A.I. in warfare: new critical aspect or shallow hotword

[JasonJ]

More and more I hear about an importance to expand on military R&D in artificial intelligence. Will it really be that critical? Or is it a fancy concept that won't really evolve to be revolutionary in practical affect?

Edited September 17, 2020 by JasonJ

[Ssnake]

In the beginning it won't look like much of a big deal. But it accumulates, and compounds faster than the mechanical and electronic/sensor evolution. So it definitely has the potential to become a big thing. The challenge is data handling.

All the "AI" happening in your smartphone today (predictive typing, voice-controlled assistants) depends on a datacenter far away, and the constrant transmission of local data for remote analysis. On a battlefield with EM control, that's not a viable concept. But we might be recording a lot, then feeding it for analysis, and receive a configured neuronal network application for local use. That might work. And then of course, whenever you emit like crazy in a benign EW environment you can of course afford real-time support.

Of course, setting up secure, mobile datacenters in a war zone is an entirely different challenge. 😜 

[JasonJ]

That makes sense, EW was another getting more attention. Always regarded it for messing up radar or communication airwaves. EW would have an even more important role if AI becomes significant.

 

I wonder if the A.I mentioned as necessary R&D in various talks and such is limited to the smartphone stuff you mentioned like predictive typing. Or if they are thinking about literal decision making, as in two-shoot-or-not-to-shoot kind of thing. Like maybe for sentry turrets.

 

Or maybe like the AI god giving recommendation to ship captain to make the quick decisions for him when hyper speed missiles are flying around.

 

[RETAC21]

16 minutes ago, Ssnake said:

In the beginning it won't look like much of a big deal. But it accumulates, and compounds faster than the mechanical and electronic/sensor evolution. So it definitely has the potential to become a big thing. The challenge is data handling.

All the "AI" happening in your smartphone today (predictive typing, voice-controlled assistants) depends on a datacenter far away, and the constrant transmission of local data for remote analysis. On a battlefield with EM control, that's not a viable concept. But we might be recording a lot, then feeding it for analysis, and receive a configured neuronal network application for local use. That might work. And then of course, whenever you emit like crazy in a benign EW environment you can of course afford real-time support.

Of course, setting up secure, mobile datacenters in a war zone is an entirely different challenge. 😜 

If you don't need stealth, then frequency hopping radios will enable that remote control. Recently Microsoft "rescued" a submerged data center, which would be pretty secure against most oponents

https://news.microsoft.com/innovation-stories/project-natick-underwater-datacenter/

[Ssnake]

Just the other day I saw some article showing (or possibly illustrating with "artist's impression") thermal sights with human heat signatures and assigning threat likelihoods to them (as in "combatant or not"). I have no doubt that you can train a neural network like that and integrate it to some weapon sight. I do have my doubts whether this is reliable and useful.

But irrespective of whether it works or not - and the US Military has introduced algorithmic kill decisions a good while ago with the CIA & NSA assigning "terror scores" to every mobile phone in the world; that the AI's verdict who needs some killin' must still be signed off by the president and then executed by some UAV operator team doesn't change the fact that the target selection is largely driven by some algorithm about which we know very little - irrespective of all that, some will try. So yeah, we'll also see inept AI making kill decisions.

Yay.

[Stuart Galbraith]

1 hour ago, RETAC21 said:

If you don't need stealth, then frequency hopping radios will enable that remote control. Recently Microsoft "rescued" a submerged data center, which would be pretty secure against most oponents

https://news.microsoft.com/innovation-stories/project-natick-underwater-datacenter/

I think those are a seriously bad idea, bearing in mind how much effort the Russians have put into minisubs and means to deploy them. A future Ivy Bells seems inevitable at this point.

 

As for AI, its difficult to conceive of the uses it will be put to. Although logically speaking, the amount of sources and information must be picked up by drones, F35's, Awacs etc, on the modern battlefield must be difficult to reconcile. OK, you have people like Bellingcat you can spend months searching out some guy who worked with some unit and put facebook postsup. But if you can imagine a machine that could do all that in minutes, the intelligence take could be massive.

