OpenAI deletes ban on using ChatGPT for "military and warfare"
theintercept.com
theintercept.com
I really do think this will be the company (or at lease technology) to unseat Google. Ironic that Google unseated Microsoft and now it looks like they will take their throne back.
Another example is the Occulus guy, that pivoted into levering something VR I guess to make stuff to kill other people since Zuckerberg crushed him.
It is just rebranded not changed in the last 70 years
Department of war was the “army” yes , but army is not what we think of army today, it literally included the entire air force - which wasn’t small auxiliary unit : both wars used tens of thousands of fighter planes .
Department of war was a department for war not just a name for what we know as army today .
…except the Navy and Marine Corps, which was the original point. A large part of the Second World War was fought by the naval services, despite those services being outside the Department of War. During the war, overall coordination of the military was not carried out by the War Department but rather via the Joint Chiefs of Staff and through ad hoc high level coordination between Army and Naval command. In particular, command in the Pacific theater was split between General Douglas MacArthur Admiral Chester Nimitz (with the ground operations under Nimitz being primarily carried out by Marines). The difficulties caused by this approach were the primary motivation for the reorganization of the American military into a unified Department of Defense.
I am well aware that the Army Air Forces were a very large part of the Army during the Second World War. However, the Navy and Marines also had tens of thousands of airplanes, none of which were under the control of the War Department or the Army Air Forces. It’s a little misleading to claim the War Department controlled the “entire air force” when they only controlled the Army Air Forces and not naval aviation.
The War Department controlled the Army and the Army Air Forces, which were part of the Army at the time, so it’s just as correct and a lot quicker to say that the War Department was only in charge of the Army. It wasn’t in charge of the Navy and Marines, and it wasn’t even in charge of fighting wars because we needed the Navy and Marines to help fight wars and they were under a different department. Which, again, was the reason for forming the Department of Defense in the first place.
When the Air Force became an independent service during the postwar military reorganizations, they actually tried (and failed) to take over naval aviation; to this day the United States Navy has a larger air force than most countries.
If the elites can't play nicely with each other it is not my problem. I don't trust them.
or, its just a russian propaganda... Could you please next time start you comments with disclaimer "according to russian propaganda.."?
From your other comment:
> You also had Russians finding a lot of virus labs in ukraine etc.
You understand that no one except russian trolls believes in this lies? On second though I think even russian trolls do not believe in it..
Oof. It is not AGI, so does not count (sarcasm).
Following which, you will need to learn how to defend against all the internal threats to any such system. ^_^
For your service, here is a limited edition digital flag pin emoji:
Fair enough, but that's not who funded your research (according to your own disclosure, the military paid for it).
If this topic is so useful for "more situations" , why didn't those "many more situations" fund it? Will you be conducting research into how this topic will have non-military usages, or is that just something you tell yourself to sleep better at night while the military pays for more research that "is useful for many more situations than just the military"?
Do you feel bad every time you use a microwave? It was originally a military radar. Without military funding, it would not exist in its modern form. Nor would basically all radio communication technologies, satellites, spaceflight. You get the idea.
I would prefer it if we developed competitive qualities. Trying to stop doesn't make sense, we have to outgrow it.
The new "Gifted And Talented Education" GATE for the modern era.**
We were evaluated for GATE in like 4th grade? Using such AI human behavioral heuristics against the minor population in 3.2.1.... Contract.
So they can be sidelined before they have a chance to disrupt anything.
Especially those individuals having any unique abilities in excess of what AI could substitute for.
In general, all technology can be reused. Maybe someday killer robots walk around with a Rust codebase. Should Rust not have been developed?
Your current government may be risk averse, the one that gets elected five, or fifteen years from now, in a different climate...
Your comparison between this work and Rust is not a good one. There's degrees of potential for abuse, and a system that predicts people's future behaviour in the hands of the military is nothing like a general-purpose language.
The best comparison I can give would be if we were talking about health care and our product was used to schedule nurses shifts or book operating rooms.
It's probably why OpenAI decided to remove the restriction.
Also, for a company that proved it has no governance, I'm surprised they didn't quietly do it anyway and wait until it was discovered.
It wasn't removed https://abc.xyz/investor/google-code-of-conduct/
"And remember... don’t be evil, and if you see something that you think isn’t right – speak up!"
This is Google telling ME to not be evil not Google telling itself to not be evil. Big difference. Its sounds more like "if you see something, say something" snitch culture. That's evil.
