AI Downs 'Top Gun' Pilot In Dogfights (dailymail.co.uk)

schwit1 writes from a report via Daily Mail: [Daily Mail reports:] "The Artificial intelligence (AI) developed by a University of Cincinnati doctoral graduate was recently assessed by retired USAF Colonel Gene Lee -- who holds extensive aerial combat experience as an instructor and Air Battle Manager with considerable fighter aircraft expertise. He took on the software in a simulator. Lee was not able to score a kill after repeated attempts. He was shot out of the air every time during protracted engagements, and according to Lee, is 'the most aggressive, responsive, dynamic and credible AI I've seen to date.'" And why is the US still throwing money at the F35, unless it can be flown without pilots. The AI, dubbed ALPHA, features a genetic fuzzy tree decision-making system, which is a subtype of fuzzy logic algorithms. The system breaks larger tasks into smaller tasks, which include high-level tactics, firing, evasion, and defensiveness. It can calculate the best maneuvers in various, changing environments over 250 times faster than its human opponent can blink. Lee says, "I was surprised at how aware and reactive it was. It seemed to be aware of my intentions and reacting instantly to my changes in flight and my missile deployment. It knew how to defeat the shot I was taking. It moved instantly between defensive and offensive actions as needed."

Comments

Unsurprising

[fredgiblet]

It was only a matter of time, computers are able to keep complete situational awareness while analyzing what the target is doing. The only question is how long until we can trust them to work totally autonomously. THAT probably won't come for a while.

Re:Unsurprising

[Anonymous Coward]

Completely unsurprising since game bots have been able to outmaneuver human players for decades now. The only thing game bots were lacking was adequate sensor input to gain area awareness in the real world without oversimplified preprocessed maps and precisely placed path nodes.

Re: Unsurprising

[jcr]

Maybe, and here's a concept, we can outfit piloted planes with systems to blow this lab environment victor out of the real skies.

Or maybe the fact that human beings can't stand the kind of acceleration levels that have no effect at all on computers will make this whole question moot.

-jcr

Re: Unsurprising

[jcr]

If the drone can pull 20G turns, it's game over for the human pilot.

-jcr

Re: Unsurprising

[arth1]

Humans doing less dangerous and menial jobs is a good thing, not a bad thing.

That's dogmatic, and not necessarily true.

I would think that humans doing dangerous things for which there are rewards[*] helps provide an evolutionary pressure against those not doing dangerous things, and those failing at them.

[*]: Primary, as in winning wars, or secondary, as in being better paid than average or attracting more mates.

That you can toss a wrapper into the wastebin from across the room, that you can walk for miles, and that you can balance on a bike are likely all because of your ancestors doing dangerous things. It paid off.

As for menial tasks, the same applies, Being good at those too lends an advantage.

We have this big thing on top of our necks, and really complicated protein factory patterns. We can afford to be good at a lot of things, much more so than most of our cousin species. But that's only to our advantage if we do become good at things, and fill that squishy bulb.
I firmly believe that that includes doing both dangerous and menial things.

Which is why I'm now getting into my car, challenging death on the county road to do menial tasks like benchmarking at work. Have a nice day!

Re: Unsurprising

[Richard_at_work]

Do you realise that modern military aircraft *already* identify targets on radar and through the HUD, and present them to the pilot as such? The onboard avionics already highlight to the pilot the ideal point at which to shoot (literally, on the F/A-18 the box on the HUD turns from a square to a diamond and presents the word "SHOOT" underneath it).

Onboard avionics targeting systems are already advanced beyond the state which you think they lack.

Re: Unsurprising

[Somebody Is Using My]

Plus, most missiles don't actually have that much maneuvering capability. They are usually solid-fuel boosters so you can't throttle the thrust significantly and their tiny winglets are more to keep them stabilized than to help them turn (in fact, most missiles only have an initial boost and then glide the rest of the way to their target). It's a commonly used trope in Hollywood to have missiles unerringly follow the Ace Hero Fighter Pilot as he does Immelmans and S-turns and daringly weaves through the narrow canyon with the missile just seconds behind, but that is nothing like real life. A missile's main advantage is its speed; it closes on you faster than you can maneuver out of its vision cone, but if you manage that you've usually beaten the weapon. Ground-to-air missiles are even more limited because so much of their thrust is wasted just getting the weapon up to speed and altitude.

It is possible to make a missile that could be more aggressive (longer thrust, better maneuverability), but this would drive the cost up of the weapon significantly; you would essentially be building a kamikaze aircraft, which is an expensive way to take down another plane. If you are going to make an autonomous drone with that sort of chase capability, better to make it re-usable and then hang cheaper, stupider weapons off of /that/.

Perhaps the future is fighters carrying drones carrying missiles? ;-)

Re:Unsurprising

[The Grim Reefer]

Cars will have no windows. Why, if you could watch Netflix instead?

