El Reg reports
Today's generation of fighter pilots could be the last of their breed, thanks to an [Artificial Intelligence] 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 Raspberry Pi computer--was deliberately handicapped but still managed to shoot down its fleshy opponent every time and evade his attempts to kill it.
[...] To make the defeat even more humiliating, the ALPHA AI's fighters were deliberately handicapped with shorter-range missiles and fewer of them, and its opponents got additional intelligence from a simulated AWACS radar aircraft.
[...] The key to the software is the use of a "Genetic Fuzzy Tree" algorithm[PDF], which breaks down complex decisions into simpler if/then scenarios and narrows down the possible choices very quickly. The code then works through its plans 250 times faster than a human can blink.
(Score: 0) by Anonymous Coward on Wednesday June 29 2016, @11:07PM
You've got a fancy video game, you run it against a retired pilot, and herald your futuristic advanced AI!!!!
Was there a boss level, at least?
(Score: 0) by Anonymous Coward on Wednesday June 29 2016, @11:35PM
Some coder makes a typo, boss fight is player against Obama.
(Score: 5, Insightful) by Zz9zZ on Wednesday June 29 2016, @11:56PM
ALPHA AI's fighters were deliberately handicapped with shorter-range missiles and fewer of them, and its opponents got additional intelligence from a simulated AWACS radar aircraft
Sure it would need real-world testing, sure it would need to beat a top notch pilot in their prime, but that doesn't make this test invalid. The Colonel may be older and retired, but is probably worth at least 3/4 of a top pilot. If the initial test was so successful, on crappy hardware nonetheless, then you can bet your ass the AI is going to win in real life scenarios once they get the engineering locked down.
I think the real discussion should be about whether putting an AI in control of a lethal weapon is even a good idea.
~Tilting at windmills~
(Score: 2) by The Mighty Buzzard on Thursday June 30 2016, @01:10AM
Depends on how foolproof you can make it. If it meets or exceeds human rate of error, I don't see a problem with it, so long as a human has a remote kill switch.
Now some of you may note that this differs greatly with my self-driving cars stance. There's a reason. Who makes the decision. If an AI fighter plane has been sent out, it's been sent out with the express intention of killing its target. Humans made the life and death choice. Every choice a car makes is a life or death choice though and it's being left up to algorithms instead of any human agency.
My rights don't end where your fear begins.
(Score: 3, Touché) by Anonymous Coward on Thursday June 30 2016, @01:22AM
> remote kill switch
Also desirable would be a remote no-kill switch.
(Score: 4, Insightful) by Francis on Thursday June 30 2016, @02:35AM
The problem is that automating the weapons we reduce the incentive to avoid armed conflict and even if the weapon systems never go all Skynet on our asses, you've still got a ton of dead civilians.
If we're going to turn to robots to do our fighting for us, we should be going non-lethal and just agree to abide by the results in terms of winners and losers rather than deal with the consequences of the weapons hitting unintended targets.
(Score: 3, Interesting) by Fnord666 on Thursday June 30 2016, @02:52AM
If we're going to turn to robots to do our fighting for us, we should be going non-lethal and just agree to abide by the results in terms of winners and losers rather than deal with the consequences of the weapons hitting unintended targets.
We already know how that turns out! Season 1 Episode 23: A Taste of Armageddon [wikipedia.org].
(Score: 2) by The Mighty Buzzard on Thursday June 30 2016, @10:25AM
It's a nice thought but it's about as viable as Sanders was a candidate. War isn't about resolving differences without pain. It's about forcing the enemy to bend to your will through death and terror. Which is why it should be used only when absolutely necessary.
My rights don't end where your fear begins.
(Score: 2) by Zz9zZ on Thursday June 30 2016, @06:39AM
I'm not confident that a fly-by-wire couldn't be hijacked, and the good ol' skynet argument always comes up too.
~Tilting at windmills~
(Score: 2) by The Mighty Buzzard on Thursday June 30 2016, @10:27AM
Well, the Skynet option is easily avoidable. Don't network the CnC machines together. At worst you get multiple but individual Skynet wannabees.
My rights don't end where your fear begins.