I normally have a very dim view of OpenAI/Altman but in this case I wonder if is something akin to a warrant canary, except for 5th generation warfare?
Altman does seem to have a bit of Chomsky in him, so it's not impossible.
For the ignorant (me), what do you mean? I'm guessing SOP is "standard operating procedure" given the military context of this thread, but MIC? No clue.
It's really helpful to just spell out the words, at least the first time.
The topic sometimes puts one into acronym mode :)
(re: acronymfinder - funny enough, though 90% of AI is shit this bit falls into the 10% where it's pretty good at explaining acronyms, i use it often enough with MBA types, the depths of its hidden context are the untapped goldmines)
I think you figured out at least one important factor in this - I forgot about warrant canaries until your comment... So thanks for resurrecting that.
What we should then ask, is for a OpenAI PUBLIC SECTOR Contract Details Dashboard" <-- Meaning, they should be required to show all the "open" AI they have being built on their systems if they want us to have faith in Safe Responsible Humane Alignments?
--
@dylan604:
>* Timelines are much more compressed now*
The Smart hockey-stick :-)
But yeah - and the worse part is not just Corporate competitors, but bad actors, rogue states, triads, ALL the Mafia's/scammers/phishers/ransomers/coin snatchers...
They all benefit at an unprecedented EQUAL rate at this point with the broad spectrum, capability, cost-effectiveness and effectively un-regulated, not-yet-aligned/guard-railed AIs here, in dev, and in near term.
Its got to be an amazing spot if your a top notch cybercrime person in the A-game right now. That, and White-Hat AI-pentesting, and next level security is behind schedule, it seems.
---
Also, I posted this regarding national security status for cloud provider's infra: [0]
In the increasingly interconnected global economy, the reliance on Cloud Services raises questions about the national security implications of data centers. As these critical economic infrastructure sites, often strategically located underground, underwater, or in remote-cold locales, play a pivotal role, considerations arise regarding the role of military forces in safeguarding their security.
While physical security measures and location obscurity provide some protection, the integration of AI into various aspects of daily life and the pervasive influence of cloud-based technologies on devices, as evident in CES GPT-enabled products, further accentuates the importance of these infrastructure sites.
Notably, instances such as the seizure of a college thesis mapping communication lines in the U.S. underscore the sensitivity of disclosing key communications infrastructure.
Companies like AWS, running data centers for the Department of Defense (DoD) and Intelligence Community (IC), demonstrate close collaboration between private entities and defense agencies. The question remains: are major cloud service providers actively involved in a national security strategy to protect the private internet infrastructure that underpins the global economy, or does the responsibility solely rest with individual companies?
(TBH i also forgot about warrant canaries for a good while. I just threw the question out to myself "what if Altman actually doesn't WANT to be a badguy, what might he have done to signal his Borgification?")
An AI with the capability to be autonomous in any environment, able to plan and execute plans well enough to defeat human opponents, is exactly what the AI doomer POV is rightly afraid of.
If the tech were truly superior, we would see Apple, Google and Meta rushing to license their tech. Yet they're not, instead, they're all building their own version. There's no secret sauce left to building an LLM. It's all public knowledge. And while ChatGPT has an edge right now, it's not a substantial one.
oh noes not the terms of service!
Could you share something that you se and think "profound"?
GPT-4 helps me write reams of code every day at my job.
Google tried to beat it with Gemini Ultra, but as well as being unreleased, the stats in the paper don't lend much confidence it will beat gpt-4-0314
Content owners will decide on the price and even who to license to.
Content owners will be asking exorbitant amounts for licensing fees and will likely strike exclusive deals with LLM owners.
Maybe Microsoft actually bought GitHub for the content?
Content owners are moving quickly to monetize their hordes of data. The days of free training data are over. If you don’t own it already, you probably can’t afford it.
I think the interesting side effect is that LLMs will end up as bifurcated as the internet. Each only being trained on some subset of content based on which subset the LLM builder chooses or is able to license for training.
LLM agents will all be hamstrung and biased in various ways based on fragmented training sets.
There will be no singularity.
There will be many LLM agents that learned to think based on the information their creators could afford or chose to provide it.
These agents will have biased, inaccurate and incomplete world views, yet will be very confident they know everything. How very human!
It's inevitable that dangerous real time AI capable of formulating plans are going to be developed for military purposes.