So will these cars of yours have vomit receptacles built in too? Motion sickness will start to become a more common problem without windows.

Re:Unsurprising

[brianwski]

> will not surrender control to a computer

I think the kaur (the user you were responding to) is wrong, buses and airplanes have windows you can open to watch the interesting and colorful world go by - in addition to window shades if you want to watch Netflix - it will be your choice. But you are also wrong, you already surrender control to a computer when it lands the commercial aircraft you are riding in. You even surrender control to your ABS brakes (occasionally) in your car which make better and faster decisions than you can about which ONE of your four car wheels to brake 10 times a second.

I see a bright happy future where I am actively enjoying the scenery and actively suggesting to the car where to go, but the car will "kick in" and avoid running over a small child or deer in the road faster than my human reflexes could manage. In my 35 years of driving (every day commuter here) I still managed to let my attention waiver once and got in a minor accident (my fault).The average driver gets in 3 or 4 accidents, so I think I'm still "above average" in my driving, but some day a computer will be able to do better than I can in avoiding accidents. I look forward to the help.

Re:Unsurprising

[lgw]

Motion sickness will start to become a more common problem without windows.

I thought I was the only one with that first reaction to Linux.

Re: Unsurprising

[bytestorm]

Sometimes you don't want to put a 450 kg warhead through somebody's window when an 8 kg warhead will do.
Sometimes you don't know how many targets there are until you're near the target.
Sometimes you need to use additional missiles if the first wasn't sufficient and can't afford the non-trivial flight time for a second launch.
Sometimes you want to go home without blowing things up and without wasting 1.6M USD.

There are advantages to having a reusable launch platform in the area, whether that be a UAV or a strike fighter.

Re: Unsurprising

[rocket rancher]

Just...no. There is a fixed amount of energy available to airborne objects in a dogfight, and most of it comes from the initial velocities of the objects at the start of the encounter. Think of it like a mana pool for your caster class -- missiles just sip it while fighters gulp it down. Each new vector acquired by an aircraft or missile bleeds off available energy, so encounters are necessarily brief. And missiles have another big advantage that is energy related: You can always fire another missile, which starts with a refreshed mana pool. The fighter's mana pool never gets refreshed.

So...the push to make fighters more maneuverable was to evade missile threats from the ground and air. Forward canards, vectored thrust, and variable geometry wings were developed to decrease the amount of energy required for a given change in vector required to defend against missiles, whose significantly smaller mass moment arms (four orders of magnitude smaller) made them inherently more maneuverable. And while it is (read: was) true that defeating the first several generations of missiles was possible by knowing and evading their ever-increasing sensor cone, that is most emphatically no longer the case, and hasn't been for a decade. During my time at the rocket ranch in the late nineties-early 2000s, I saw videos of Russian air-to-air weapons systems that made the fighter types in the briefings gulp in dismay. Passive (stealth) and active ECM are the only ways we have of defeating these current threats if we insist on having big, energy gulping objects that need to defend against smaller, more maneuverable objects that only sip at the available energy pool.

And don't discount the notion of disposability -- missiles, after all, are by definition disposable. But a kinetic kill doesn't necessarily mean that *both* objects have to be destroyed in a given encounter. A hypersonic missile equipped with a chaff ejector stuffed with depleted uranium ball bearings instead of magnesium can deliver enough energy against the cockpit of a fighter (structurally the weakest point because of human pilots' need to see with their own eyes) to guarantee a kill (literally, in this case.) And it probably still has enough energy to find and attack another target or three, effectively nullifying your kamikaze-aircraft-is-too-expensive disposability argument.

Prevent the Software From Bein Subverted

[Anonymous Coward]

So maintaining air superiority now becomes an IT security issue.

"He took on the software in a simulator"

[Anonymous Coward]

Translation: he took on the software in its version of reality, with it either being omniscient or having a perfect model of its sensors' deficiencies. While having to work with its presentation of its reality filtered through its presentation devices, limiting the information available to everything the simulator builders considered important enough to bother with and which are actually physically presentable (good luck with proper accelerations, for example).

Re:"He took on the software in a simulator"

[AmiMoJo]

I don't know why you are surprised that the computer is better. Aside from anything else, it will be able to push the aircraft to the absolute limit of performance without blacking out due to G forces. All modern jets rely on computers to distil sensor data down to something that the pilot can process at a much slower rate than the machine can anyway.

The simulators are pretty good actually. They spend a lot of effort making the computer controlled opponents realistic in terms of sensor capability. If anything the human has an advantage here, since acceleration induced blackouts are not simulated.

Why are we still using Human Pilots?

