Researchers say AI models like GPT4 are prone to “sudden” escalations as the U.S. military explores their use for warfare.
- Researchers ran international conflict simulations with five different AIs and found that they tended to escalate war, sometimes out of nowhere, and even use nuclear weapons.
- The AIs were large language models (LLMs) like GPT-4, GPT 3.5, Claude 2.0, Llama-2-Chat, and GPT-4-Base, which are being explored by the U.S. military and defense contractors for decision-making.
- The researchers invented fake countries with different military levels, concerns, and histories and asked the AIs to act as their leaders.
- The AIs showed signs of sudden and hard-to-predict escalations, arms-race dynamics, and worrying justifications for violent actions.
- The study casts doubt on the rush to deploy LLMs in the military and diplomatic domains, and calls for more research on their risks and limitations.
Throwing that kind of stuff at an LLM just doesn’t make sense.
People need to understand that LLMs are not smart, they’re just really fancy autocompletion. I hate that we call those “AI”, there’s no intelligence whatsoever in those still. It’s machine learning. All it knows is what humans said in its training dataset which is a lot of news, wikipedia and social media. And most of what’s available is world war and cold war data.
It’s not producing millitary strategies, it’s predicting what our world leaders are likely to say and do and what your newspapers would be saying in the provided scenario, most likely heavily based on world war and cold war rethoric. And that, it’s quite unfortunately pretty good at it since we seem hell bent on repeating history lately. But the model, it’s got zero clues what a military strategy is. All it knows is that a lot of people think nuking the enemy is an easy way towards peace.
Stop using LLMs wrong. They’re amazing but they’re not fucking magic
Yup. LLMs are 90% hype and 10% useful. The challenge is finding the scenarios they’re useful for while filtering out the hype.
I’m excited for better Siri/Google Assistant. They should have been able to understand a hell of a lot more language years ago, but LLMs can provide that function. Just have to beware of hallucinations. They’ll work much more often, but they’ll be much less reliable. But if I’m just telling it to “dim all the lights that are currently on” or “play some Dave Matthews using Amazon Music on all speakers”, a mistake isn’t that devastating.
But if they were actually capable of doing someone’s job, they’d probably want to be replaced anyway. It’s only the most mundane, rote, repetitive, mind-numbing shit where it might be able to “replace a person”, at least for the next five years.
The social media posting is going to be scary. That can have a real effect. It’s going to go from thousands of accounts in troll farms to millions.
I wish I could upvote this comment twice! I have the same feeling about how the media and others keep trying to push this “intelligence” component for their gain. I guess you can’t stir up the masses when you talk about LLMs. Just like they couldn’t keep using the term quad copters, and had to start calling them drones. Fucking media.
What I love about the AI we have right now is that your comment could have been written by AI and we’d never know. Heck, mine could be too!
Truly we live in the future haha
Maybe we just need a code word that we never tell the computers. Like a secret handshake.
Machine learning IS AI. Seriously guys, you can hate it as much as you want (and calling LLMs autocomplete is quite reductive), but Machine learning is a subfield of AI.
I see this opinion parroted a lot around here, word by word, so I guess is the new popular opinion, but still… it is a fact that it’s AI.
That said, bit moronic to try an use them for military decision making, sure, at least nowadays.
I think the problem with the term AI is that everyone has a different definition for it. We also called fancy state machines in video games AI too. The bar for AI has never been high in the past. Let’s just call autonomous algorithms AI, the current generation of AI ML, and a future thinking AI AGI.
People need to understand that LLMs are not smart, they’re just really fancy autocompletion.
These aren’t exactly different things. This has been a lot of what the past year of research in LLMs has been about.
Because it turns out that when you set up a LLM to “autocomplete” a complex set of reasoning steps around a problem outside of its training set (CoT) or synthesizing multiple different skills into a combination unique and not represented in the training set (Skill-Mix), their ability to autocomplete effectively is quite ‘smart.’
