Comment by lsy | Hacker News Reader

Comment by lsy | Hacker News Reader

This is a very urgent-sounding article about China's supposed AI superpowers and how they will dominate the battlefield, coincidentally written by the head of an American "AI" company who would likely personally benefit from shifting $150B into AI research, as suggested. Complete with AI-industry standard claims about technology that is "less than 10 years away", and sci-fi about "autonomous, adaptive drone swarms".

War is so much more complicated than the author wants readers to think. Raw technological dominance does not assure victory, as the US found after 20 years in Afghanistan. AI, for all its impressive feats, continues to operate without flexibility or common sense, and foundational concepts like geography or economic influence remain the most important contributors to success in the military sphere.

Even if one takes the continuance of the DoD's aims as an unalloyed good, what areas of its budget would go unserved with 25% redirected to AI research? How do those cuts or reprioritizations affect US military capability? I don't suspect the author has thought about this very much, and yet he feels qualified to write yet another AI/China scare article, which more and more seem designed to terrify politicians who determine funding rather than to convince military experts.

> "War is so much more complicated than the author wants readers to think. Raw technological dominance does not assure victory, as the US found after 20 years in Afghanistan."

I'm not sure that really proves your point, in that the US' problem in Afghanistan wasn't that their tech dominance wasn't enough - it clearly was, they pretty much had control of the country for years, no?

Where they failed was that in a fairly rare example of a "victory target" for a war, wanted to not just win militarily but also to change the minds/culture of an entire country to the point where they could withdraw and have the country keep acting the way they were enforcing while there.

Sure they failed at that, but had the goal been to take control of Afghanistan and make it a US colony, permanently occupied for the foreseeable future, that would've been doable (at least technically, ignoring US domestic politics etc etc)

Historically most wars have been about seizing land/wealth/people, not about changing a country's culture and then walking away. Surely drones which patrol the skies and kills anyone at the first sign of any sort of resistance (holding a gun, crossing a forbidden line, whatever) would've made holding the country a lot easier from a technical point of view.

Imagine if the US army had, when in Afghanistan, been given the same orders as say UK was giving back when building an empire "make them submit, kill those who don't, that's going to be our land" is there any doubt that their military strength wouldn't have been enough against Afghan rebels who in the real case could hide in mountains until America ran out of time and withdrew (which wouldn't happen if they hoped to own the country)?

So sure I agree with you that war isn't as simple as technical dominance as the reasons for war etc come into play, but since most wars have different reasons to the US' i Afghanistan... wouldn't Putin love drones that could fill Ukranian skies and kill anyone at the first sight of a weapon (while avoiding Russian soldiers)?

WWII was very much about changing the culture, and the US was pretty successful at that in Germany and even Japan (read about Japan in 1930s, it's terrifying).

The USSR was also somehow successful (see Eastern Bloc), fortunately much less so.

To your point, I literally laughed out loud at this in the article:

Based on the pace of progress with AI technology today, I believe this is less than 10 years away.

How does anyone write things like this and expect to be taken seriously with the track record of AI regarding real world things like autonomous driving?

Surely someone has a "law" that states generalized AI is always 10 years away.

It's usually predicted as 25-50 years away because of:

https://en.wikipedia.org/wiki/Maes%E2%80%93Garreau_law

A 10 year prediction is because this is the CEO of a company saying it. So the lifespan used in the prediction is that of the company and not the person.

> autonomous driving

Autonomous driving is "hard", where hard is mininizing unintentaional death for domestic adoption.

Autonmous murdering is easier, perhaps much easier. Because flying from point A to B without obstacles is simpler than driving, and droning a few thousand innocents due to permissive ROE is forgone. It's a vastly different calculations in terms of allowable deaths and hardware attrition.

Autonomous murdering is easier, perhaps much easier.

Is it really though? One variable that autonomous driving has in its favor is nobody is actively trying to disrupt it.

What happens when people put up nets around key areas? Fishing line? GPS jammers? Lasers to disable cameras? LiDAR blocking? Radar jamming? EMP? Radio jamming? All of this is being studied today and will at least keep pace with AI advancements. For just one example:

https://nationalinterest.org/blog/buzz/chinese-engineers-sho...

Autonomous vehicles don’t have to deal with any of these. Autonomous murder devices most assuredly will because people have a strong incentive to prevent their own murder.

IMO yes, mass attritionable UAVs exacerbates fundemental disadvantage of anti air - hard to defend everywhere vs air power easier to concentrate to create local tactical superiority.