As long as we dont let them do our thinking for us of course, which is always a danger when people want quick, easy answers to complex questions.

[shep854]

In a first-tier, peer conflict, how long will all those emitters be permitted to function?  It seems they are just begging to have something loud and violent home on them.

[Stuart Galbraith]

Well if you are going to be using AI, they are going to be pulling in data from multiple sources, even things like facebook and twitter. If you look at the Bellingcat threads on Ukraine, you will see the Russians are horribly insecure in that kind of thing, even if they are trying to clamp down on it. Its probably impossible to clamp down on all of that, particularly stuff thats several years old that might still have relevance (military units dont move barracks very often)

I would guess that if you have an AI system capableof categorising battlefield threats, you should be able to have it point out what its most useful sources are, and not target them.  Its not that much different from WW2, where we daily had to make assessments on whether to use Ultra information, and risk losing the source, or just go ahead and pillage it. As the war drew to a close, it got easier and easier to make assessments like that. In the end they were seemingly sending out stuff via the BBC to shake the Germans up about spies on their staff.

[JasonJ]

44 minutes ago, shep854 said:

In a first-tier, peer conflict, how long will all those emitters be permitted to function?  It seems they are just begging to have something loud and violent home on them.

On mentioning of that, ISTR that it was said that the USN would need to practice using old fashion navigation means to break dependence on GPS, etc, because of the danger of EW environment in some of the talks by John M. Richardson when he was CHO.

[JasonJ]

45 minutes ago, Stuart Galbraith said:

Well if you are going to be using AI, they are going to be pulling in data from multiple sources, even things like facebook and twitter. If you look at the Bellingcat threads on Ukraine, you will see the Russians are horribly insecure in that kind of thing, even if they are trying to clamp down on it. Its probably impossible to clamp down on all of that, particularly stuff thats several years old that might still have relevance (military units dont move barracks very often)

I would guess that if you have an AI system capableof categorising battlefield threats, you should be able to have it point out what its most useful sources are, and not target them.  Its not that much different from WW2, where we daily had to make assessments on whether to use Ultra information, and risk losing the source, or just go ahead and pillage it. As the war drew to a close, it got easier and easier to make assessments like that. In the end they were seemingly sending out stuff via the BBC to shake the Germans up about spies on their staff.

The Type 10 tank apparently has such a threat identification database and feeds that info live to the crew about detected targets and provides recommendation on ammo type. So I guess that is AI stuff already.

[shep854]

3 minutes ago, JasonJ said:

On mentioning of that, ISTR that it was said that the USN would need to practice using old fashion navigation means to break dependence on GPS, etc, because of the danger of EW environment in some of the talks by John M. Richardson when he was CHO.

I believe the USN still practices EMCON, which would also require shutting down GPS.

[glenn239]

4 hours ago, JasonJ said:

More and more I hear about an importance to expand on military R&D in artificial intelligence. Will it really be that critical? Or is it a fancy concept that won't really evolve to be revolutionary in practical affect?

An article the other day said that the USAF's best fighter pilots lose in simulated dogfights against the AI now.  Why would a 3rd world country pay millions of dollars to train a mediocre pilot when they could pay a fraction of that and have the best pilot in the world?

[Josh]

16 minutes ago, shep854 said:

I believe the USN still practices EMCON, which would also require shutting down GPS.

GPS is a completely passive system. The emitters are the satellites, which is why it is easily jammed - its low strength signal coming from 10,000 miles away. But receiving and processing that signal is totally passive.

[JasonJ]

1 minute ago, glenn239 said:

An article the other day said that the USAF's best fighter pilots lose in simulated dogfights against the AI now.  Why would a 3rd world country pay millions of dollars to train a mediocre pilot when they could pay a fraction of that and have the best pilot in the world?

So in a dogfight, the AI could control a fighter aircraft to the most advantageous flight maneuvers. Sounds like a computer chess pilot.

[shep854]

4 minutes ago, Josh said:

GPS is a completely passive system. The emitters are the satellites, which is why it is easily jammed - its low strength signal coming from 10,000 miles away. But receiving and processing that signal is totally passive.