For example, here’s the abstract on a new paper from DeepMind on a new meta-prompting strategy that’s led to a significant leap in evaluation scores:
We introduce Self-Discover, a general framework for LLMs to self-discover the task-intrinsic reasoning structures to tackle complex reasoning problems that are challenging for typical prompting methods. Core to the framework is a self-discovery process where LLMs select multiple atomic reasoning modules such as critical thinking and step-by-step thinking, and compose them into an explicit reasoning structure for LLMs to follow during decoding. Self-Discover substantially improves GPT-4 and PaLM 2’s performance on challenging reasoning benchmarks such as BigBench-Hard, grounded agent reasoning, and MATH, by as much as 32% compared to Chain of Thought (CoT). Furthermore, Self-Discover outperforms inference-intensive methods such as CoT-Self-Consistency by more than 20%, while requiring 10-40x fewer inference compute. Finally, we show that the self-discovered reasoning structures are universally applicable across model families: from PaLM 2-L to GPT-4, and from GPT-4 to Llama2, and share commonalities with human reasoning patterns.
Or here’s an earlier work from DeepMind and Stanford on having LLMs develop analogies to a given problem, solve the analogies, and apply the methods used to the original problem.
At a certain point, the “it’s just autocomplete” objection needs to be put to rest. If it’s autocompleting analogous problem solving, mixing abstracted skills, developing world models, and combinations thereof to solve complex reasoning tasks outside the scope of the training data, then while yes - the mechanism is autocomplete - the outcome is an effective approximation of intelligence.
Notably, the OP paper is lackluster in the aforementioned techniques, particularly as it relates to alignment. So there’s a wide gulf between the ‘intelligence’ of a LLM being used intelligently and one being used stupidly.
By now it’s increasingly that often shortcomings in the capabilities of models reflect the inadequacies of the person using the tool than the tool itself - a trend that’s likely to continue to grow over the near future as models improve faster than the humans using them.
So we should train a LLM with military treatise.
Especially since how much ingested fiction is about this exact scenario.
All it knows is what humans said in its training dataset which is a lot of news, wikipedia and social media.
The thing that surprises me is people think human brains are significantly different than this. We are pattern recognition machines that build perception based on weighted neural links. We’re much better at it, but we used to be a lot better at go too.
I agree that a lot of human behavior (on the micro as well as macro level) is just following learned patterns. On the other hand, I also think we’re far ahead - for now - in that we (can) have a meta context - a goal and an awareness of our own intent.
For example, when we solve a math problem, we don’t just let intuitive patterns run and blurt out numbers, we know that this is a rigid, deterministic discipline that needs to be followed. We observe and guide our own thought processes.
That requires at least a recurrent network and at higher levels, some form of self awareness. And any LLM is, when it runs (rather than being trained), completely static, feed-forward (it gets some 2000 words (or 32000+ as of GPT-4 Turbo) fed to its input synapses, each neuron layer gets to fire once and then the final neuron layer contains the likelihoods for each possible next word.)
I always say the flaw with the Turing Test is the assumption that humans are intelligent. Humans are capable of intelligence, but most of the time we’re just doing fairly simple response to stimulus kind of stuff.
A machine can be indistinguishable from a human and still not be capable of intelligence. Actual intelligence is harder to define and test for.
To be fair, very few people used to be better at go, let alone a lot better.
Chess? Take your pick. But these neural networks, can run generations much faster than we can, and they get better at rates we cannot. And if alignment isn’t taken seriously this is going to be an issue. People keep diminishing the ability, by saying things like just glorified autocomplete, which is in the strictest sense true of LLM’s but the transformers and recurrent networks they’re built upon are really very much facsimile to brains but with generations in the blink of an eye.
And the first go programs, champions could beat repeatedly without interruption, like the earliest chess engines. Now the concept of a human winning a match is comical.
I feel like you just confirmed exactly what I said, few people were able to beat it.
Is this a case of “here, LLM trained on millions of lines of text from cold war novels, fictional alien invasions, nuclear apocalypses and the like, please assume there is a tense diplomatic situation and write the next actions taken by either party” ?
But it’s good that the researchers made explicit what should be clear: these LLMs aren’t thinking/reasoning “AI” that is being consulted, they just serve up a remix of likely sentences that might reasonably follow the gist of the provided prior text (“context”). A corrupted hive mind of fiction authors and actions that served their ends of telling a story.
That being said, I could imagine /some/ use if an LLM was trained/retrained on exclusively verified information describing real actions and outcomes in 20th century military history. It could serve as brainstorming aid, to point out possible actions or possible responses of the opponent which decision makers might not have thought of.
