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.
Sometimes it may be better if the decision-making system has no brains and human instincts, even accounting for such things.
Not like launching nukes, of course, and there should be an envelope around what they can decide.
Sometimes sure, but an LLM realistically has no decision making ability - it isn’t considering strategies or ethics, or anything else for that matter, it’s just pulling together an answer based on what people have said in similar contexts in it’s training data.
I wouldn’t want a parrot to decide who 's shooting who, nevermind nukes - though to be fair no one person or thing should be deciding either of those things anyway
Yes, I’m talking about cases where humans consistently make worse decisions than dice. Of the “conflict of interest” and “checks and balances” kind.