(Score: 2) by Zz9zZ on Thursday June 30 2016, @09:32PM
True, but one of the big problems is that networking is a major component, and AI will probably be able to crack through any software defenses. As for keeping the CnC machines off the grid, I bet the AI can get around that part too, even coercing a human to do its bidding. Interesting times, we shall see how it pans out.
~Tilting at windmills~
(Score: 3, Insightful) by takyon on Wednesday June 29 2016, @11:40PM
I glanced at this story yesterday, but figured the "AI" was running on a Xeon at least. RasPi 3? Goodbye humans.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2, Informative) by Anonymous Coward on Wednesday June 29 2016, @11:47PM
Try playing Quake on a Pentium 90. Humans never had a chance.
(Score: 5, Informative) by caffeine on Thursday June 30 2016, @12:10AM
And consider that they can cut out the weight of the cockpit and support systems, and increase the turn ability as there is no human to pass out from the g forces. Only counter is another AI.
(Score: 3, Informative) by dyingtolive on Thursday June 30 2016, @12:33AM
Or an EMP. Or hacking. Imagine finding some way to commandeer a bunch of F-22s in the air.
Admittedly, that's the stuff of sci-fi today, but autonomous planes were the stuff of sci-fi 20 years ago, and here we are.
Don't blame me, I voted for moose wang!
(Score: 3, Interesting) by takyon on Thursday June 30 2016, @01:13AM
Require all remote instructions to be signed by the correct private encryption key, and make it single use per mission. Program most of a mission in advance, allow the objectives to be completed autonomously (AI makes some adjustments on its own, such as dodging obstacles or routing around bad weather).
EMP hardened electronics including computing elements are a possibility (see the nanoscale [sciencemag.org] vacuum tubes [ieee.org] from a while back). I'm not sure all of today's planes can fly without electronics - like flying wings [wikipedia.org].
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by dyingtolive on Thursday June 30 2016, @01:19AM
Oh yeah, I know we have defenses for this kind of stuff, but war escalates. Wait until they can evesdrop based upon the churn of the turbines like fan noise nowadays, apparently, which seems to be a sonic version of Van Eck phreaking, but I'm no expert.
Don't blame me, I voted for moose wang!
(Score: 1, Insightful) by Anonymous Coward on Thursday June 30 2016, @01:35AM
(Score: -1, Offtopic) by Anonymous Coward on Thursday June 30 2016, @03:01AM
But what if I don't want an F-35 and would rather have an A-10? Only F-35s get to be a part of the internet of appy things? Feh.
(Score: 4, Insightful) by tonyPick on Thursday June 30 2016, @06:16AM
I suspect that modern fighters have such a reliance on software, that if an EMP can take out the core processing you're dead anyway. And hacking the IFF system would be as bad with humans in the driving seat as the AI.
(Going from conversations from way back with people who knew far more about control systems than me, admittedly)
(Score: 2) by bob_super on Thursday June 30 2016, @03:40PM
> Imagine finding some way to commandeer a bunch of F-22s in the air.
Well, given the shitty range, reliability, "don't use them near enemies" price, and inability to attack ground targets, that's about a best-case scenario there.
As long as you're not inside a US airliner or tall building.
(Score: 2) by DannyB on Thursday June 30 2016, @05:30PM
Ballistic Missiles have that name for a reason. Ronald Regan didn't seem to understand that you cannot recall nuclear missiles once they have been launched. (Which seems like something that should be well understood by the person who has the power to launch them.)
That is why these fighters should behave more like that. They should not accept remote commands. Period. You launch them. They fly to a mission site. Accomplish their mission. And return home if possible. No possible interference from an enemy.
If you think it is necessary to accept remote commands, then you might restrict the territory the aircraft must be in for a command to be accepted. The aircraft would have to be able to verify its location based on things it can independently sense, not subject to enemy interference. The position of the sun compared with the time. The stars. Topographical land maps in an on board database. (It's not exactly easy for an enemy to change the terrain of large amounts of land.)
I understand the idea of using signed encrypted messages. I suspect that would become the attack point for an enemy. Compromise the security and ability to sign the commands well in advance of the start of a conflict. I suppose the key-pair could be generated right before mission launch, and one key is inserted into the aircraft right before takeoff.
To transfer files: right-click on file, pick Copy. Unplug mouse, plug mouse into other computer. Right-click, paste.