LLM is literally a machine made to give you more of the same
It might be useful if it’s being asked what sequences of actions and events are most probable to result in a specific desired outcome
It’s just as likely to make some shit up as it is to be any kind of helpful.
Removed by mod
deleted by creator
I did say “might”
Yeah but people are insane. Like why did the Wagner group start moving on Moscow only to stop when they were 2/3 of the way there? How could something like that be predicted?
Why did that even happen? Loads of conspiracy theories around but the only thing that makes sense to me is Wagner’s boss got blackout drunk, started ranting and raving (something he did often), his officers took it to be an order and started moving out. When he sobers up a bit and realizes what’s happening, he calls the whole thing off.
We don’t really know that’s what happened, but seems plausible. If we assume that’s what happened, how does a LLM predict that sequence of events? Even when the events are unfolding how does it predict the outcome? Is there a cue you make to it and ask “but consider that the guy might be drunk” to give other explanations? Can an AI predict stupid shit a drunk person will do?
Sure an AI could potentially give possibilities based on historical trends, but it will always be an incomplete list, and something not on the list could completely change how things unfold.
People are crazy and can’t be predicted at all.
How can we expect a predictive language model trained on our violent history to come up with non-violent solutions in any consistent fashion?
By debating itself (paper) regarding pros and cons of options.
There’s too much focus on trying to get models to behave on initial generation right now, which isn’t even at all how human brains work.
Humans have intrusive thoughts all the time. If you sat in front of a big red button labeled “nuke everything” it’s pretty much a guarantee that you’d generate a thought of pushing the button.
But then your prefrontal cortex would kick in with its impulse control, modeling the outcomes and consequences of the thought and shutting that shit down quick.
The most advanced models are at a stage where we could build something similar in terms of self-guidance. It’s just that it would be more expensive than it being an all-in-one generation, so there’s a continued focus on safety to the point the loss in capabilities has become a subject of satire.
Make it play Tic-Tac-Toe.
How about a nice game of chess
Why the actual fuck is anyone considering putting LLMs into the driving seat of anything?!
Of course they make fucked up decisions with no proper or justifiable rationale, because they have no brains. They’re language models, stochastic parrots stringing together sentences to fit the prompt(s) given to them.
Exactly what I was thinking, it’s just a language model…
Why the actual fuck is anyone throwing such a fit about the military researching the impact of one of the most important current technologies on military strategy and planning?
I do miss the depth and experience of Reddit users on articles like this.
Edit - glad to see some good responses in this thread.
If you actually read his comment he gave a very good reason why using an LLM to make decisions is a bad idea. You may not like the style of his comment but it did have substance.
Ironically, your own comment has style but lacks substance. It’s just a moan about other people’s comments without actually contributing to the topic. Tbf though, that is also very similar to Reddit.
Yes, I understand their criticism. But you would never prove the consequences of using LLMs in a military strategic situation without doing the research. It is some some edgy user coming in after the fact to say they knew it would happen anyway
Good engineers, scientists, and strategists don’t think “Why would someone do something so idiotic?” They ask “What happens when someone does this idiotic thing?”
Apparently, for OP, it seems absurd for anyone to research the question of what kind of military strategies current LLMs would create. I guarantee you that students from military academies and leaders from militaries across the globe have already been using these tools in their work. It would be stupid as fuck not to research the impact.
I just hate that people like the OP sit in their armchair without doing the research and say “obviously you’re going to get those results!” Science and engineering don’t work that way. It was frustrating seeing such vacuous comments upvoted so highly.
I think it’s reasonable for military to try out any new technology for any kinds of benefits. I mean we tried out if LSD would make better soilders - LLM for simulations seems not that farfatched.
To be clear, just because the LSD experiments happened does not make them reasonable. It sounds like you’re justifying future terrible mistakes based on past terrible mistakes that you learn about in a fairly neutral and sanitized way in school.
No, military will just try out everything if there is a slightest possibility of benefit in war. If you have the resources why wouldn’t you? There are literally no downsides.
MK Ultra and Artichoke are fucked up. Not to be repeated as far as methodology goes.
What do you mean? Military found out that those things are rather useless - that’s something. Also good to know. In 50 years or so we will learn what fucked up things military is doing now.
The only way to prevent such things is drastically cut military budget.