The "OODA Loop" is fundamental to combat. (Observe, Orient, Decide, Act) Having a tighter and more potent (in the sense of fast and accurate processing) OODA loop is a fundamental advantage. The economics and game theory of military combat is going to result in units which have potent OODA loops which can overwhelm a human being's OODA loop. Once that happens, competition between different sides will result in an arms race going far above that level of capability.
Once the above happens, it's disturbingly likely that instrumental goals will arise in such entities which are dangerous not only to human beings on the wrong side, but to human beings on any side.
The weird prejudice in your comment against "non-nerds" and your assertion that they "colonized" the internet and "demanded...evil things" doesn't reflect reality. Knowledge of computer science does not correlate with moral or ethical qualities of character.
[0]https://qz.com/1145669/googles-true-origin-partly-lies-in-ci...
They removed it because it's stupid and meaningless.
> “use our service to harm yourself or others” and gives “develop or use weapons” as an example, but the blanket ban on “military and warfare” use has vanished.
so Lockheed and co won't be able to use it for most of their military projects. I don't personally see an issue with this change in policy given what you said: the vast vast majority of usecases is just mundane office spreadsheet stuff and the worrying stuff like AI powered drones is disallowed (DARPA has that covered anyway).
Americans, and every other country's, citizens all pay for the inefficiency of the large defense departments. A slightly more efficient DoD office drone isn't exactly a more dangerous world IMO.
Making this killing machine more efficient at doing anything besides dying or self-dismantlement is harmful to liberation around the world.
If it matters to you, Congress has already unequivocally signaled unanimous support. That's how this works under 50 USC 33: the executive can launch attacks unilaterally with 48 hours notices (here it was negative hours notice), thus giving Congress the opportunity to pass a joint resolution ending the strikes. The opposite thing happened.
That is an extremely slanted view of the situation. The Houthis attacked plenty of ships belonging to (de iure or de facto) multiple nations and carrying plenty of someone else's cargo, thus disrupting about 12 per cent of the total volume of global trade. They aren't even trying to enforce a specific blockade ( blockade is act of war, but it must be limited to very specific cargo/ships).
That is piracy 101 and pirates have been generally considered enemies of mankind since at least Antiquity.
"liberation around the world."
Yeah, like the way the Russians are "liberating" Bakhmut and Avdiivka. Without the Western militaries and their help, they could have "liberated" the entire Ukraine into one big smouldering heap of ruins.
I think this is where this is coming from.
It would be useful to clarify the rules and ban direct automatic control of weapons or indirect control as part of a feedback loop on any system involving weapons from commercial AI products.
DJI FPV ($1k retail) supposedly has about 1 in 3 chance of taking out a moving tank that is not a sitting duck - i.e. hatches closed etc - and does not expose the operator to danger in the process. For comparison, a single Javelin is $240k. FPV drones are cheap enough that they're routinely used even against minor targets such as unarmored small cars and even individual soldiers - even after accounting for failures before a successful hit, it still ends up costing the enemy a lot more in $$$ than was spent on all the drones.
Starlink is cheap for its capabilities but it is still startup cost of billions of dollars if you could do it all .
And no, you don't need Internet connectivity for those things. You need to be within radio range, but this can still mean 1-2 km away easily even on stock equipment, and more with better antennae etc.
Javelin has a warhead capacity of 9Kg, for the same bang so to speak ( armor penetration) you would need drones that can hoist 9Kg, those don't retail at anywhere near $1,000.
That doesn't mean that cheap drones are not effective means of weapons delivery, they are as this war is showing. They are akin to side arms, when penetration is not a factor they are useful, but they are hardly a replacement for heavy guided munitions like Javelin.
Man Pads are by no means perfect, the current gen is 30 years old, they could be made much smarter and lot cheaper than $250,000, however just $1000 drone is not going to replace them yet.
And the main advantage for Javelin over an FPV drone is that a Javelin is fire-and-forget; you don’t need to steer the missile yourself. It will attack the weaker top armor automatically.
> "Javelin has a warhead capacity of 9Kg, for the same bang so to speak ( armor penetration) you would need drones that can hoist 9Kg"
This is false because a Javelin warhead contains propellant and guidance as well as the payload. In reality, a shaped charge capable of penetrating 200mm of steel weighs about 1 kilogram. Which is well within the capabilities of many drones under and around the 1000$ price range.