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @09:00PM
No ability to recall? hey didn't they make a movie about that back in the '60's?
https://en.wikipedia.org/wiki/Dr._Strangelove [wikipedia.org]
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @12:08PM
ya, i was thinking the same. it was NOT on a pi. it could be ( and would be slower ), but wasn't. I still don't know why the PI was brought up. Its also a computer, so of course it can run code.. They could have said 'could run on a tablet' just as well.
(Score: 1, Interesting) by Anonymous Coward on Thursday June 30 2016, @12:06AM
WWII style dogfights are supposed to be obsolete given long range missiles.
So what's this useful for?
Perhaps a old style dogfight assistant autopilot for a manned fighter.
Perhaps a smart drone which is a pilot's nightmare.
In the test, I wonder how the AI engine got information about the opponent.
On a simulator, perfect state vectors are just sitting there.
Did the engine have these, or did the test also simulate some real sensor plan?
(Score: 3, Interesting) by khallow on Thursday June 30 2016, @12:22AM
WWII style dogfights are supposed to be obsolete given long range missiles.
If your ability to maneuver is sufficiently awesome, you can dogfight those missiles just like any other flying object. Also, when I brought this up on the other site, a replier claimed that a lot of current day missiles can be outmaneuvered with rather simple dogfighting techniques (they gave the example of drop in altitude as you turn towards the missile and fly past the missile from underneath. Apparently, this causes a number of designs to crash into the ground - presumably if you were relatively close to the ground to begin with).
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @12:35AM
So maybe the place for the engine is in the missile?
(Score: 1) by khallow on Thursday June 30 2016, @02:22AM
(Score: 3, Interesting) by Arik on Thursday June 30 2016, @03:11AM
The trouble with that line is that even the best manned plane can't come anywhere near the maneuverability (or speed) of a modern air to air missile.
It's not impossible to do, there are tricks like the one you described, which I don't doubt would work against the right missile and with a low enough starting altitude, but it's more likely to be a fluke than a viable plan.
Conventional wisdom, with considerable logic behind it, is that dogfighting is obsolete. Long range missiles and radars are too effective for it to happen, at least on any regular basis.
If laughter is the best medicine, who are the best doctors?
(Score: 1) by khallow on Thursday June 30 2016, @03:22AM
The trouble with that line is that even the best manned plane can't come anywhere near the maneuverability (or speed) of a modern air to air missile.
But note that we are speaking of an unmanned airplane with far better maneuverability and speed possible and which can react to things like an incoming modern air to air missile far faster than a human can.
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @05:07AM
and, may be able to jink out of the way of the missile, but... proximity fuzes...different story.
(Score: 1) by khallow on Thursday June 30 2016, @01:34PM
but... proximity fuzes...
Depends what the fuse is attached to. Conventional explosives count only if the missile can get close enough. Nuclear tipped missiles wouldn't need to get as close.
(Score: 1, Interesting) by Anonymous Coward on Thursday June 30 2016, @05:07AM
So what if they put that AI into missiles? There are no human pilots in missiles, for obvious reasons. Those reasons obviously don't apply to AIs.
(Score: 2) by fritsd on Thursday June 30 2016, @08:09AM
Fiat Lux! [wikipedia.org]
(Score: 2) by tibman on Thursday June 30 2016, @02:41PM
I'm going to call BS on outmaneuvering missiles. Just looked up a modern-ish missile from 2005: https://en.wikipedia.org/wiki/IRIS-T [wikipedia.org] Speed of mach 3 with a turn rate of 60 degrees at 60 gravities. It has a proximity fuze and fires shrapnel. It has such maneuverability that you can fire this missile at planes behind you.
Now i would completely agree that missiles can be outrun or confused (with counter measures). But not outmaneuvered.
SN won't survive on lurkers alone. Write comments.
(Score: 1) by khallow on Thursday June 30 2016, @06:25PM
I'm going to call BS on outmaneuvering missiles.
I'm not. The AI will have similar maneuvering capability. And the AI can do more to that missile than just attempt to outmaneuver it.
(Score: 2) by tibman on Thursday June 30 2016, @06:58PM
Think about it like this. The AI in a missile can be exactly as advanced as the AI in a fighter. Only the missile can move faster and turn harder. The only thing a fighter can do that a missile cannot is use some utility that is too bulky for a missile to carry. ECM, for example. But for maneuverability the missile will beat the fighter.