Hm.  I thought I had read that the ground unit had to 'shake hands' with satellites to obtain position data.

[Josh]

Where AI will (and already is) be used the most is in intelligence gathering and target detection. At the highest level this will amount to very high end pattern analysis across domains - sensor information, social media, the nightly news. At the pointing end this will involve identifying and prioritizing targets in sensor clutter. We are already seeing aircraft systems and individual missiles that have their own libraries of target types (ship classes in AShMs for instance). AI will extend this to recognizing targets not previously loaded by their common characteristics and making choices about weapon engagements - see the Golden Horde project. AI will also be absolutely necessary for electronic warfare, since modern AESA radars can operate across frequencies with an essentially random wave form. Detecting a radar not previously encountered or employing a non traditional pattern requires AI, as does any hope of attempting to jam it. And AI is necessary for anything like the necessary reaction time - by the time a human looks at a screen to see he's being illuminated by a radar, a modern system already as established a weapon delivery grade track.

[Josh]

11 minutes ago, glenn239 said:

An article the other day said that the USAF's best fighter pilots lose in simulated dogfights against the AI now.  Why would a 3rd world country pay millions of dollars to train a mediocre pilot when they could pay a fraction of that and have the best pilot in the world?

A dog fighting AI amounts to more of a parlor trick - it was a gun battle that the AI won, and in the age 180 degree, 60G IIR AAMs I think all WVR fighting between peers is going to be a single turn event. But to your point, yes it does seem like AI enabled aircraft could allow a previously non peer country to challenge a more capable opponent. However the best AI and platforms will likely be limited to top tier militaries, and those they chose to export to. So it likely will still be more of a fight between peers, IMO.

[glenn239]

1 hour ago, JasonJ said:

So in a dogfight, the AI could control a fighter aircraft to the most advantageous flight maneuvers. Sounds like a computer chess pilot.

They had a video playing of an actual pilot-AI dogfight.  It was in simulator, not real, but the AI was in the action-reaction cycle to a degree that the pilot could not match. 

[Ssnake]

5 hours ago, Stuart Galbraith said:

As long as we dont let them do our thinking for us of course, which is always a danger when people want quick, easy answers to complex questions.

But that's precisely what's going to happen - arguably it already is happening WRT algorithmic drone strike kill "recommendations" - because the analytical part of our brain, System II, is lazy (especially if you're tired, with low blood sugar level, ...), and because lawyers second-guessing all your decisions won't be as much on your tail if you follow the computer's recommendation, unless it's obviously, egregiously absurd (and/or your case happens to acquire a very high publicity profile).

[glenn239]

4 hours ago, Josh said:

Where AI will (and already is) be used the most is in intelligence gathering and target detection. At the highest level this will amount to very high end pattern analysis across domains - sensor information, social media, the nightly news. At the pointing end this will involve identifying and prioritizing targets in sensor clutter. We are already seeing aircraft systems and individual missiles that have their own libraries of target types (ship classes in AShMs for instance). AI will extend this to recognizing targets not previously loaded by their common characteristics and making choices about weapon engagements - see the Golden Horde project. AI will also be absolutely necessary for electronic warfare, since modern AESA radars can operate across frequencies with an essentially random wave form. Detecting a radar not previously encountered or employing a non traditional pattern requires AI, as does any hope of attempting to jam it. And AI is necessary for anything like the necessary reaction time - by the time a human looks at a screen to see he's being illuminated by a radar, a modern system already as established a weapon delivery grade track.

Taken to its ultimate level, AI systems will make war increasingly more symbolic, as it will be machine vs. machine with the humans deciding the treaty based on the outcome of how the robots perform.  In terms of air defense, more along the lines of thousands, then tens of thousands, of autonomous drones so that an aircraft or missile can't get to a target without going through an area where a drone can move into its path.

In human terms, wars will be increasingly targeted at the decision makers and provocateurs that start them, these on both sides identified and systematically hunted to destruction by enemy AI that are looking for them, and only them.