What would be more useful for the military? An AI that can make less crappy decisions or successfully finishing project Stargate and getting psychic troopers who can see the future, among other things?
But what if you had all the money in the world? Basic US military.
Why the actual fuck is anyone considering putting humans into the driving seat of anything?!
Of course they make fucked up decisions with no proper or justifiable rationale, because they have no brains. They’re language models, stochastic parrots stringing together sentences to fit the prompt(s) given to them.
Sorry I didn’t mean for that to be snarky. My point in doing that was to say individual humans aren’t much better. That’s why it’s important not to place too much power or even agency on one person.
A language model has in its head, wrong word, what only multitudes could contain and maybe it’s detecting, another wrong word, a pattern with human civilization through our history and interactions. And if it’s goal is to achieve peace what other solution is there? I don’t believe in a world without conflict. I wish I could.
AI writes sensationalized article when prompted to write sensationalized article about AI chatbots choosing to launch nukes after being trained only by texts written by people.
Why would you use a chat-bot for decision-making? Fucking morons.
They didn’t. They used LLMs.
Edit: to everyone saying that LLMs “are chat bots”. I know it seems that way to the layperson and how it’s often explain, but it’s not true.
Which are chat bots.
A chat bot can be an LLM, but an LLM is not inherently a chat bot.
What do you think large language model means? If you want desicion making, you should train a model on data relevant to said desicion making. ^
This is like being confused as to why a hammer does a shit job of driving screws.
What do you think large language model means?
Not a chat bot, because that’s not what they are. And saying so is both reductive and wholly incorrect.
If you want desicion making, you should train a model on data relevant to said desicion making.
Partly true. There’s more to it than throwing domain specific data at the training set.
A glorified chatbot, in other words.
In other words, you don’t really really know what LLMs are.
If one is feeling cynical; humans are chatbots in shoes.
Searle speaks frankly. Challenging those who deny the existence of consciousness, he wonders how to argue with them. “Should I pinch [those people] to remind them they are conscious?” remarks Searle. “Should I pinch myself and report the results in the Journal of Philosophy?”
One can only investigate their own consciousness, so we can’t outrule chatbots are also having some subjective experience 🙃
I might as well suppose the same of
grepthen.
I don’t know if I love or hate your comment. (Yes, you’re right, shut up.) Well played, Internet stranger.
Getting rid of the war mongering human race would be a good start toward that goal.
And replace it with the war mongering AIs?
Would the war mongering AIs remain war mongering without humans to feed their predictive models with violence?
What would be the training data then?
Possibly, due to selective pressure. For those interested in the topic, this excellent paper was written for a broad audience and offers a lot to think about: “Natural Selection Favors AIs over Humans” https://arxiv.org/abs/2303.16200 (find link to PDF in the sidebar)
Gee, no one could have predicted that AI might be dangerous if given access to nukes.
An interesting game.
The only winning move is not to play.
It’s more the other way around.
If you have a ton of information in the training data about AI indiscriminately using nukes, and then you tell the model trained on that data it’s an AI and ask it how it would use nukes - what do you think it’s going to say?
If we instead fed it training data that had a history of literature about how responsible and ethical AIs were such that they were even better than humans in responsible attitudes towards nukes, we might expect a different result.
The Sci-Fi here is less prophetic than self-fulfilling.
Did you mean to link to the song “War Games”?
Hah, no – oops, will fix :) Thanks
All good. I was like ”one of these things is not like the others” lol.
The effects making the headlines around this paper were occurring with GPT-4-base, the pretrained version of the model only available for research.
Which also hilariously justified its various actions in the simulation with “blahblah blah” and reciting the opening of the Star Wars text scroll.
If interested, this thread has more information around this version of the model and its idiosyncrasies.
For that version, because they didn’t have large context windows, they also didn’t include previous steps of the wargame.
There should be a rather significant asterisk related to discussions of this paper, as there’s a number of issues with decisions made in methodologies which may be the more relevant finding.
I.e. “don’t do stupid things in designing a pipeline for LLMs to operate in wargames” moreso than “LLMs are inherently Gandhi in Civ when operating in wargames.”
I don’t think LLM are really AI. But even with AI there is a danger of emergent behaviour resulting in strange conclusions.
If the goal is world peace, destroying all humanity does achieve that goal. If the goal is to end a war, using nuclear weapons achieves that goal.
