10 ms·

For those who had not watched the AlphaDogfight competition, the most remarkable thing was not that the AI beat the Human fighter pilot. The most remarkable thi

For those who had not watched the AlphaDogfight competition, the most remarkable thing was not that the AI beat the Human fighter pilot. The most remarkable thing was the extreme gap in performance between the best AI and second and third place AI's. It wasn't even close. An AI that is marginally worse could still be able to beat a human but will lose consistently over and over against a better AI. This bodes very poorly for any nation that falls behind in the AI race. As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms.

belter 5y ago
Nothing a proper EMP weapon will not sort out. https://www.doh.wa.gov/portals/1/documents/pubs/320-090_elecpuls_fs.pdf https://www.doh.wa.gov/portals/1/documents/pubs/320-090_elec...
- CodeGlitch 5y ago
  From your link: > There are two basic ways to protect or harden items against EMP effects. The first method is metallic shielding. Shields are made of a continuous piece of metal such as steel or copper. A metal enclosure generally does not fully shield the interior because of the small holes that are likely to exist. Therefore, this type of shielding often contains additional elements to create the barrier. Commonly, only a fraction of a millimeter of a metal is needed to supply adequate protection. This shield must completely surround the item to be hardened. > The second method, tailored hardening, is a more cost-effective way of hardening. In this method, only the most vulnerable elements and circuits are redesigned to be more rugged. The more rugged elements will be able to withstand much higher currents. This method has shown unpredictable failures in testing, though it is thought it may be useful to make existing systems less vulnerable. Seems reasonable to provide EMP shielding to a drone?
  - sklargh 5y ago
    Since we are capable of EMP-hardening very large airframes (see Nightwatch, Looking Glass, SAM) I assume shielding crucial systems on a drone would be an initial design challenge but not impossible.
- oneoff786 5y ago
  Do we actually have non nuclear emp weapons
  - belter 5y ago
    Also from the reference above: "For weapons purposes EMP producing sources other than nuclear detonations have been successfully developed. Several nations, with United States at the forefront, are reported to have developed non-nuclear bombs capable of generating EMPs. Electromagnetic bombs (E-bombs) are specialized, non-nuclear tools designed to destroy information systems. These devices are primarily intended for battlefield application..."
  - triactual 5y ago
    Yes, but they really only work on civilian targets because military hardware is EMP hardened and has been since a few weeks after the advent of nuclear weapons. The scary stuff is electronic warfare where you instruments seem to be working but they're not.
jancsika 5y ago
> This bodes very poorly for any nation that falls behind in the AI race. As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms. Curious how you think such an arms race avoids complete annihilation of both sides.
- systemvoltage 5y ago
  There is a great course by Prof. William Spaniel about Game Theory in the context of Nuclear Proliferation. Highly recommend it for anyone who wants to know about the nuts and bolts of how nuclear policy is shaped: https://www.youtube.com/playlist?list=PLKI1h_nAkaQqivWh2_XEtRacEW7ttnq1W https://www.youtube.com/playlist?list=PLKI1h_nAkaQqivWh2_XEt... Basically, you have to choose between MAD or asymmetrical destruction. With exponentially improving AI, the situation is more like an inverted pendulum. Any small perturbation (say resources or breakthroughs in AI) would lead to catastrophic asymmetrical advantage for one side.
onion2k 5y ago
This bodes very poorly for any nation that falls behind in the AI race. Not really. The chance of any sort of aerial dogfighting happening in a war are next to zero. The cost of planes, and pilots, is so high that almost all the effort spent these days is on killing your enemy with over-the-horizon weapons fired from hundreds of miles away. AI-powered dogfighting in aircraft is military marketing. A nation could develop world-class AI planes and they'd be absolutely useless against a hypersonic missile fired from mainland China or a Russian submarine, nor would they be any use against a guerilla army in Afghanistan or Iraq that doesn't have the capability to fight in the air or that has 30 year old planes that can be shot down with 50 year old tech. War has changed; making better versions of old weapons is a waste of time.
- gloriana 5y ago
  This is true. But the barrier to using over-the-horizon weapons like icbm / hypersonic is much greater than say drone attacks or fighter jet bombs. Have over-the-horizon weapons ever been used? No. But drone attacks and fighter jet activities are daily occurrences because they can be used with little repercussions.
  - regularfry 5y ago
    Over-the-horizon weapons absolutely have been used. Frequently, repeatedly, and (allegedly) effectively. Tomahawks count, unless they're retrospectively reclassified as drones.
- fnord77 5y ago
  > A nation could develop world-class AI planes... such a nation would probably already have nuclear weapons. Any nation capable of hypersonic skip glide weapons also probably has nukes. So this hypothetical attack wouldn't be happening. At least w/ non-nuclear weapons.
- atlasunshrugged 5y ago
  Can you extrapolate on this? In my understanding in recent conflicts (e.g. Armenia and Azerbaijan) and currently in Ethiopia, drones (okay, we're not talking about fighter jets here but still, autonomous or near autonomous aerial systems) have played critical roles in eliminating air defenses, heavy weaponry, etc. and in future conflicts now that parties on all sides know how important the use of such drones are, will be deploying their own drones as countermeasures.
sAbakumoff 5y ago