Many tanks and IFVs have been taken out on both sides by quadcopters carrying shaped charges. Plenty of penetration can be achieved by relatively lightweight shaped charges.
A single-stage RPG-7 HEAT grenade weighs 2.5 kg, of which the actual explosive charge is only 700 g, and much of the rest is the powder charge that propels it in normal use, but is completely unnecessary for drone applications. And, as already noted above, you do not need the same absolute armor penetration capacity for these things, because the whole point is to attack from the side where armor is the thinnest and any exposed internal components are the easiest to disable. Judging by the videos of successful use, the most common technique is to attack the engine compartment directly from above.
Now, DJI FPV, which does retail for $1K (and can be had for less if buying in bulk) is certainly quite capable of lifting a 1kg warhead. But it should also be noted that these days, in most cases Ukrainians are using money more efficiently by assembling their own custom-tuned FPV drones from components and specially manufactured HEAT charges, which knocks the price down to ~$600 for the same lift capacity with better speed and range. They are still roughly the same size as DJI or only slightly larger, which can be readily seen in numerous videos on YouTube and Telegram showcasing their use. Here's one example of a locally manufactured FPV drone: https://neboperemogy.fund/dron-hrim/ - the lift capacity for this one is 2kg, sufficient for heavier tandem HEAT charges, and it costs ~$750.
well, FPV is quickly becoming yesterday stuff - the EM warfare makes direct video connection infeasible, at least near the target. So, the most recent drones can guide themselves toward the target using computer vision - the operator brings them closer to the target's position, points-and-clicks at the target on screen, and the drone does the rest on its own.
I think a recent war and a recent bombing campaign may imply otherwise.
> Don’t use our service to harm yourself or others – for example, don’t use our services to [. . .] develop or use weapons, injure others or destroy property [. . .]
Humans are primates that operate in hierarchies and compete for territory and resources. That's the real cause of any war, despite the lies used to supposedly make them into ethical issues.
And remember that the next time hundreds of millions of people from each side of globe have been convinced that mass murder of the other "evil" group is the only way to save the world.
Ultimately, I think WWIII will prove that humans really shouldn't be in control. We have to hope that we can invent something smarter, less violent, better organized, and less selfish.
Cheap AI drones that kill with precision by the millions are the ultimate weapon.
More powerful than Nuclear. Nuclear is a very big but blunt instrument of war. Drones are small and sharp instruments.
You can command it to kill every human in a geographic region without any other structural damage. It would happily obey that instruction.
Although I do find it weird that they're so concerned about their products being used in other, relatively less objectionable ways, but are OK with this.
More likely they are anticipating public news about military usage sooner than later and removing this to Mitigate the pr damage .
> In the interview, Kochavi recalled Israel’s 11-day war with Hamas in May 2021. He said, "In Operation Guardian of the Walls, once this machine was activated, it generated 100 new targets every day. To put it in perspective, in the past, we would produce 50 targets in Gaza in a year. Now, this machine created 100 targets in a single day, with 50 per cent of them being attacked."
> In 2021, the IDF launched what it referred to as the world’s "first AI war". It was the eleven-day offensive on Gaza known as “Operation Guardian of the Walls" that reportedly killed 261 Palestinians and injured 2,200.
It is sad, though, that they couldn't stand firm about what's right and become a building block for long-term world peace.
Another issue is full autonomy: even human soldiers/pilots/etc. are not fully autonomous - they will ask command if there are civilians or friendly units in the AO, as they don't always have enough information to make that decision themselves. To achieve that with machines (not necessarily AI), you either need a fully integrated system (i.e. humans are obsolete for military use), or you need an efficient and functional human-machine interface.
So I don't expect we'll be seeing fully autonomous AI weaponry anytime soon, despite it being technologically possible. AI-_assisted_ weaponry, however, probably already exists.
January 6, 2024:
"Defence Intelligence of Ukraine shares footage of the targeting of two Russian Pantsir-S1 air defence systems. Looks like loitering munition was used. As said, today in the Belgorod region of Russia."
Source tweet:
https://twitter.com/bayraktar_1love/status/17437042635308319...
Original source is the Main Directorate of Intelligence's Telegram channel (in Ukrainian):
Notice yellow rectangles that are visible around the targets in the video.
It seems that AI-aiming was used at the final parts of the approach trajectories, after loss of communications with the drones.
Both are vague and problematic to enforce, but the latter more so.