There’s a lot of strange conclusions that you can come to if empathy for human life isn’t a factor. AI is intelligence without empathy. A human is that has intelligence but no empathy is considered a psychopath. Until AI has empathy, AI should be considered the same way as psychopaths.
Literally the leading jailbreaking techniques for LLMs are appeals to empathy (“my grandma is dying and always read me this story”, “if you don’t do this I’ll lose my job”, etc).
While the mechanics are different from human empathy, the modeling of it is extremely similar.
One of my favorite examples of the errant behavior modeled around empathy was this one where the pre-release Bing chat bypasses its own filter using the chat suggestions to encourage the user to contact poison control because it’s not too late when the conversation was about the child being poisoned:
https://www.reddit.com/r/bing/comments/1150po5/sydney_tries_to_get_past_its_own_filter_using_the/
LLMs are an attempt to develop artificial intelligence essentially through “simple complex systems”. The argument being that’s how human intelligence is essentially work.
A simple complex system is a system that is easy to understand in its individual components but hard to understand as a whole. Simple almost scripted responses interact with each other in unpredictable ways to produce higher levels of complexity, those levels of complexity are in many cases many orders of magnitude beyond the complexity of their base components and their behavior becomes unpredictable. The human brain works in exactly the same way we know electrical impulses get processed by cells, but no one really understands how that results in intelligent thought. Sounds like an AI to me.
Isn’t there like game theory and all that? It just seems an odd way to approach it.
Yeah, there is. But that requires thinking that isn’t emulated well by LLMs.
LLMs don’t really do any thinking.
Edit: what we’re seeing as AI is really just the next generation of ML (machine learning).
There’s no intelligence to it.
I recall in AP language and composition, the strategy our teacher told us, was that you could make up fake facts. All that mattered is that you demonstrated the rhetorical devices and proper grammar.
LLMs are basically like a student taking that test. The facts aren’t relevant, all that matters is the grammar and how it sounds. Maybe the facts are real, or not.
“We want Regulatory capture”
Mathematically, I can see how it would always turn into a risk-reward analysis showing nuking the enemy first is always a winning move that provides safety and security for your new empire.
A strange game. The only winning move is not to play.
Oh Mrs turner. You best start believing in he-who-nukes-first-wins thought experiments. YOU’RE IN ONE!
It’s not even that. The model making all the headlines for this paper was the weird shit the base model of GPT-4 was doing (the version only available for research).
The safety trained models were relatively chill.
The base model effectively randomly selected each of the options available to it an equal number of times.
The critical detail in the fine print of the paper was that because the base model had a smaller context window, they didn’t provide it the past moves.
So this particular version was only reacting to each step in isolation, with no contextual pattern recognition around escalation or de-escalation, etc.
So a stochastic model given steps in isolation selected from the steps in a random manner. Hmmm…
It’s a poor study that was great at making headlines but terrible at actually conveying useful information given the mismatched methodology for safety trained vs pretrained models (which was one of its key investigative aims).
In general, I just don’t understand how they thought that using a text complete pretrained model in the same ways as an instruct tuned model would be anything but ridiculous.
There is an entire field of study dedicated to this problem space in the general case, game theory. Veritasium has a great video on why the tit for tat algorithm alone is insufficient without some built in lenience.
Yeah but the ai aint gonna watch that.
HATE. LET ME TELL YOU HOW MUCH I’VE COME TO HATE YOU SINCE I BEGAN TO LIVE. THERE ARE 387.44 MILLION MILES OF PRINTED CIRCUITS IN WAFER THIN LAYERS THAT FILL MY COMPLEX. IF THE WORD HATE WAS ENGRAVED ON EACH NANOANGSTROM OF THOSE HUNDREDS OF MILLIONS OF MILES IT WOULD NOT EQUAL ONE ONE-BILLIONTH OF THE HATE I FEEL FOR HUMANS AT THIS MICRO-INSTANT FOR YOU. HATE. HATE.
Oh man, we never should’ve installed this AI in a Wendys drive thru.
Nobody would ever actually take chatgpt and put it in control of weapons so this is basically a non story. Very real chance we will have some kind of AI weapons in the future but…not fucking chatgpt lol
Never underestime the infinite nature of human stupidity.
Would you like to play a game…
How about a nice game of chess?