>> As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms. Wait, what? Are you saying that if a scientist refuses to contribute to AI development that will be used to kill human beings then he plays into the hands of nation's adversaries ?
- CalRobert 5y ago
  Not to put words in their mouth but isn't that pretty clear? If you don't develop advanced weaponry, you are going to be killed by people who _do_ develop advanced weaponry. Unless you manage to stop the advanced weaponry from being developed at all, which is admirable, but incredibly difficult, and with huge consequences for being wrong.
  - atlasunshrugged 5y ago
    Or unless you have a significant enough deterrent and the credibility to use it in basically any sort of conflict (N. Korea may fit into this category)
  - pasabagi 5y ago
    That depends on your model of international relations. Most states have recognized that international relationships are far from zero sum, and that wars between peer competitors are almost never worth it. Indeed, the success stories since WW2 have been those countries that have spent very little on weapons, and engaged in few wars (with the US being the exception). Even if you take WW2 and WW1 into account, the biggest winner of those conflicts (the US) spent the least on armaments prior, and had probably the least aggressive foreign policy of any major power.
    - boppo1 5y ago
      > have been those countries that have spent very little on weapons, and engaged in few wars (with the US being the exception). Perhaps, because the US is an exception.
scandox 5y ago
Arthur C Clarke has a good story on the dangers of warfare at the bleeding edge https://en.m.wikipedia.org/wiki/Superiority_(short_story) https://en.m.wikipedia.org/wiki/Superiority_(short_story)
beerandt 5y ago
Just as advances in war tech in the past have led to the more "advanced" war machine "fighting the last war," these advances only serve to change the behaviors, rules, and context of any future engagement. Keeping in mind that war is ultimately a "battle" of resources and expenditures, no one in a position of great disadvantage will invest in that particular effort, but seek a way to circumvent it. Whereas the "winner" has already sunk that cost. (Even without AI, how common are dogfights or air-to-air confrontation in the last 50 years?) There are ways for the more advanced participant to defeat this game, but so far it has required violating human norms that are held sacrosanct even in war. AI's advantage might end up being the ability to navigate this evolving game theory, but only if it's able to properly value what's acceptable to human nature and what isn't. A "kill all humans as a solution to end spam" doesn't work here (unless the AI has taken ultimate charge, which becomes an entirely different issue). On the other hand, certain near-peer potential adversaries have signaled a willingness to violate whatever norms need be violated to secure victory.
minimuffins 5y ago
> As well as any nation who's best and brightest minds refuse to contribute Sickening pro robo war jingoism!
torginus 5y ago
I remember I read a fighter pilot's analysis on the events: - Bullet drop and travel time were not modelled, the guns were literally lasers - Most of the kills the AI did were high aspect shots, meaning the aircraft flew perpendicularly to each other. In real life, these shots are almost impossible to hit, due to the aforementioned ignored factors - A human pilot in a simulator doesn't have the situational awareness of flying in a physical plane
- KineticLensman 5y ago
  Yes. >> The pilot, dressed in an olive-green flight suit, sat in a high-backed gaming chair, his face obscured by a virtual-reality headset. So nothing like the real cockpit they would have trained in, with low-resolution, high-latency visuals. This alone could have removed a lot of the expert advantage of a real pilot in a real plane.
  - VectorLock 5y ago
    If they didn't model bullet drop, did they model effects of high-G on the pilots?
imtringued 5y ago
I could say the same thing about lots of obvious economic policies... The world doesn't work like that.
oneoff786 5y ago
That doesn’t make much sense. You’re implying that because this AI beat another AI by a lot that all skill gaps between AI must be significant. What if the second and third place versions just weren’t as good and over a relatively short term it all converges?
XorNot 5y ago
This is an extremified version of the general progress of technology in warfighting though: while we have the stories of automotive factories being converted to produce tanks during World War 2, that was something that was only possible in the environment of the time. More then ever, you go to war with the military you have, the outcome is going to be decided very quickly because the rate of equipment attrition is enormous: military victory will be decided in weeks, not years (you can of course still lose the occupation, but that's a different matter).
stubish 5y ago
An AI system will consistently beat an opposing AI system if it has been trained to beat the opposing AI system, as it will only engage if it can predict a favorable outcome. Fully autonomous systems are going to have to be very, very secret because one leak will destroy their utility. This alone might be a good reason to keep them as pilot assist systems rather than pilot replacement.
Gortal278 5y ago
> As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms. I am worried about this somewhat. I know a few too many great engineers reluctant to work on defense work because they are concerned about what their friends will think. It's not particularly tasteful in progressive circles.
- the_only_law 5y ago
  Those fucking defense companies can start paying more and maybe I’ll consider building weapons to be sold to the Saudis.
dehrmann 5y ago
> This bodes very poorly for any nation that falls behind in the AI race. How much do flight characteristics matter? A better AI might still lose in a worse plane.