[dwillden]

Because when we automate war and remove the risk of losses on our side, it becomes too easy to just throw more robots into a situation. War is not something that should be automated, we need to retain the potential of real losses to restrain our desire to engage in war. Even extensive use of drones is taking us dangerously down that path. We can kill those who oppose or offend us without risk of our own losses and thus we have little cause for showing restraint in using such equipment to conduct our foreign policy.

Oh and Skynet!!!

G-force limits, too

[Richard Kirk]

This is probably old data, but few pilots in special, elasticated suits can get beyond 10g without blacking out. As we approach our limit, our peripheral vision goes, so even if we don't black out, we are not working well if we keep this up for long. It is possible to make conventional airframes that can take 25g if you don't have to cut big holes in the airframe for the cockpit. So, a computer in a plane built for a computer ought to rule.

Re:G-force limits, too

[MTEK]

Hybrid solution, though not something I'd want to sign up for...

• 1. Pilot identifies threat aircraft.
• 2. Pilot engages combat AI.
• 3. Pilot wakes up fives minutes later with a headache and a kill.

AIs don't have G-force limits

[PvtVoid]

It's worse than that: the AI in this test won when piloting evenly matched planes. But the weak point in modern fighter jet design is the squishy fragile thing in the cockpit, which can't take more than 8 g-s or so, and not even close to that for negative g-forces. Get rid of the pilot, and you can design a plane whose performance is vastly better than a piloted plane. Now put that AI in it and send it head-to-head against an F-35. No contest.

Not a real world test

[sjbe]

He took on the software in a simulator.

So he was fighting in a computer game, not in a real jet and certainly not in real combat conditions. This is a limited scenario with limited conditions. Keep this in mind.

And why is the US still throwing money at the F35, unless it can be flown without pilots.

See above. There is a HUGE difference between a computer game and flying a real jet in combat conditions. We've had computer "AI" (using the term loosely) that could beat people at games for a long time. That isn't the same thing as having an AI that is ready for real world combat and it is even further from having an AI that is trustworthy on decisions of whether to shoot or not. To the best of my knowledge we do not presently nor are we likely to any time soon have an AI that we can or should trust to make judgements about what to shoot or when to shoot it. It's not clear to me that we ever can or should take humans out of that loop. It might be necessary to take them out of the vehicle physically (what with us being bags of fluid and all) but we'd be idiots to trust any current AI with complete control of combat.

Furthermore an F35 does a lot more than just dog fighting. In fact its primary role is likely to be air to ground combat far more often than air to air. That's why they call it a Strike Fighter. I'm not moving the goal posts here either. Yes it is reasonable that a computer AI could outperform a human in air combat maneuvering. Especially when the jet doesn't have a human on board with the physical limitations of a human, particularly in relation to G forces. We've had jets for decades that can generate more g forces than a human can handle and we've had to artificially limit them. The problem is that we still need humans in the loop for decision making and for the most part that is a good thing. Even our drones don't shoot automatically because we cannot trust them to make appropriate firing decisions in most cases.

Most A/A kills result from not being seen

[bkmoore]

Since the first world war, most air to air kills were scored against opponents that did not see their attacker. The preferred tactic was to come out of the sun or attack from a blind spot. The Red Baron stated, "I get real close, pull the trigger, and he blows up", or something to that effect. An AI- piloted airplane would have this same limitation, as it would only be aware of what its sensors tell it. If you jam its on board sensors and data-link capability, all that AI won't be worth anything. What this has to do with the F-35, I don't know? Unless it's just to flame an airplane that a lot of arm-chair experts don't like. There are lots of missions for a manned airplane, and "dogfighting" (or BFM) is a tactic and not not a mission.

Fighters becoming an anachronism, like horse cav

[perpenso]

A lot of people overinterpret the lessons of the Korean war where missiles were overstressed versus the technology of the time... and have taken it as some universal lesson which will apply forever into the future, that close-range dogfighting will always be the most critical aspect of aircraft design.

Vietnam not Korea. Personally I expect the AI to go into the missiles not the aircraft. Fighters becoming a romantic anachronism, like horse cavalry. And like horse cavalry they will last longer than people expect. My local National Guard unit is cavalry, reconnaissance, and had horse as late as the 1930s. In certain terrain guys sneaking around on horse was still more effective than vehicles. They were just the eyes for armored formations and not expect to fight themselves. Sort of like modern Long Range Reconnaissance Patrols, if you are firing your weapons something has gone terribly wrong. Note some US Special Forces briefly operated as cavalry in Afghanistan. I believe the US Marines sometimes use dirt bikes. Recon may also be a role for repurposed fighters. Actually it has been such a role, removing guns an armor and adding cameras. Sometime there are gaps with satellite and drone coverage and a fast mover flying low and masking its approach with terrain fills that gap. A role not unlike that 1930s horse cavalry role, eyes, not direct combat.