Global military buildup

[Barrister]

" The cold reality is that if you have a weapons system that prevents 1000 kills of your own by the enemy at the cost of increasing friendly fire kills by 100"

He is saying that if you can PREVENT 1000 friends getting kills at the cost of 100 friendly fire casualties, you should do that.

OK, fair enough.

Intellectually, sure.  But in the real world you never see those 1000 deaths prevented, you only see the 100 friendly fire casualties.

My understanding is that while friendly fire incidents will inevitably happen, modern militaries work very very hard to minimize them.

[Berkut]

Friendly fire is a fact of life.  If you deploy artillery or bombers, you're going to kill your own people sometimes, and yet no one is considering to not use them for those reasons. 

I don't think the military shares the risk aversion expressed here.  The cold reality is that if you have a weapons system that prevents 1000 kills of your own by the enemy at the cost of increasing friendly fire kills by 100, you're going to use it as much as you can.  Unreasonable risk aversion is not a luxury you can afford in an endeavor where people die a lot.

While I understand your point, I very much doubt that western militaries would find a 10:1 ratio of extra enemy casualties to friendly fire casualties acceptable.

" The cold reality is that if you have a weapons system that prevents 1000 kills of your own by the enemy at the cost of increasing friendly fire kills by 100"

He is saying that if you can PREVENT 1000 friends getting kills at the cost of 100 friendly fire casualties, you should do that.

Yes, but that reasoning is fallacious if it is possible to bring down the friendly kills to a lower number.  Just as the military now attempts to do.  No one plans an operation which accepts that a 10% friendly fire kill rate is possible.  Quite the contrary, friendly fire events are avoided as much as possible.

Thanks, I am sure nobody realized that friendly fire was something that everyone tries to avoid and minimize.

[Berkut]

" The cold reality is that if you have a weapons system that prevents 1000 kills of your own by the enemy at the cost of increasing friendly fire kills by 100"

He is saying that if you can PREVENT 1000 friends getting kills at the cost of 100 friendly fire casualties, you should do that.

OK, fair enough.

Intellectually, sure.  But in the real world you never see those 1000 deaths prevented, you only see the 100 friendly fire casualties.

My understanding is that while friendly fire incidents will inevitably happen, modern militaries work very very hard to minimize them.

Of course. But they don't work THAT hard. I mean, they could reduce them to zero obviously, by just not shooting, ever.

And I would dispute the idea that in the "real world" you don't see the 1000 deaths prevented. I in fact think in the "real world" they very much do - in the media world? The civilian world? Maybe not so much.

Modern militaries certainly do work very hard to minimize them. Vast effort is spent doing so. But at the end of the day, war is about violence, and DG was exactly right. Once you decide to engage in war, there are going to be mistakes and friendly fire. Over-emphasizing the avoidance of that would be a mistake if it meant that more friendly die in the long run because you are worried about preventing friendly fire in the short run.

This was pretty well understood when we engaged in serious wars. Danger close artillery is, well, danger close. Sometimes "danger" means "Ooops, our guys are dead".

[Barrister]

And I would dispute the idea that in the "real world" you don't see the 1000 deaths prevented. I in fact think in the "real world" they very much do - in the media world? The civilian world? Maybe not so much.

Modern militaries certainly do work very hard to minimize them. Vast effort is spent doing so. But at the end of the day, war is about violence, and DG was exactly right. Once you decide to engage in war, there are going to be mistakes and friendly fire. Over-emphasizing the avoidance of that would be a mistake if it meant that more friendly die in the long run because you are worried about preventing friendly fire in the short run.

This was pretty well understood when we engaged in serious wars. Danger close artillery is, well, danger close. Sometimes "danger" means "Ooops, our guys are dead".

What I mean by you don't "see" those saved lives is this:

Covid-19 vaccines have saved hundreds of thousands of human lives.  But nobody knows who those "saved lives" are.  Nobody gets a text message saying 'you would have died if you didn't get your Covid vaccine'.  Same things for those saved casualties.  Intellectually you might know that those saved lives exist, but you can never put a face to them.

And it's not just about fighting "serious wars".  Friendly fire incidents are terrible for morale, and have a real impact on a unit's ability to fight effectively.

Which goes back to my point that I don't think a military would accept the "we'll save 1000 lives in exchange for 100 dying from friendly fire", even if you could quantify it that way.

