A thread for unboxing AI

I have certainly written no such thing, but I don't control how others read my posts or what they think I say. I would say it that ignoring the past, the present and the immediate predictable future is an incredibly stupid thing to do. Those will always be the most reliable indicators of future war and doctrine.

As for what the future brings for war and drone warfare, you haven't really made much in the way of predictions. You obviously think there is great horror coming, but you haven't really gone into any specifics.

And yes, that does make me dubious about the argument. Firing thousands of drones of various types at a city can certainly be horrific, but plenty of weapons are horrific in war. Anyone who thinks that is downplaying the horrors of drone warfare or future of drone warfare really needs to go and study the history of war. It is the exact opposite of that.

I'm starting to realize that people are so desensitized to war and weapons of war that they don't realize that making comparisons to consequences and cost of using of traditional weapons like mines, bombs, guided bombs, cruise missiles and conventional indirect fire weapons (airborne drone weapons generally combine elements of all these) is not an endorsement.

People are also conflating capacity with outcome. You can certainly slaughter an entire city with drones. However, you can slaughter an entire city with knives and clubs, just look to the Rwandan genocide - to date one of the most gruesomely "effective" genocides in history that outpaced the Holocaust in killings per day.

This is why I am very hesitant to say that war will become "worse", because while such a debate can be intriguing, for the most part I think it reflects how how little many people know of war.

Oh okay its just not knowing its already alreally really horrible.

I built an MCP Gateway for the Datacenter yesterday. Don't know why I didn't do it sooner. This way we can make sure folks are using read only tokens for production systems and get audit logs for SOX.

I hooked my new monitoring platform lab MCP Server to my local Codex and bam AIOps inya face. Local Codex has integrations into the user's slack, outlook, SharePoint. So it can be pretty smart about what's going on in the company.

Yes, the problem in this discussion is obviously my ignorance of contemporary warfare.

I don't even know what you are trying to say. Apparently, war is "worse" and war will be "even worse", but you are not willing to provide details. That sucks for everyone else trying to understand you, because "worse" is actually a comparative form.

My best guess is that you think war now involves new threats that some nations that have traditionally enjoyed supremacy are not equipped to deal with, and therefore you find yourself uncertain of how future wars will play out.

However, that isn't war itself getting worse. That is just you finding on how the other half lives.

I never said you were.

I don't have any interest in trying to persuade people that AI robotics/etc will be more horrific than land mines. Any more than i have any interest in persuading people climate change is a real danger etc etc. I think trying to persuade people of these things becomes a counter productive distraction after a certain point - which is long past. If you don't think there's anything to see here then fine. We can just disagree.

You could start by explaining what "more horrific" and "worse" refers to. These are comparative terms. For example, are you arguing that there will come wars that are worse than has ever been seen before?

I doubt it will help but:

a) The potential to terrorize, control, destroy and provide an existential threat to individuals, groups, peoples, places and countries will increase

b) It's a bit like nukes but harder to contain as a) there's no big moment and b) it can be selective.

Your turn. If your thoughts the risks were real then what political action would you support differently? Nothing, as before with renewed urgency or something new? (I'm in the urgency camp)

It's always amusing to watch another poster get chezzed for the first time. Welcome to the club!

You will see the same vague flailing whenever you ask an AI doomer to name a specific threat that does not already exist.

It's not teh first time.

I this case it's pretty simple and not exactly controversial but T_D is trying to shoehorn it into a position I don't hold at all.

I'm not a doomer.

I think we're seeing the same sort of denial we see from those arguing against climate change. It's not about doom from me, it's just about the need for good politics in the face of new threats or developments

I was saying, similar to what chez is posting, you will also hear this vague etc from ai doomers. Idk what you are.

Climate Change deniers have to deny reams of science. AI Doom deniers have to deny science fiction plots.

I am a Capitalism Doomer. Its's all going down anyway. I want AGI out of it. Soft Accelerationism.

I think capitalism as we have known it is dying but although change in itself can be very painful, I dont think it's a bad thing overall.