belter 5y ago

Nothing a proper EMP weapon will not sort out. https://www.doh.wa.gov/portals/1/documents/pubs/320-090_elecpuls_fs.pdf https://www.doh.wa.gov/portals/1/documents/pubs/320-090_elec...

CodeGlitch 5y ago
From your link: > There are two basic ways to protect or harden items against EMP effects. The first method is metallic shielding. Shields are made of a continuous piece of metal such as steel or copper. A metal enclosure generally does not fully shield the interior because of the small holes that are likely to exist. Therefore, this type of shielding often contains additional elements to create the barrier. Commonly, only a fraction of a millimeter of a metal is needed to supply adequate protection. This shield must completely surround the item to be hardened. > The second method, tailored hardening, is a more cost-effective way of hardening. In this method, only the most vulnerable elements and circuits are redesigned to be more rugged. The more rugged elements will be able to withstand much higher currents. This method has shown unpredictable failures in testing, though it is thought it may be useful to make existing systems less vulnerable. Seems reasonable to provide EMP shielding to a drone?
- sklargh 5y ago
  Since we are capable of EMP-hardening very large airframes (see Nightwatch, Looking Glass, SAM) I assume shielding crucial systems on a drone would be an initial design challenge but not impossible.
oneoff786 5y ago
Do we actually have non nuclear emp weapons
- belter 5y ago
  Also from the reference above: "For weapons purposes EMP producing sources other than nuclear detonations have been successfully developed. Several nations, with United States at the forefront, are reported to have developed non-nuclear bombs capable of generating EMPs. Electromagnetic bombs (E-bombs) are specialized, non-nuclear tools designed to destroy information systems. These devices are primarily intended for battlefield application..."
- triactual 5y ago
  Yes, but they really only work on civilian targets because military hardware is EMP hardened and has been since a few weeks after the advent of nuclear weapons. The scary stuff is electronic warfare where you instruments seem to be working but they're not.

