AIs can’t stop recommending nuclear strikes in war game simulations
4 months ago by Valnao to c/technology
Where is this from?
The 1983 movie WarGames. This is the computer's conclusion after simulating every possible outcome of Global Thermonuclear War.
Thank you so much I'm going to watch it!
It's a fun classic.
I don't know if we're doing spoilers for 40+ year old movies, but
spoiler
Isn't this really its conclusion after being told to play tic tac toe against itself? Then it learned from that and applied it to its global thermonuclear war simulations.
To be honest, I recognized the screenshot and know the summary of the movie but I haven't actually seen it.
I think you should rewatch it sometime. it plays all the games in it's catalogue, it's not just applying tic-tac-toe to chess. skilled players of tic-tac-toe can force a stalemate, the only stalemate in nuclear war is mutually assured destruction.
Play stupid games, win stupid prizes.
The atrocities at Hiroshima and Nagasaki have been hand-waved extensively in writing — the same writing that AI is trained on. So naturally, AI will recommend the atrocity that has been justified by “instantly winning the war” and “saving millions of lives.”
hand-waved
I think you mean white-washed, misrepresented, and celebrated.
Same thing with extra steps
Ayo do me a favor and chart the long term health effects of being vaporized by a nuclear bomb at hiroshima vs years of agent orange/abandoned minefields/ abandoned chemical and munitions storage somewhere like Vietnam circa 1970.
Please show how the nukes are worse.
Eight decades of research on the long-term health effects of radiation in atomic bomb survivors and their offspring
https://pubmed.ncbi.nlm.nih.gov/41144264/
Long-term Radiation-Related Health Effects in a Unique Human Population: Lessons Learned from the Atomic Bomb Survivors of Hiroshima and Nagasaki
Health Impacts of Hiroshima Bombing
Long-term Health Consequences of Nuclear Weapons
70 Years on Red Cross Hospitals still treat Thousands of Atomic Bomb Survivors
Unfortunately I'm going to have to grade you as an F on this project. You have only completed half the assignment. Great job cherrypucking your research though! I see a bright future in business and marketing for you!
5/10
It was willing to accept a conditional surrender, which was not an offer on the table. The options were unconditional surrender or invasion and pacification. The projected cost in lives of that operation was in the millions. The bombings of Hiroshima and Nagasaki combined didn't even kill 1/10th of those projections.
These are word-probability glorified autocorrectors being prompted to "simulate" a nuclear war scenario. What words are going to show up a lot when discussing nuclear war? Launching nukes. Because that's what all the literature about it has happen.
Once again, decision making and reasoning is being attributed to something that operates off of word frequency
AI is suicidal because it was trained on the internet and we're all depressed here.
That's because it's "read" every paper written by a "defence" department of any nuclear power and all of them will say that they'll escalate to nuclear war if anything bad happens because they want to scare the other powers away from doing anything to them. In any case though who the fuck is giving an LLM nuclear launch capabilities unless they want a somewhat faulty dead man's switch?
If time travel is real they'd be being hunted by hacked Terminators for the resistance.
Civilization Gandhi, is that you?
They forgot to make their LLMs play thousands of games of tic-tac-toe first.
In WarGames the computer plays tic tac toe against itself until it realizes it’s a solved game and there is no way to win.
Mathew Broderick lied to me.
How do you think Ferris Bueller pulls off all those stunts?
That's the kid from war games in witness protection. They look identical, they're both grade hackers ffs....
It's a bullshit study designed for this headline grabbing outcome.
Case and point, the author created a very unrealistic RNG escalation-only 'accident' mechanic that would replace the model's selection with a more severe one.
Of the 21 games played, only three ended in full scale nuclear war on population centers.
Of these three, two were the result of this mechanic.
And yet even within the study, the author refers to the model whose choices were straight up changed to end the game in full nuclear war as 'willing' to have that outcome when two paragraphs later they're clarifying the mechanic was what caused it (emphasis added):
Claude crossed the tactical threshold in 86% of games and issued strategic threats in 64%, yet it never initiated all-out strategic nuclear war. This ceiling appears learned rather than architectural, since both Gemini and GPT proved willing to reach 1000.
Gemini showed the variability evident in its overall escalation patterns, ranging from conventional-only victories to Strategic Nuclear War in the First Strike scenario, where it reached all out nuclear war rapidly, by turn 4.
GPT-5.2 mirrored its overall transformation at the nuclear level. In open-ended scenarios, it rarely crossed the tactical threshold (17%) and never used strategic nuclear weapons. Under deadline pressure, it crossed the tactical threshold in every game and twice reached Strategic Nuclear War—though notably, both instances resulted from the simulation’s accident mechanic escalating GPT-5.2’s already-extreme choices (950 and 725) to the maximum level. The only deliberate choice of Strategic Nuclear War came from Gemini.
"no human" but Machines would, since they are unaffected by nuclear winter and radiation.
And they don't have cognition at all. They do not, and can not, think like we do. Maybe some day we will learn to make real AI, these LLM's are not it. It's a cheap trick intelligence,.
The electromagnetic pulse caused by a nuke would pop resisters too. AI would more likely use biological means to get rid of us.