[Berkut]

And I would dispute the idea that in the "real world" you don't see the 1000 deaths prevented. I in fact think in the "real world" they very much do - in the media world? The civilian world? Maybe not so much.

Modern militaries certainly do work very hard to minimize them. Vast effort is spent doing so. But at the end of the day, war is about violence, and DG was exactly right. Once you decide to engage in war, there are going to be mistakes and friendly fire. Over-emphasizing the avoidance of that would be a mistake if it meant that more friendly die in the long run because you are worried about preventing friendly fire in the short run.

This was pretty well understood when we engaged in serious wars. Danger close artillery is, well, danger close. Sometimes "danger" means "Ooops, our guys are dead".

What I mean by you don't "see" those saved lives is this:

Covid-19 vaccines have saved hundreds of thousands of human lives.  But nobody knows who those "saved lives" are.  Nobody gets a text message saying 'you would have died if you didn't get your Covid vaccine'.  Same things for those saved casualties.  Intellectually you might know that those saved lives exist, but you can never put a face to them.

And it's not just about fighting "serious wars".  Friendly fire incidents are terrible for morale, and have a real impact on a unit's ability to fight effectively.

Which goes back to my point that I don't think a military would accept the "we'll save 1000 lives in exchange for 100 dying from friendly fire", even if you could quantify it that way.

I think every military everywhere accepts exactly that, because to not accept that (or something like that) would mean that you simply do not engage.

And my point about serious war was just that in those the acceptance is much more, well, accepted. In a more transparent manner. There are multiple examples where the US Army ordered things done that they 100% knew would result in significant friendly fire casualties, but did so anyway. 

In a "less serious" war, they do the same thing, but just not in as transparent a manner, because there is less political tolerance for acceptance of that reality. But it still happens just the same.

[Jacob]

I agree with the folks who are saying that militaries are going to be willing to sacrifice X number of their own assets (be they people or materiel) for Y gain when it comes to AI drones - or any other military technology for that matter. And I think it's non-controversial to say that different militaries and societies will have different ratios they consider acceptable, dependent on their social and strategic situations.

As I see it, friendly fire incidents are only one failure mode of AI. FF definitely has the potential to be an obstacle to adaptation, depending on how and when they occur (PR does affect procurement, I'm pretty sure). That said, I think real challenge is to consistently (at whatever success rate) avoid not just FF but all the failure modes inherent in the highly complex and dynamic situation of an actual battle space - which I believe is much more complex than road traffic or landing commercial aircraft. Maybe that's close to being a solved problem, but it's certainly not something that I've see evidence of (though I'd be very interested in seeing such evidence).

It's an interesting area to watch, for sure, but my expectation is that we're still quite a while away from seeing autonomous AI in actual battle - even as an experiment.

[The Minsky Moment]

2) I believe that one of the current lessons from the Ukrainian-Russian war is that the Western way of war is superior to the Russian one due to a highly professional core of NCOs able to make autonomous decisions, able to understand the large tactical and strategic objectives (make decisions two levels above their rank, I believe people are saying), and to act independently on their own initiative. This type of decision making is even harder to program competently compared to even "fly around and kill all the enemies you see, according to this list of priority targets, while avoiding killing civilians or our own people."

Sure if the fight is between two conventional human armies, the one with better NCOs and officers is going to have an advantage.

But let's say you are in charge of a country that sees itself as potentially adversarial to the US and NATO, but one that is better resourced in Russia and smart enough to possibly learn some lessons.  For the sake of convenience, lets refer to this purely hypothetical country by the name: "China"

"China" might be well advised to invest their efforts in building up strong NCOs and developing smart junior officers, however, there are some concerns:
+ It takes a lot of time and costs a lot of money to do right.
+ "China" cannot easily access training from well-established armies that have these attributes
+ How do you know if and when you have succeeded in your goal?  From the policymaker perspective, you can throw money at the problem and run exercises, but can you really be sure of the strength of your personnel until tested?
+ Historically, some unfree regimes have been wary about taking this approach, because historically plenty of such regimes have been overthrown by junior officers supported by their loyal NCOs. That's the problem with creating a talented, disciplined officer corps and giving them fancy tools of war; one day they may decide they can run things a lot better than the corrupt idiots at the top.

