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.
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.