I'd say I worry about democracy more than anything else. If we don't get our act together then the china model is likely to look increasingly attractive.

Indeed not. Thus my joke.

Ah I've been chezzed by rococo

Wait what's the downside.

fascism in some form seems like a real risk

Fascism is a given, not a risk.

You said China model, I took that as meaning a mixed economy with socialist tendencies what did you mean.

oh okay the china model only. I think democracy is the only moral option. But few seem to value it very much beyond any advantage it delivers.

The last week was sort of a turning point, with general AI doming up with a counter example to Erdos' unit distance problem in a pretty unexpected way. People have tried to prove it or find counter examples for a while and it did so without using purely geometric methods and in a way i don't think humans would have felt is likely to work.

Thanks, I missed the Erdos thing.

Terry Tao is also 100% in on AI assisted proof and releases a YouTube video once in a while.

Also in news, Karpathy, who confounded OpenAI, used to run Teslas AI, and invented the term vibe coding, is joining Anthropic to work on model self-improvement.

Drones in its current format serves a doctrinal purpose that covers a fairly wide variety of traditional weapons and combinations of those.

These would be:

Typically, these uses are complimentary. Drones have advantages over these weapons, but also disadvantages. A drone can be cheaper than a missile, but slower. It can be fired in far larger numbers than ballistic missiles, but is much easier to shoot down. It can be more difficult to mass produce than an artillery shell, but it is more precise, easier to shoot and scoot and you can keep the human element much farther back from the front-line.

The question of how much harm they can do is a question of doctrine, as it is with all these weapons. For example, using wide-area-explosives in urban areas is known to come with up to a 1:9 military:civilian casualty rate (carpet bombing) and down to something like 50:50 if you take extreme care, but doctrines will make that number vary wildly. Drones have the potential to be both more precise than current weapons, but on the flip side they also allow for scale of attack on a bigger level and for attacks to be carried out with less precision than current weaponry.

The AI element as of today is varied, but is often used as a fallback option against EW-warfare or long-distance attacks where remote control is complicated. Basically, the drones can assume their own targeting for bypassing EW-interference aimed at the remote-control signals. There does also exists drones with autonomy that ranges from fully autonomous after launch, autonomous when set in that mode and autonomous with human oversight. AI-control also exists on a wide spectrum. Comparative guided systems that at some point take over control of the ammunition / weapon has been a reality for almost 80 years. Similarly, mine warfare where the weapon / ammunition is fully outside human control once deployed has existed since the mid 1800s, (far longer than that if we just count the technology, but we will limited ourselves to actual functional doctrine and tech here). Fully autonomous weapons and weapons platforms is new, and the most likely area that would be the target of policy, doctrine and potential agreements.

In short, there is no single form of drone or drone warfare that could or would be controlled by a single policy. Their form and use case exist across such a wide spectrum that such a document would have a scope so large to be unwieldy and also pretty much have zero chance of success of ever being passed as an agreement between nations.

Hence, it makes more sense to view some policy on drones as potential extensions to existing doctrines, some policy as referring to specific types of drone classifications (which do not really yet exist). Policy in regard to AI control follow a similar vein. A drone using AI-control in the same fashion that a traditional guided missile uses automated targeting is a very different thing from a loitering AI-guided weapons platform that engages or can engage fully outside human control after launch.

Will such policy or agreements pass? That depends. Nations are generally eager to agree on the use of weaponry if it offers them an advantage or no added disadvantage, but reluctant to agree on the use of weaponry where they think it offers them a disadvantage or takes away an advantage. The modern history of mine warfare is the the best case study for this trend. Agreements are broken or withdrawn from following the same logic.

This is a short summary of the background I use to respond to your bullet points:

I'm unsure if the increase is the correct term. I think change is more apt. Nations that have traditionally enjoyed the safety of military supremacy might see the threat increase in the foreseeable future, but nations that have traditionally been at a disadvantage might see a more even playing field. Nations that have traditionally been very weak in this regard will see little change.

This will likely shift as drone warfare and technology evolve. Industrial capacity and technological advantage will likely slowly shift the window in favour of nations with higher capacity in those areas.