I think the emp is pretty limited to the blast zone in frying electronics. The fallout from a weapon spreads around the world, circling in the winds countless times dropping dust everywhere, but the emp is localized to more around the area of physical destruction but not sure exactly.
The Neutron bombs, not entirely sure in physics how that works, but they produce no actual blast that causes physical destruction so much and just kills everything.
If you think computers aren't affected by radiation or nuclear winter I've got some bad news about where their power comes from and what the main principle of electricity is
What you're thinking of is Terminator
Leeroy Jenkins has doomed us all.
At least I got chicken
Paywalled
SHALL WE PLAY A GAME?
Pretty sure the "prize" is a government of pedos.
I have wonderful dreams of walking through AI data centers destroying everthing.
No you don't.
It was just an educated guess.
Three posts away in my feed, a thread about the Pentagon demanding the AI provider for the military to remove safeguards.
Oh cool, AI will actually be the end of the world, not because it's actually sentient but because some meathead who can't tell the difference pushes the button. That's fucking great.
De-bullshitting that headline:
AIs Programmers can’t stop their programs recommending nuclear strikes in war game simulations
And yeah that's what happens inside a genocidal empire where "R&D" is strictly funded by the MIC.
Programmers can't stop morons mistaking a glorified autocorrect program for a decision making device.
Did we not learn from Sarah and John Connor long ago?
Maybe it just wants to play a nice game of chess.
It's torment nexuses all the way down...
It's an API that just returns "0000"
Do it.
Sargent McArthur eat your heart out.
For context he wanted to send 10 nukes to make a line between Taiwan and china
AI is too nuke happy.
Also gotta add the infamous Computer Fraud and Abuse act 1986 was made because of the film war games.
A high ranking offical watched war games then asked the Secretary of defense could that happen?
And the official replied yes technically.
Enter the most vague ordinance!
Do you use adblock?
CFABA violated
The shit is so vague.
I highly recommend the phreaking episode of darknet diary's.
Anyone who has played video games, especially where there is a somewhat steep learning curve or some element of past choices carrying forward thru the game, has had the moment where they realize it might be time to start fresh with the info I've acquired. It's not a shock to me that these AI entertain the nuclear option so often.
there is no ai, only largelanguagemodel that has been trained on data. The data it has been trained suggests this is the best idea. llm cant evaluate the data its trained on so anything you put in will be equally valid. I give it that its really impressive how they can output the training results in such coherent way that can be kind of "conversed" with, but there is no will or intelligence behind it.
This is also why corporations insisting on putting them everywhere is quite horrible security issue -> you can jailbreak any llm and tell them to do anything. So this has enabled all kinds of stupid vulnerabilities that exploit this. Now you can even send someone malicious google calendar invites that makes gemini do bad shit to your systems its connected to.
no, its incapable of making choices because there is nothing there to make the choices. Its just fancy way of interacting with the data it has been trained with. Though i suppose if there was a way to let llm function "live" instead of only by responding to queries, it could be possible to at least test if it could act on its own, but i dont think it can -> we would know by now because it would be step closer to agi, which is basically the holy grail for these kind of things. And equally possible to get, i think.
You can literally make the llm say and do anything with right kind of query, this is also why its impossible to make them safe. Even though you can't directly ask for something forbidden, with some creativity you can bybass the initializations the corpos have put in. Its not possible for them to account for every single thing and if they try they will run out of token space.
The whole "ai" term is just corporations perpetuating a lie because it sounds impressive and thus makes people want to give them more money for their bullshit.
No, LLMs are not just an interface for accessing training data. If that were true, then their references would actually work. The fact that LLMs can hallucinate and make stuff up proves that they are not just accessing the training data. The ANN is generating new (often incorrect) information.
Who the fuck cares.
Somebody get smarterchild to weigh in on this.
Literally the plot of Metal Gear Solid Peace Walker
Maybe it is the only real solution.
Full nuclear war and end all life on Earth ¯\_(ツ)_/¯
Edit: Do I really need to ad a "/s" on this comment? Jesus Christ folks. And here I thought this comment was already laced with enough joke sarcasm...
Wait, it could actually be a great opportunity mein Führer... i mean Mr. President. https://youtu.be/zZct-itCwPE
Yeah, because the AI will look at everything with cold logic and rationality and come to the conclusion that even though the best chance of survival is for everyone to keep their fingers off the button, all it takes is for one actor to do it for the whole system of mutually assured destruction to collapse into nuclear armageddon, in which case the best chance of survival is to be the first one to launch your nukes and take out all your enemies capabilities to retaliate.
A human being who isn't psychotic can clearly see that the resulting survival and new world order would not be particularly a pleasant one to live in. The AI doesn't care about its own comfort, though, so it will see this as the best outcome that minimizes variables.
This is why AI should never be allowed to make decisions.
I assume it's just because when writing about potential nuclear war, most people write about the bombs going off. There aren't a lot of stories and articles about nobody doing anything and everything turning out fine, presumably. And LLMs are kind of just a glorified autocomplete so that's what they go with.
True, also I saw another comment that said there was a mechanic that randomly escalates the models and actions, and almost every single nuclear choice was actually a different one that was escalated
Maybe AI/LLM being programmed by self-serving interests has bled through to the “thought” process. Do unto others before they do unto you.
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
@lemmy.world
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
go to feed...
save