jancsika 5y ago

> This bodes very poorly for any nation that falls behind in the AI race. As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms. Curious how you think such an arms race avoids complete annihilation of both sides.

systemvoltage 5y ago
There is a great course by Prof. William Spaniel about Game Theory in the context of Nuclear Proliferation. Highly recommend it for anyone who wants to know about the nuts and bolts of how nuclear policy is shaped: https://www.youtube.com/playlist?list=PLKI1h_nAkaQqivWh2_XEtRacEW7ttnq1W https://www.youtube.com/playlist?list=PLKI1h_nAkaQqivWh2_XEt... Basically, you have to choose between MAD or asymmetrical destruction. With exponentially improving AI, the situation is more like an inverted pendulum. Any small perturbation (say resources or breakthroughs in AI) would lead to catastrophic asymmetrical advantage for one side.

onion2k 5y ago

This bodes very poorly for any nation that falls behind in the AI race. Not really. The chance of any sort of aerial dogfighting happening in a war are next to zero. The cost of planes, and pilots, is so high that almost all the effort spent these days is on killing your enemy with over-the-horizon weapons fired from hundreds of miles away. AI-powered dogfighting in aircraft is military marketing. A nation could develop world-class AI planes and they'd be absolutely useless against a hypersonic missile fired from mainland China or a Russian submarine, nor would they be any use against a guerilla army in Afghanistan or Iraq that doesn't have the capability to fight in the air or that has 30 year old planes that can be shot down with 50 year old tech. War has changed; making better versions of old weapons is a waste of time.

gloriana 5y ago
This is true. But the barrier to using over-the-horizon weapons like icbm / hypersonic is much greater than say drone attacks or fighter jet bombs. Have over-the-horizon weapons ever been used? No. But drone attacks and fighter jet activities are daily occurrences because they can be used with little repercussions.
- regularfry 5y ago
  Over-the-horizon weapons absolutely have been used. Frequently, repeatedly, and (allegedly) effectively. Tomahawks count, unless they're retrospectively reclassified as drones.
fnord77 5y ago
> A nation could develop world-class AI planes... such a nation would probably already have nuclear weapons. Any nation capable of hypersonic skip glide weapons also probably has nukes. So this hypothetical attack wouldn't be happening. At least w/ non-nuclear weapons.
atlasunshrugged 5y ago
Can you extrapolate on this? In my understanding in recent conflicts (e.g. Armenia and Azerbaijan) and currently in Ethiopia, drones (okay, we're not talking about fighter jets here but still, autonomous or near autonomous aerial systems) have played critical roles in eliminating air defenses, heavy weaponry, etc. and in future conflicts now that parties on all sides know how important the use of such drones are, will be deploying their own drones as countermeasures.

sAbakumoff 5y ago

>> As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms. Wait, what? Are you saying that if a scientist refuses to contribute to AI development that will be used to kill human beings then he plays into the hands of nation's adversaries ?

CalRobert 5y ago
Not to put words in their mouth but isn't that pretty clear? If you don't develop advanced weaponry, you are going to be killed by people who _do_ develop advanced weaponry. Unless you manage to stop the advanced weaponry from being developed at all, which is admirable, but incredibly difficult, and with huge consequences for being wrong.
- atlasunshrugged 5y ago
  Or unless you have a significant enough deterrent and the credibility to use it in basically any sort of conflict (N. Korea may fit into this category)
- pasabagi 5y ago
  That depends on your model of international relations. Most states have recognized that international relationships are far from zero sum, and that wars between peer competitors are almost never worth it. Indeed, the success stories since WW2 have been those countries that have spent very little on weapons, and engaged in few wars (with the US being the exception). Even if you take WW2 and WW1 into account, the biggest winner of those conflicts (the US) spent the least on armaments prior, and had probably the least aggressive foreign policy of any major power.
  - boppo1 5y ago
    > have been those countries that have spent very little on weapons, and engaged in few wars (with the US being the exception). Perhaps, because the US is an exception.