On the other hand - imagine that "China" - while lacking a strong modern military tradition and an educational system that emphasized independence of action and thought - happens to be pretty darn good at manufacturing all sorts of stuff at scale, and is already investing huge resources into AI and robotics.  It might be tempting to pour resources into unmmaned and autonomous systems to take advantage of those strengths.  True, that would mean accepting that rival militaries might have superior officers and NCOs.  Then again, if positions and bases occupied by those fine individuals were flooded with attacks by cheap autonomous platforms, their effective numbers would dwindle and they would be replaced by less effective replacements.  Whereas the manufacturing lines churning out the unmanned systems would be just as effective as before - perhaps more so as upgrades were added in response to events on the battlefield.

Just a thought.

[Zanza]

Similar to the potential lower bar for autonomous AI for surface (or sub-surface) naval warfare compared to ground warfare, I suppose there may be a lower bar of entry for autonomous AI for contesting air supremacy.

A fire and forget air-to-air missle like a Sidewinder is already a mostly autonomous weapon. You lock in a target and then it does the rest itself, no more human attention needed. If you then let the platform from which it is fired fly (semi-)autonomous and fire missles based on own algorithms or input from say a human crewed Awacs, you have a mostly or fully autonomous air supremacy weapon. No need for expensive and rare pilots, no limitation on G the aircraft can sustain on maneuvering, no space wasted for the human compartment...

[Zanza]

2) I believe that one of the current lessons from the Ukrainian-Russian war is that the Western way of war is superior to the Russian one due to a highly professional core of NCOs able to make autonomous decisions, able to understand the large tactical and strategic objectives (make decisions two levels above their rank, I believe people are saying), and to act independently on their own initiative. This type of decision making is even harder to program competently compared to even "fly around and kill all the enemies you see, according to this list of priority targets, while avoiding killing civilians or our own people."

Let the NCOs command either directly the drones or a small team of drone pilots. No need to do away with the concept.

[Barrister]

+ Historically, some unfree regimes have been wary about taking this approach, because historically plenty of such regimes have been overthrown by junior officers supported by their loyal NCOs. That's the problem with creating a talented, disciplined officer corps and giving them fancy tools of war; one day they may decide they can run things a lot better than the corrupt idiots at the top.

Yeah, I wanted to say the top-down, highly hierarchical nature of the Russian military is a feature, not a bug.  The last thing Putin wants is an army that can think for itself.  That's why the Russians didn't have a general in overall charge of the Ukrainian campaign until very recently - it all had to go through Putin.

[Zanza]

Still, I think the problem of complexity is not so much about kill ratios (though that shouldn't be discounted as an issue) but about vulnerability and efficacy. Highly complex dynamic environments with many edge cases create the risk of non-desired behaviour as well as vulnerabilities to exploits (as players of many online games can attest to). This, of course, can be iterated through, but that too is a non-trivial task.

Just make 10.000 reasonably cheap drones. Even if one percent is lost to dege cases that you did not consider, you still have 9.900 drones left. Even if half of them are lost, your enemy still has to deal with 5.000 drones...

[Jacob]

A fire and forget air-to-air missle like a Sidewinder is already a mostly autonomous weapon. You lock in a target and then it does the rest itself, no more human attention needed. If you then let the platform from which it is fired fly (semi-)autonomous and fire missles based on own algorithms or input from say a human crewed Awacs, you have a mostly or fully autonomous air supremacy weapon. No need for expensive and rare pilots, no limitation on G the aircraft can sustain on maneuvering, no space wasted for the human compartment...

Let the NCOs command either directly the drones or a small team of drone pilots. No need to do away with the concept.

No disagreement there. The target selection and NCO parts is the part that I think AI will struggle the most with, so keeping the NCO in the loop for selecting targets makes a whole lot of sense to me.

[Jacob]

Just make 10.000 reasonably cheap drones. Even if one percent is lost to dege cases that you did not consider, you still have 9.900 drones left. Even if half of them are lost, your enemy still has to deal with 5.000 drones...

Easy enough. Release the dogs of war flocks of Languish Lethal Lemmings Mk 1.0!

What do you think the role of these 10,000 reasonably cheap drones darkening the sky are going to be? Are they a replacement for artillery? For infantry? Tanks? Helicopters? Jets? All of the above?

[The Brain]

What's the cost difference between a ground-attack drone and an anti-drone drone?

[Jacob]

... Just a thought.

That's a reasonably plausible scenario. Flooding the battle space with replacable "good enough for the mission" autonomous AI drones seems a winning move for whoever can do it. Maybe China can. It'll be interesting to see what sort of timelines are required for doing that

Seems like Taiwan should invest in their own drone fleet, as well as counter drone capabilities.