 

[Josh]

I suspect drones won't be programmed to be nearly that selective but all players. With a big enough swarm and a cheap enough drone, you can easily exchange them one for one with enemy soldiers. Picture something like PERDIX being blanketed across an infantry battalion, only smart enough to actually ID humans as targets and carrying a few grams of HE. Hopefully such usage will at least make some attempt to separate soldiers from civilians, but it seems unlikely a government like China's or Russia's would really be that interested in the difference.

[Ssnake]

Exactly. I'm not even sure how much our own governments respect that difference, and be it just out of apathy.

[Ssnake]

8 hours ago, glenn239 said:

Taken to its ultimate level, AI systems will make war increasingly more symbolic, ... In human terms, wars will be increasingly targeted at the decision makers and provocateurs that start them, these on both sides identified and systematically hunted to destruction by enemy AI that are looking for them, and only them.

I think that's wishful thinking.

Since 2001 we have hunted every mid level Taliban leader with a cellphone. 20 years later we have killed three to four leader generations, but guess who's still there and winning the war while we are pulling out. The "node targeting" doctrine has been tried since the 1940s, exemplified with the air attack on the FAG Kugelfischer ball bearing factory in Schweinfurt in the attempt to grind the Nazi war machine to a halt, to the DEA's strategy to target drug cartel leaders, air strikes on Saddam's communication and air defense centers, etc.

The only predictable result of these strategies was that the targeting list was expanded. I know of no case where enough critical nodes were hit to bring the promised swift end to a campaign. And in any case, with a sufficiently desparate or ruthless leader, even after he lost all his robots but still has soldiers, he'll keep feeding them to the grinder. When war is a fight for survival, nobody gives up just because he ran out of robots.

[DB]

I can imagine that smartphone level processing to assign threat levels to thermal images of people will result in a lot of dead photojournalists.

Certainly, identifying people is now trivial - it must be more than a decade since someone put together a laptop, camera and a two-axis controlled paintball "!sentry gun" in his back garden, which shot things with arms and legs.

This may be the the earliest version, but it's working against a white background and probably not pattern matching to a person in this version. 

 

[Stuart Galbraith]

10 hours ago, Ssnake said:

I think that's wishful thinking.

Since 2001 we have hunted every mid level Taliban leader with a cellphone. 20 years later we have killed three to four leader generations, but guess who's still there and winning the war while we are pulling out. The "node targeting" doctrine has been tried since the 1940s, exemplified with the air attack on the FAG Kugelfischer ball bearing factory in Schweinfurt in the attempt to grind the Nazi war machine to a halt, to the DEA's strategy to target drug cartel leaders, air strikes on Saddam's communication and air defense centers, etc.

The only predictable result of these strategies was that the targeting list was expanded. I know of no case where enough critical nodes were hit to bring the promised swift end to a campaign. And in any case, with a sufficiently desparate or ruthless leader, even after he lost all his robots but still has soldiers, he'll keep feeding them to the grinder. When war is a fight for survival, nobody gives up just because he ran out of robots.

I always thought vietnam was quite instructive, where the US was using 20th Century means to defeat a 2 or 3 hundred year old problem. You cant develop a smart bomb to target an idea in someones head, something we seem to have rediscovered in Afghanistan.

AI wont help with that. It will perhaps identify messengers and agitators easier. Although when you see so many wannabe terrorists in Europe who are apparently essentially self radicalizing, then ive not much hope its going to make much difference.

As for your last, Ive been reading up recently on the Iran/Iraq war. Iran had essentially won the war by 1982, having pushed Iraq back to their own borders. It was politics, pure and simple, that demanded they push on for complete victory, and expended thousands of Iranians, Including some Child Soldiers, in an effort to export the revolution to Iraq. What would AI do for what are essentially human wave tactics, driven by religious fervor and misguided politics? Nothing. The only thing AI will help with is large structred militaries, because they have an organizational layout its possible to identify. I get the impression with Iran, systems that are supposed to be basically heirarchical, when it comes down to it just arent.