scandox 5y ago

Arthur C Clarke has a good story on the dangers of warfare at the bleeding edge https://en.m.wikipedia.org/wiki/Superiority_(short_story) https://en.m.wikipedia.org/wiki/Superiority_(short_story)

beerandt 5y ago

Just as advances in war tech in the past have led to the more "advanced" war machine "fighting the last war," these advances only serve to change the behaviors, rules, and context of any future engagement. Keeping in mind that war is ultimately a "battle" of resources and expenditures, no one in a position of great disadvantage will invest in that particular effort, but seek a way to circumvent it. Whereas the "winner" has already sunk that cost. (Even without AI, how common are dogfights or air-to-air confrontation in the last 50 years?) There are ways for the more advanced participant to defeat this game, but so far it has required violating human norms that are held sacrosanct even in war. AI's advantage might end up being the ability to navigate this evolving game theory, but only if it's able to properly value what's acceptable to human nature and what isn't. A "kill all humans as a solution to end spam" doesn't work here (unless the AI has taken ultimate charge, which becomes an entirely different issue). On the other hand, certain near-peer potential adversaries have signaled a willingness to violate whatever norms need be violated to secure victory.

minimuffins 5y ago

> As well as any nation who's best and brightest minds refuse to contribute Sickening pro robo war jingoism!

torginus 5y ago

I remember I read a fighter pilot's analysis on the events: - Bullet drop and travel time were not modelled, the guns were literally lasers - Most of the kills the AI did were high aspect shots, meaning the aircraft flew perpendicularly to each other. In real life, these shots are almost impossible to hit, due to the aforementioned ignored factors - A human pilot in a simulator doesn't have the situational awareness of flying in a physical plane

KineticLensman 5y ago
Yes. >> The pilot, dressed in an olive-green flight suit, sat in a high-backed gaming chair, his face obscured by a virtual-reality headset. So nothing like the real cockpit they would have trained in, with low-resolution, high-latency visuals. This alone could have removed a lot of the expert advantage of a real pilot in a real plane.
- VectorLock 5y ago
  If they didn't model bullet drop, did they model effects of high-G on the pilots?

imtringued 5y ago

I could say the same thing about lots of obvious economic policies... The world doesn't work like that.

oneoff786 5y ago

That doesn’t make much sense. You’re implying that because this AI beat another AI by a lot that all skill gaps between AI must be significant. What if the second and third place versions just weren’t as good and over a relatively short term it all converges?

XorNot 5y ago

This is an extremified version of the general progress of technology in warfighting though: while we have the stories of automotive factories being converted to produce tanks during World War 2, that was something that was only possible in the environment of the time. More then ever, you go to war with the military you have, the outcome is going to be decided very quickly because the rate of equipment attrition is enormous: military victory will be decided in weeks, not years (you can of course still lose the occupation, but that's a different matter).

stubish 5y ago

An AI system will consistently beat an opposing AI system if it has been trained to beat the opposing AI system, as it will only engage if it can predict a favorable outcome. Fully autonomous systems are going to have to be very, very secret because one leak will destroy their utility. This alone might be a good reason to keep them as pilot assist systems rather than pilot replacement.

Gortal278 5y ago

> As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms. I am worried about this somewhat. I know a few too many great engineers reluctant to work on defense work because they are concerned about what their friends will think. It's not particularly tasteful in progressive circles.

the_only_law 5y ago
Those fucking defense companies can start paying more and maybe I’ll consider building weapons to be sold to the Saudis.

dehrmann 5y ago

> This bodes very poorly for any nation that falls behind in the AI race. How much do flight characteristics matter? A better AI might still lose in a worse plane.