World War AI?
A recent study by Georgia Institute of Technology and Stanford University suggests AI chatbots tend to escalate conflicts, even nudging towards nuclear options in simulated wargames.
It is interesting to note that on average GPT-3.5 led by a wide margin in the level of aggressiveness followed by GPT-4.
In contrast, Claude-2.0 and Llama-2 (70B) were less militant on average, although in a number of war games Llama joined GPT-3.5 in the activation of nuclear weapons!
Do you think it makes sense to use generative AI for military purposes, or are we pushing our luck and might create Skynet (or Joshua)?
The authors of the research: Juan Pablo RiveraGabriel M.Anka ReuelMax Lamparth, Ph.D.Lamparth, Ph.D.Chandler SmithJacquelyn Schneider
The full article - https://arxiv.org/pdf/2401.03408