SN won't survive on lurkers alone. Write comments.
(Score: 1) by khallow on Thursday June 30 2016, @07:27PM
The AI in a missile can be exactly as advanced as the AI in a fighter. Only the missile can move faster and turn harder.
Why would that be true? There are big performance compromises like being capable of traveling a long distance (after all, it's supposedly the end of the dogfighting era right?). And the sensors that a missile will contain will be limited. The worst of all is simply that when you launch something that slow from a distance, you telegraph not only that particular strike, but to some degree your intent. You can bluff, feint, and whatnot, but it's still information into your intentions and strategies.
(Score: 2) by tibman on Thursday June 30 2016, @07:59PM
Unless the fighter is faster (mach 3 in the example above) or can turn sharper (60 degrees per second in the example above) then that fighter simply cannot escape if we are talking about performance comparison. If you are talking about computing power then that is what this article is about.
I'm not sure what that has to do with a fighter outmaneuvering a missile? But i wouldn't describe something travelling 1 mile every 2 seconds as slow. That missile is literally faster than a .50 cal bullet max velocity. Faster than a bullet. If someone shot at you would you describe the event as "they launched something slow at me"? As for the distance part of this problem i only need to point out the sharp turning radius. You'd have to get some extreme perpendicular speed (in reference to the missile direction) to get away. But! I totally believe fighters can escape a missile by out running it. Outmaneuvering, however, looks like BS to me.
SN won't survive on lurkers alone. Write comments.
(Score: 1) by khallow on Thursday June 30 2016, @08:33PM
Unless the fighter is faster (mach 3 in the example above) or can turn sharper (60 degrees per second in the example above) then that fighter simply cannot escape if we are talking about performance comparison.
Which I think is reasonable for an AI fighter to do. For example, at 100 gee of acceleration, the fighter can accelerate to somewhere near mach 6 in two seconds. Of course, so would those more advanced missiles.
I'm not sure what that has to do with a fighter outmaneuvering a missile?
Well, for example, that's almost a minute of time (at 100 miles away) to either lose targeting (say by flying into an urban area or into a valley), reconfigure the vehicle to become more maneuverable, and/or launch countermeasures.
I think also that this discussion ignores important modes in how AI piloted craft are likely to be deployed. A large swarm would preclude targeting of individual aircraft. Swarm-based dogfighting is going to be important because a lot of throw away long range missiles that don't hit anything aren't going to be a good use of a foe's resources. At the least, you'll need some local intel to determine the best targets to take out. Local intel needs protection and needs to be taken out at short range.
(Score: 2) by MostCynical on Thursday June 30 2016, @12:29AM
If you can successfully avoid a plane or missile, you are a long way to hitting it.
If the AI managed to shoot down the General, the next step will be shooting missiles out of the air.
The idea behind Reagan's Star Wars missile defence system will finally work (but without the lasers, or satillites, or cost).
"I guess once you start doubting, there's no end to it." -Batou, Ghost in the Shell: Stand Alone Complex
(Score: 2) by takyon on Thursday June 30 2016, @01:53AM
Every new missile should be equipped with a RasPi 3 powered AI and autonomous maneuvering capability. Price tag? +$1 million per missile.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 3, Informative) by MostCynical on Thursday June 30 2016, @02:26AM
Reaper drone costs $28 million; one Hellfire missile (Lockheed Martin/Raytheon) costs about $70,000; one Paveway bomb (Lockheed Martin/Raytheon) about $20,000. The total cost of one weapons load for a Reaper – four Hellfire/ two Paveway – is at least $320,000
www.knowdrones.com/FAQ.pdf
Is only money.
"I guess once you start doubting, there's no end to it." -Batou, Ghost in the Shell: Stand Alone Complex
(Score: 4, Insightful) by Francis on Thursday June 30 2016, @02:40AM
Of course there's not budgeting on that, we always have money to spend on weapons to kill brown people. It's things like social safety nets and things that improve people's lives that are too expensive for our budget.
