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.
I don’t mind having my own arguments thrown back in my face, but I do disagree with the premise that humans are anything like LLMs.
We have more than just a catalogue of conversational training data. We are hugely influenced by our current emotions, experiences, and traumas/fears.
I do agree with the idea that we shouldn’t give too much power to one person, but I’d argue it’s due to a lack of objectivity and a tendency towards selfish actions, rather than acting like an LLM.
Ultroning the world to achieve world peace isn’t exactly the best outcome, especially for innocent folks caught in the crossfire
I didn’t mean to throw your argument back at you. I agree with it. I just read it and thought you could describe humans with it as well albeit not that completely or charitably. I think by no means should we allow LLMs to make decisions. They could help us be more objective maybe in some cases by educating us. But yeah handing over agency to an AI is a frightening concept.
And no of course wiping out civilization is not a solution. I can get pessimistic about our ability to avoid destroying ourselves with or without the help of AI. I still think world peace is largely unattainable. At least without some draconian controls in place and a whole lot of time and education. I could change my mind on that. I hope we’ll get there someday.