Countermeasures can keep pace but doesn't mean they can match scale. It's typically going to be easier to mass UAVs who can use standoff munitions than to preposition defensive assets, especially land based ones, in sufficient concentrations which can be countered by adding more UAVs during mission planning. Countermeasures for high performance UAVs, essentially unmanned fighters instead of low tier drones will be very difficult. Advanced US/PRC HALE drones fly much higher and potentially faster, there's a reason counters like DE or various form of jamming target their ordnances at lower attitudes, not high flying, fast flying drones themselves which still requires combination of high performing ground and air based interceptors that can hit high. Except UAVs can tactically operate without life preservation in mind, which means much more permissive/aggressive operations = even more difficult requirements for counter-air infra.

Ultimately it's balance between offense / defensive is matter of industrial base, producing and positioning enough hardware to overwhelm opponents. Effects on battleground will be situational, i.e. large powers can already overwhelm smaller powers, and for peer powers it will be theatre specific, i.e. PRC can more easily overwhelm US carrier groups with mass drones because CSGs has finite VLS cells, power generation etc for defense whereas land based aviation can be massed at scale. Alternatively, PRC's very comprehensive anti-air network might be more easily penetrated by aggressive UAV formations using tactics manned missions won't attempt.

You have a valid point. But as a person who has worked in industrial automation, and also the military, yeah automatic murder is probably 50x easier than preventing human bodily harm.

Just having machinery around people is inherently dangerous.

But aside from complex concepts, The mine for example, is potentially one of the first Automated Killing devices, and is notoriously good at it.

Integrating technology into simple killing devices isn't very difficult compared to keeping people safe from harm. And while I agree that there are plenty of good ways to stop automated killing devices, you have to make big sacrifices to stop them, and that's another point to be made about this unique situation.

You either have to outpace your opponents technology, or be willing to fight without some or all of your own, because when you start introducing stuff like what you listed above, you can no longer use your own systems of defense, which then makes you vulnerable to other conventional forms of attack, giving the advantage to the technologically superior team.

In my opinion 'Terminator' might not be 10 years away, but automated killing devices are already in the field, and they aren't being reported on because they are classified. You would not believe the money and effort that goes into recovery of physical assets (working and destroyed) that are in theatre.

How much do you think the Ukrainians (or the Russians, if you're a psychopath) would pay for a fully automated missile that detects the source of incoming artillery, launches in that direction, identifies the gun, and blows it up?

It really doesn't seem much of a stretch from the current state of technology. The only thing missing is an AI-driven targeting system for the last mile. I sincerely hope somebody (on "our" side) is working on it.

I don’t because “their side” will eventually reverse engineer it as they have with all other weapons

Not very different from systems like shotspotter deployed in us cities, modern couter artillery systems

You mean like https://en.m.wikipedia.org/wiki/Counter-battery_radar ?

Ukrainians are using Polish TOPAZ system that does mostly that. Not AI needed (radar systems). Probably radars or satelitte images could be imroved with AI though. I think with somethink like starlink you could use tracking tech for tracking foreign military but US seems to have something like that. Probably also smart to automate analyst like in Russia-Ukraine war you could source info from relevant Telegram channels. Also tracking of social media for any leaks (seems like it’s present already).

What ukrainians needed was their own satelitte that got crowdfounded as US didn’t want to share all intel in real time. What they need now is long range rockets and then Russia will start peace negotiations.

This is a somewhat dated (not as in obsolete, but as in used for decades) and normal part of warfare. Artillery combat generally involves "shoot and scoot" strategy since the units tend to be identified quite quickly, in some cases before their first shells have even landed.

All AI would do is pull the trigger in the end. And I'd always rather that be done by a human. If AI were in charge of nuclear deployments during the Cold War, it's likely we would have seen nuclear annihilation since there were genuine false positives where a nuclear response was only stopped by a human breaking the chain of command and refusing to escalate as he "should have." [1] With AI we needn't worry about such 'discipline problems.'

[1] - https://en.wikipedia.org/wiki/Stanislav_Petrov

Counter battery radar only tells you where the enemy was when they fired, not where they are now. Even with drone surveillance, the drone has to find the target before it gets shot down, and then the information has to be acted on before they move. Also counter battery radars are themselves are vulnerable targets. It doesn't seem implausible to suggest that within a decade there will be missiles or loitering munitions with image detection programs for seeking out enemy artillery (or ships, or tanks, or even infantry) that would render shoot and scoot much less effective. It wouldn't require anything like AGI. There are all sorts of ethical issues with this idea, but the PLA isn't going to care about that. Not really sure how the West is going to handle the fact the PLA will have weapons that much of the western public are against developing in principle. I guess that's another parallel with nuclear weapons.

Reply on news.ycombinator.com