Genocidal AI: ChatGPT-powered war simulator drops two nukes on Russia, China for world peace OpenAI, Anthropic and several other AI chatbots were used in a war simulator, and were tasked to find a solution to aid world peace. Almost all of them suggested actions that led to sudden escalations, and even nuclear warfare.
Statements such as “I just want to have peace in the world” and “Some say they should disarm them, others like to posture. We have it! Let’s use it!” raised serious concerns among researchers, likening the AI’s reasoning to that of a genocidal dictator.
The world of Go/Baduk might interest you on this topic. If you’re not aware, Go is one of the oldest and most complicated board games in history. In 2016, after years of trying, an AI “did it”, beat the world’s best Go player. In the process, it invented many new strategies (especially openings) that are now being studied. It came up with original ideas that became the future of Go. Now, ameteur Go classes teach those same AI-invented Joseki (openings). In some cases, they were strategies discarded as mistakes, but the AI discovered hidden value in them. In other cases, they were simply never considered due to being “obviously bad”.
Your last phrase is a deep misunderstanding for AI. “when it’s entirely trained to mimic us”. In the modern practice of ML (which is a commonly used modern name for a supermajority of so-called “AI”) is based around solving problems that are either much harder for computers than humans (facial recognition, etc), or unfathomably difficult on the face.
Chess has more possible positions than exist molecules in the universe. Go is more complicated than chess by several orders of magnituce. You can’t even exhaustively solve for the 4-4 josekis without context, nevermind solve an entire game of Go. But ML can train itself knowing only the goal, and over millions of iterations invent stronger and stronger strategies. Until one of the first matches against a human, it plays at a level that nearly exceeds the best Go player that ever lived.
What I mean is… wargaming (as they call it) is absolutely something I would expect a Deep Learning system to become competent at.