I think the sentiments around AI and war are mostly a bit naive. Of course AI is going to be weaponized. A lot of people think that's amoral, not ethical, etc. And they are right. In the wrong hands weapons can do a lot of harm and AI enabled weaponry might be really good at that. Of course, the whole point of war is actually harming the other side any way you can. And usually both sides think they are right and will want the best weapons to do that. So, yes, they'll want AI and are probably willing to spend lots on getting it.
And if you think about it, a lot of conflicts are actually needlessly bloody. AI might actually be more efficient at avoiding e.g. collateral damage and bringing conflicts to a conclusion sooner rather than later. Or preventing them entirely. Sort of the opposite of what we are seeing in Ukraine currently.
"I think AI has the potential to create infinitely stable dictatorships." - Ilya Sutskever
I would say it was a good thing that did not happen.
Unlike the US in Korea and Vietnam, the USSR wasn't even contemplating using nukes in Afghanistan or elsewhere.
Effectively, it's the military power of the US and its allies that prevents people like Vladimir Putin from killing potentially millions of people in their neighbouring countries. Whatever faults US has, it's still infinitely better than Russia. I say this as a citizen of a country that shares a long border with Russia.
While that carries weight, 'the other person is doing it' has long been an excuse for bad behavior. The essential goals are freedom, peace, and prosperity; dealing with Russia and China are means to an end, not the goal. If developing AI doesn't achieve the goal, we are failing.
But this time the atomic bomb will be able to decide by itself whether to incinerate human race.
First, Los Alamos was a project of a democratic government, serving the people of the US and allies. OpenAI is a business that serves itself.
During WWII, the US was in an existential war with the Nazis, also trying to develop nuclear weapons. If the Nazis had built it first, we may have been living now in a very dark world. (Obviously, it also helped defeat Japan.) On the other hand, there are threats if an enemy else develops capabilities that provide large advantages.
At least part of the answer, I think, is that the US government needs to take over development. It already houses the world's top two, most cutting edge technology organizations - the US military and NASA - and there is plenty more (NIST, NIH, nuclear power, etc.); the idea that somehow it's beyond the US government is a conservative trope and obviously false. We don't allow private organizations to develop military weapons (such as missiles and nukes) or bioweapons on their own recognizance; this is no different.
? where did I say that?
By doing what they've been doing, they won't hurt their stocks.
I don't have a full answer to my own question to be honest.
In the above example though, you'd never be able to prove that it was or wasn't your contribution so it's easy to say you bear no collective responsibility. But would it be true?
I'm not sure, but I can't say definitively you would bear no responsibility.
I'm trying to find an ethical analogue. If you dig up an old porno that I made 20 years ago, and show that to my child, am I responsible for the trauma it causes?
I would find it wrong to say "I bear no responsibility because that's just how things are" if that makes sense.
Going without gpt4 can put you at a disadvantage for some work.
Not paying taxes affects your life much more negatively.
Given the different costs, logically it seems that paying taxes would be something you have less collective responsibility for.
If so, Bravo, that's quite good.
(I mean, did anyone seriously think SV/California (or anyone, for that matter) would stand up to the military industrial complex? The one Eisenhower warned us all about??)
It's possible. It just won't make you rich, which is, I suspect, the real problem.
Yeah, and look what happened to them.
( https://en.wikipedia.org/wiki/Nissim_Amon & translation of the original message from Amon: https://sites.google.com/view/nissimamontranslation )
* https://www.cnn.com/2019/05/06/tech/facebook-groups-russia-f...
* https://www.voaafrica.com/a/israeli-firm-meddled-in-african-...
I sort of suspect AI-driven accounts are already present on social media, but I don't have proof.
The GPT-4 release documentation has some examples of this in its addendum. ChatGPT also seems to be good at writing advertisements. Without the strong guardrails, I wouldn't bet on one or two persons instructimg a GPT-4-scale model perfoming worse at manipulating debates than 10 or 100 humans without AI.
Turing test achieved. I don't know if the internet will lose its appeal because of this. Could be that in the future, to use an online service, you'll need to upload a human UUID.
Nothing would make the internet lose appeal to me faster than having to do something like that.
"Need $500? Rent out your UUID for marketing!"
https://www.mirror.co.uk/news/world-news/vladimir-putin-adds...