As for terrorism, there is certainly scope for drones to complicate that. But this will likely take the form of new forms of attacks rather than some great overall increase in threat level. Some morons with a drone-carried pipe bomb will be about the same threat as morons with the pipe bombs in a bag, but the drone could change how the attack that takes place and what it strikes. If we start to see such attacks, I think it is likely that the outcome will become drone bans.

This is currently more sci-fi territory, as if we assume drones or drone warfare that are the equivalent of these weapons without carrying the same warheads, then this does not yet exist. It could also be the same question, as drones could easily be used as platforms to carry weapons of mass destruction. On one level, it is a natural fit, as warheads of such types can be very light. On the other hand, it is a bad fit, as the chance of your warhead falling into the wrong hands is much greater (you generally want to fit such weapons on platforms where you know the weapon will either detonate or destruct).

If we ignore the latter, we can certainly envision drones and AI combining to create horrifying scenarios. AI-controlled weapon satellites that makes space exploration impossible or render satellite technology defunct, Naval drones and AUVs that make seafaring dangerous and landlock us similar to how we were centuries ago, loitering weapons platforms that kill indiscriminately or even weapons, weapon platforms or command centres that go rogue in Skynet-esque fashion as the AI logic that control them fail. These things exist on a range of "already possible" to "not so far-fetched as it was 5 years ago".

However, it does not really offer up something new. If international cooperation keeps breaking down, these developments could happen even without drone technology or AI technology. The most destructive war in human history was carried out some 700 years ago and was carried out with medieval siege weapons, primitive explosives and lance, swords, spears and bow.

Securing peace and prosperity is done via diplomacy, cooperation and being expensive to conquer. There exists no level of technology that can save you or anyone else when international order breaks down.

For what it is worth: If the reply to this post is a flippant remark or a one-liner, I will assume you have absolutely zero interest in discussing this topic and are just shitposting.

Thank you for the detailed response. I don't think I can do it justice in one response but hopefully it is the basis for a much better discussion.

Firstly a few things I think we much agree on:

Development is not going to be stopped by policies/treaties. Usage will not be prevented by treaties if the international order breaks down. Personally, I'm of the view that treaties against use or development are very dangerous if it allows people to imagine they wont be developed/used.

Totally agree. it is the point I was trying to make at the beginning. Renewed urgency is required, complacency will kill us. Even if we disagree on how much more dangerous the new developments are, simply the fact of such fast change/development is destabilising.

I have no quarrel with your wide definition of drones. Whatever we call them I'm including crawling, walking etc as well as well as flying.

Where we maybe start to diverge is you seem very focused on the current state of affairs whereas I think 'now' in the political sense covers likely developments over the next few decades as a minimum. We can agree on no science fiction but science fiction is stuff like AI consciousness, singularities etc. None of what is happening now has been far fetched for 30 years at least - fwiw it's happening faster than I expected but there's nothing remotely unexpected or requiring uncertain leaps. And plenty more of it is on the way. Maybe 3d printing is newer.

In particular

was't about science fiction. I'm saying that unlike a nuclear bomb which is a very noticeable well defined event, deployment of AI/robotics etc happens gradually and in a myriad of different ways. That makes treaties on use/development more or less impossible. I think you made this point yourself so I assume I was being unclear.

Re selective - this is where I think a lot of the 'worse' comes in. Mass deployment of cheap 'drones' that can wait, hide, hunt and select targets is a new threat. Their use to target areas or people matching various criteria (or sparing only those who do) is going to be a new order of scary/controlling. Where we may disagree is that I don't think this is remotely far fetched from a tech pov. So as above we need to be focusing on defense and international cooperation.

Another part of it being worse is the need for people. I take your point about wars in the past but at least they required huge armies. 'boots on the ground' with lots of body bags coming home is a significant deterrent. Obviously horrors can happen anyway but it's so much easier to do the stuff that doesn't require your own casualties. And so much easier to replenish.

Another part (outside of science fiction) is the speed of deployment when AI is deciding on targets rather than humans doing it. This creates it's own terror/control.