How does an AI decide how to interpret vague prompts? How does it “think”?

[bakysengs]

I’ve been having a ton of fun with Midjourney AI, and it’s immensely fascinating seeing how it interprets vague prompts. When there are many different ways in which something could be interpreted, of which the internet surely has several, how does it “decide” what to show you? For example, my favorite prompt I’ve ever done: “forbidden digital love” produced a man hugging a digital woman in one, and a girl looking at a texting box with heart emojis in the other (see my recent posts if you wanna see them!) Is it just an amalgamation of everything the AI has been fed or is it referencing something in particular? I find it so amazing because that symbolism is very human-like. Another interesting one was when “the end of time” produced a group of people looking at a huge futuristic monolith.