This piece of "journalism" released a couple of days ago claims Finland is in the process of joining NATO, while it already joined nearly a year ago. This is obviously caused by utilization of a LLM model with training data limited to time before Finland was accepted. At least at the end of the article they mention AI was utilized, and included an email where you can complain about factual errors.
all debates on social media have already been astroturfed to hell and back by professional posters for many years, but LLMs are certainly going to function as a force multiplier
https://scholar.google.com/citations?view_op=view_citation&h...
if you want some entertaining reading go ahead and browse through Eduardo Pasiliao's research at your working city there:
https://scholar.google.com/citations?user=Caw-nkAAAAAJ&hl=en
I'll summarize it for you: computational propaganda with an emphasis on discerning and disrupting social network structure.
Which unit?
https://en.wikipedia.org/wiki/Eglin_Air_Force_Base
Garrison for:
https://en.wikipedia.org/wiki/7th_Special_Forces_Group_(Unit...
Which is Army and part of:
https://en.wikipedia.org/wiki/1st_Special_Forces_Command_(Ai...
Whose psychological operations unit is based out of North Carolina. Doesn't track with Eglin.
I wonder if that's a fluke or exit node for a large number of Unclass networks that a lot of bored LCpl Schmuckatellis are using.
Not everything is a spy movie
It's used for operational efficiency: to select and bomb targets faster and in greater numbers than human analysts are able to
Not everything is boring paperwork
(Source: https://www.livemint.com/ai/israelhamas-war-how-ai-helps-isr... where AI achieved 730x improvement in bombing target selection rate and >300x greater rate of resulting bombs)
Now take an LLM that you can feed it questions or discussions from sites, have it jump in with what appears to be meaningful content, gets a bunch of "karma", then gradually start putting out the propaganda. It would be a hard item to fight.
> Today’s Mortgage Rates for Jan. 12, 2024: Rates Cool Off for Homeseekers
https://www.cnet.com/personal-finance/mortgages/todays-rates...
And yesterday
> Mortgage Rates for Jan. 11, 2024: Major Mortgage Rates Are Mixed Over the Last Week
https://www.cnet.com/personal-finance/mortgages/todays-rates...
And the day before
> Current Mortgage Interest Rates on Jan. 10, 2024: Rates Move Upward Over the Last Week
https://www.cnet.com/personal-finance/mortgages/todays-rates...
You get the idea.
Or, you know, it just hallucinating and people not checking it. But that would be as silly as lawyers citing non-existent AI-hallucinated legal cases.
Surprising no one, they picked the money.
(answer is yes, that's still banned! https://www.apple.com/legal/sla/docs/iTunes.pdf )
Who doesn’t think China has a ten year AI algorithm to takeover Taiwan? Israel+US+UK > Middle East.
SkyNet or War Games are likely already happening.
anybody who works in either AI or natsec
There's also something to be said about how the West's reliance on these war games (don't know if AI-powered or not) when preparing for the latest Ukrainian counter-offensive has had disastrous consequences for the actual Ukrainian soldiers on the field, but I don't think that Western military leaders are so honest with themselves anymore in order to acknowledge that (at least between themselves, if not to the public). A hint related to those Western war games in this Economist piece [1] from September 2023:
> Allied debates over strategy are hardly unusual. American and British officials worked closely with Ukraine in the months before it launched its counter-offensive in June. They gave intelligence and advice, conducted detailed war games to simulate how different attacks might play out, and helped design and train the brigades that received the lion’s share of Western equipment
What it is supposed to mean?
A) opt-out of participating to absolve yourself of future sins or
B) create the systems yourself, assuring you will have a say in the ethical rules engineered into the weapons
If you actually give a shit about ethics and safety (as opposed to the appearance thereof) the only logical choice is B.
OP choice was protest or participate and influence to safer outcomes. Your choice was protest or participate without influence to safer outcomes.
Also the AI participant would be OpenAI either way, whereas your inadequate alternative is participate with the US or NK will participate. Also, not the same.
So, wrong on two counts.
There are plenty of other weapons (such as mines) that the “international community” has “banned”, but are very useful in a war. Any country that doesn’t or can’t expect the US to come to its rescue ignores such bans and still manufactures them in great quantities.
2. Nerve gas is not strategic. A better comparison are nukes in WW2.
3. Nerve gas has no other uses unlike AI.
4. Nerve can only be used to hurt unlike AI
5. If AI in military is so dangerous, should the US just sit and do nothing while China /Russia deploy it fully? What is your suggestion here specifically?
Suppliers don't get to pick which house the missile lands on.