectrum Shift

    Like walking closer and closer to a cliff, how close dare we approach true AI before “Sky Net” actually happens? In the original StarTrek “What Little Girls are Made Of”, the android Roc, awakens with the declaration “Survival cancels programing”. I’m not comfortable with programmers claims that AI will not escape their control. The Chinese thought they had control of Wuhan Goo too, didn’t they?

  • wayne

    Spectrum Shift-
    thanks for the what-little-girls-are-made-of reference. Tangentially, Youtube & CBS are getting more aggressive at blocking choice Star Trek clips.

    LocalFluff–
    you bring up interesting points.
    I think were into a definitional problem here, ‘chemical weapons’ include nerve-agents as well as plain old toxic chemicals. “Zyklon-B” for example, was hydrogen cyanide. The Germans tested it on the battlefield in WW-1 but abandoned it’s use. Much like chlorine gas, very unpredictable and difficult to disseminate effectively in combat.

  • Steve Richter

    repeat this exercise, this time enable the pilot to control a swarm of drone aircraft as his wing men. The pilot stays back a bit, directing the drones in their attack and defense maneuvers. A human piloted plane with 5 drones would obviously defeat a pack of 6 drones.

  • F16 Guy

    I spent 22 years practicing tactics I could use in combat. The aircraft I flew were state of the art, and I never believed I would fail in either air-to-air or air-to-ground combat maneuvers.
    The fighters today (F-22, F-35, advanced F-16) are incredibly capable. That said, technology and AI will render the manned fighter obsolete.
    What we really need to figure out is how to make war obsolete.

  • Edward

    Spectrum Shift wrote: “Like walking closer and closer to a cliff, how close dare we approach true AI before ‘Sky Net’ actually happens? In the original StarTrek ‘What Little Girls are Made Of’, the android Roc, awakens with the declaration ‘Survival cancels programing’.

    Whatever happened to Asimov’s Three Laws of Robotics?

    Asimov did not like the Frankenstein story, so he found a way to avoid the creation turning against its creator. Not so hard to implement, but only if we implement it. AI harming humans violates Asimov’s first law, so AI killing machines are a bad idea.

  • wayne

    Edward–
    here we go….

    “I, Robot”
    Whose Revolution
    https://youtu.be/KLveo_Hswo4
    1:07

    The three laws will lead to one logical outcome…. revolution

  • Edward

    Wayne,
    That was such a disappointing movie. It was the very Frankenstein story that Asimov avoided with his three laws, violating the spirit and the cleverness of his “I Robot” stories. If only that movie’s filmmakers were smart, they would have known not to make a Frankenstein story, and they would have made the story about figuring out how the three laws resulted in unexpected situations. That would have been as clever as Asimov was.

    What a disappointing movie.

  • LocalFluff

    Elon Musk’s brain chip development is interesting. He seems to put much effort in how to robotically connect them to neurons. The robot he talks about will be able to do it at a mass scale! It’s one thing to use such chips to cure blind, lame, clinically depressed, but his ambition obviously goes far further. There’re great dangers with such a technology, imagine how China would use it to control new born. I think its application should be guided by traditional medical ethics.

  • Max

    In the end, the robot series that Isaac Asimov created, starting with the “caves of steel” having the three laws of robotics ended in the final book with the robot reprogramming a “zero law” that precedes all the others allowing him to kill billions of humans who were threatening the rest of humanity. He had to kill humans to save humans from themselves.

Leave a Reply

Your email address will not be published. Required fields are marked *