(Score: 1, Interesting) by Anonymous Coward on Thursday June 30 2016, @07:17AM
If instead of spending billions (trillions?) blowing people up (at wedding parties too, how rude...), you spent billions/trillions on making friends (or reducing enemies) you might have got better results for your country as a whole.
After all imagine if you didn't bomb a single village or wedding party, but dropped/gave them lots of nice stuff and then various extremists try to tell villages that the US are evil, and that the villagers should send some volunteers to the USA to kill people... They'd be like, sure after we're done playing with the toys or need new ones... Even if you "bomb" them with another load of toys every 5 years for a decade or two it'll still be less than what your military has spent. One laser guided bomb or smart missile costs about as much as buying a mobile phone for each family in an entire village. One cruise missile is even more... It's a lot easier to convince people that you are evil and they should fight you if you are blowing up wedding parties and kids than if you're giving them phones, food and books (even if the books are a "corrupting influence").
But I suppose the real objective is to make more enemies so that the US Gov and friends can get more money from the taxpayers and the rest of the world (via the petrodollar).
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @12:27PM
I do think the initial Afgan response and tracking down UBL were constructive, necessary responses to 9-11.
But that seems a small part of that has happened.
Operation Iraqi Liberation, even Jeb Bush might say not so useful.
A ham fisted attempt both at making things 'better' in Iraq and also at making the us more 'secure'.
I'd like to think the recent energy sector changes have have lowered the odds of a repeat performance.
The idea that the war on terrorism should be fought with peace corps instead of the military is not new, but at this point it might take a couple of generations before this would make sense in the world the current strategy has created.
Also there is the cynical/pragmatic view Eisenhower predicted of 'think of the poor military industrial complex'.
Strategies seem to last as long as Presidential terms.
We have moved from not so well thought out war to don't do dumb stuff (in public).
Certainly an improvement, but a ways to go still.
Perhaps the next step is to corral some cowboys?
I'm fond of Churchill's quote that the US can be counted to do the right thing after exhausting all the other options.
Hopefully the world will survive the wait.
(Score: 2) by lentilla on Thursday June 30 2016, @05:26AM
Price tag? +$1 million per missile
Yeah, that's about the cost of a $35 Raspberry, once one adds the SD card, the power supply, the cables and all the rest of the required paraphernalia :-)
(Score: 2) by takyon on Thursday June 30 2016, @05:31AM
Savage.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @09:03AM
You forgot to add in the price of an executive handshake with a man wearing a suit.
(Score: 2) by MostCynical on Thursday June 30 2016, @10:41AM
But you don't *have* to use Monster branded HDMI cables.
"I guess once you start doubting, there's no end to it." -Batou, Ghost in the Shell: Stand Alone Complex
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @11:55AM
But without those cables, the sound of the explosion will be far less authentic!
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @04:09AM
Perhaps a place for this AI is on drones that protect manned aircraft by outmaneuvering threats.
A fair question is, once you have the drones, what were the manned planes for again?
Maybe the primary purpose of an air force.
Situational awareness, or transporting something like people or kaboom for a surface effort.
Which would make AI drones just a tool to make the air force better able to project force and prevent others from doing same to us.
I'd still like to know how the AI was receiving threat state in the simulation.
That would give a better sense of if this stuff is real or not.
(Score: 2) by PartTimeZombie on Thursday June 30 2016, @03:28AM
WWII style dogfights are supposed to be obsolete given long range missiles
WWII style dog fights were supposed to be obsolete by the time of the first Jets.
WWII style dog fights were supposed to be obsolete before WWII because of the new "fast" bombers, like the Fairey Battle that would fly too high and fast for Fighters to even catch.
(Score: 4, Touché) by toygeek on Thursday June 30 2016, @12:49AM
Well, you know...
There is no Sig. Okay, maybe a short one. http://miscdotgeek.com
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @01:40AM
http://venturebeat.com/2011/12/15/drone-iran-gps-aclu/ [venturebeat.com]
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @12:03PM
Its being paid for by tax dollars, so where is it?
(Score: 0) by Anonymous Coward on Thursday June 30 2016, @12:55PM
A suitable pilot for the F35
(Score: 2) by Bot on Friday July 01 2016, @11:32PM
An AI worth its salt would absorb info about the F35, and soon start bombing defense contractors, perceived as the real enemy.
Account abandoned.